title
#58 Dr. Ben Goertzel - Artificial General Intelligence

description
Patreon: https://www.patreon.com/mlst Discord: https://discord.gg/ESrGqhf5CB The field of Artificial Intelligence was founded in the mid 1950s with the aim of constructing “thinking machines” - that is to say, computer systems with human-like general intelligence. Think of humanoid robots that not only look but act and think with intelligence equal to and ultimately greater than that of human beings. But in the intervening years, the field has drifted far from its ambitious old-fashioned roots. Dr. Ben Goertzel is an artificial intelligence researcher, CEO and founder of SingularityNET. A project combining artificial intelligence and blockchain to democratize access to artificial intelligence. Ben seeks to fulfil the original ambitions of the field. Ben graduated with a PhD in Mathematics from Temple University in 1990. Ben’s approach to AGI over many decades now has been inspired by many disciplines, but in particular from human cognitive psychology and computer science perspective. To date Ben’s work has been mostly theoretically-driven. Ben thinks that most of the deep learning approaches to AGI today try to model the brain. They may have a loose analogy to human neuroscience but they have not tried to derive the details of an AGI architecture from an overall conception of what a mind is. Ben thinks that what matters for creating human-level (or greater) intelligence is having the right information processing architecture, not the underlying mechanics via which the architecture is implemented. Ben thinks that there is a certain set of key cognitive processes and interactions that AGI systems must implement explicitly such as; working and long-term memory, deliberative and reactive processing, perc biological systems tend to be messy, complex and integrative; searching for a single “algorithm of general intelligence” is an inappropriate attempt to project the aesthetics of physics or theoretical computer science into a qualitatively different domain. Panel: Dr. Tim Scarfe, Dr. Yannic Kilcher, Dr. Keith Duggar Pod version: https://anchor.fm/machinelearningstreettalk/episodes/58-Dr--Ben-Goertzel---Artificial-General-Intelligence-e15p20i Artificial General Intelligence: Concept, State of the Art, and Future Prospects https://sciendo.com/abstract/journals/jagi/5/1/article-p1.xml The General Theory of General Intelligence: A Pragmatic Patternist Perspective https://arxiv.org/abs/2103.15100 [00:00:00] Lex Skit [00:03:00] Intro to Ben [00:10:42] Concept paper [00:20:50] Minsky [00:21:42] OpenCog [00:25:50] SinglularityNet [00:27:19] Patternist Paper [00:30:13] Short Intro [00:35:43] Cognitive Synergy [00:41:29] Hypergraphs vs vectors: focus operations and algebra, not representations [00:47:46] Does brain structure form a hypergraph? [00:51:21] What's missing from neural networks today? [00:56:52] Sensory knowledge, bottom-up and top-down reasoning [01:02:02] If the brain is a continous computer, then why graphs? [01:08:54] Forgetting is as important as learning [01:11:55] Should we ressurrect analog computing? [01:18:18] AIXI - limitations [01:25:20] AIXI - the reductio absurdum of reinforcement learning [01:27:56] Defining intelligence [01:33:34] Pure Intelligence [01:40:08] SingularityNET - a decentralized path to practical AGI [01:47:18] SingularityNET - can we automate API discovery and understanding? [01:53:36] Wrap up [01:56:36] A true polymath [01:59:58] SigularityNET and the API problem [02:04:45] Dynamic AGI vs reliable engineering [02:10:42] Can intelligence emerge in SingularityNET? [02:19:10] How is AIXI a useful mental exercise? https://opencog.org/ https://singularitynet.io/ Yannic's video on SingularityNet https://www.youtube.com/watch?v=yFAuXmcGk2Y Music credit: https://soundcloud.com/vskymusic

detail
{'title': '#58 Dr. Ben Goertzel - Artificial General Intelligence', 'heatmap': [], 'summary': "Delving into contrasting reactions towards ai and discussing the pursuit of artificial general intelligence, emphasizing the potential emergence of ai with human-level intelligence by 2029 and surpassing human intelligence by 2045, while exploring mlst's view on ai knowledge acquisition, the importance of cognitive synergy in achieving human-like intelligence in agi systems, ai research categorization, brain plasticity, gödel's theorem's impact on human and digital computer creativity, limitations of reward functions in ai, dr. ben gertzl's insights on ai, future internet services, gpt-3 limitations, predictive abilities in bacterial systems, and the constraints of aixi in representing intelligence.", 'chapters': [{'end': 399.478, 'segs': [{'end': 56.343, 'src': 'embed', 'start': 10.089, 'weight': 1, 'content': [{'end': 13.571, 'text': "either people are really excited about it or they're really terrified of it.", 'start': 10.089, 'duration': 3.482}, {'end': 15.672, 'text': 'It seems to be the two responses.', 'start': 13.611, 'duration': 2.061}, {'end': 26.016, 'text': "I've been working on AI for three decades, and I first started thinking about AI when I was a little kid in the late 60s and early 70s,", 'start': 15.732, 'duration': 10.284}, {'end': 29.538, 'text': 'when I saw AIs and robots on the original Star Trek.', 'start': 26.016, 'duration': 3.522}, {'end': 39.123, 'text': "So I guess I've had a lot of cycles to process the positives and negatives of it, whereas now, suddenly,", 'start': 29.558, 'duration': 9.565}, {'end': 42.447, 'text': 'most of the world is thinking through all this for the first time.', 'start': 39.123, 'duration': 3.324}, {'end': 51.061, 'text': "Now we're at a time in history when a lot of these ideas can be made real, which is amazing and scary.", 'start': 42.628, 'duration': 8.433}, {'end': 56.343, 'text': 'Not only are we on the cusp of creating a form of intelligence going beyond ourselves,', 'start': 51.161, 'duration': 5.182}], 'summary': 'Ai elicits excitement and fear, with potential for real-world impact.', 'duration': 46.254, 'max_score': 10.089, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY10089.jpg'}, {'end': 102.764, 'src': 'embed', 'start': 78.186, 'weight': 4, 'content': [{'end': 88.092, 'text': 'who is one of the most influential people in the field of artificial general intelligence and spearheads many famous projects such as OpenCog,', 'start': 78.186, 'duration': 9.906}, {'end': 90.654, 'text': 'SingularityNet and Sophia the Robot.', 'start': 88.092, 'duration': 2.562}, {'end': 101.103, 'text': "We read the SingularityNet paper by Ben and his collaborators and then listened to Ben's appearance on the excellent Lex Friedman podcast,", 'start': 91.738, 'duration': 9.365}, {'end': 102.764, 'text': 'and we knew we had to talk to him.', 'start': 101.103, 'duration': 1.661}], 'summary': 'Ben is a key figure in ai, leading projects like opencog and singularitynet.', 'duration': 24.578, 'max_score': 78.186, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY78186.jpg'}, {'end': 151.325, 'src': 'embed', 'start': 126.935, 'weight': 2, 'content': [{'end': 136.183, 'text': 'Think of humanoid robots that not only look but act and think with intelligence equal to and ultimately greater than that of human beings.', 'start': 126.935, 'duration': 9.248}, {'end': 142.291, 'text': 'In the intervening years, though, the field has drifted far from its ambitious old-fashioned roots.', 'start': 137.384, 'duration': 4.907}, {'end': 151.325, 'text': 'We should again be pursuing artificial general intelligence as a distinct effort from our current narrow approaches.', 'start': 142.732, 'duration': 8.593}], 'summary': 'Pursue artificial general intelligence separate from narrow approaches.', 'duration': 24.39, 'max_score': 126.935, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY126935.jpg'}, {'end': 215.632, 'src': 'embed', 'start': 192.14, 'weight': 0, 'content': [{'end': 203.982, 'text': 'Suppose my good friend Ray Kurzweil is correct, that 2029 we get AIs as smart as people, by 2045 we have AIs boundlessly more intelligent than people.', 'start': 192.14, 'duration': 11.842}, {'end': 213.831, 'text': "We're gonna find that any AI that has human-level intelligence is gonna have some species of advanced reflective consciousness.", 'start': 204.926, 'duration': 8.905}, {'end': 215.632, 'text': "but I mean I'm open to other possibilities.", 'start': 213.831, 'duration': 1.801}], 'summary': 'By 2029, ais will match human intelligence; by 2045, ais will surpass it.', 'duration': 23.492, 'max_score': 192.14, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY192140.jpg'}], 'start': 10.089, 'title': "Ai's contrasting perspectives and future", 'summary': 'Delves into contrasting reactions towards ai, reflecting on global engagement and discussing the pursuit of artificial general intelligence, emphasizing the need for convergence on a good agi architecture and the potential emergence of ai with human-level intelligence by 2029 and surpassing human intelligence by 2045.', 'chapters': [{'end': 56.343, 'start': 10.089, 'title': 'Ai: excitement and fear', 'summary': 'Discusses the contrasting reactions of people towards ai, with the speaker reflecting on their three-decade involvement in ai and the current global engagement with its implications, highlighting the potential both for awe and apprehension.', 'duration': 46.254, 'highlights': ['The speaker has been working on AI for three decades, providing a seasoned perspective on the topic.', 'The global engagement with AI implications is highlighted, emphasizing the current first-time considerations for many.', 'The potential for creating intelligence beyond human capacity is acknowledged, indicating the awe and apprehension surrounding this advancement.']}, {'end': 191.419, 'start': 56.343, 'title': 'The future of artificial general intelligence', 'summary': 'Discusses the importance of pursuing artificial general intelligence as a distinct effort from current narrow approaches and the need for a convergence on a good agi architecture, as emphasized by ben goertzel, an influential figure in the field of artificial general intelligence.', 'duration': 135.076, 'highlights': ['Ben Goertzel is a leading figure in the field of artificial general intelligence, spearheading influential projects like OpenCog, SingularityNet, and Sophia the Robot, and emphasizing the importance of organizing early stage AIs to be as beneficial as possible.', 'The field of artificial intelligence was initially aimed at constructing thinking machines with human-like general intelligence, and it is essential to redirect efforts towards pursuing artificial general intelligence as a distinct endeavor from current narrow approaches.', 'Many in the AI community still succumb to the first step fallacy, believing that incrementally improving current AI systems will lead towards artificial general intelligence, highlighting the need for a convergence on a good AGI architecture.', 'Dr. Ben Goertzel is the CEO and founder of SingularityNet, a project that combines artificial intelligence and blockchain, showcasing his influence in merging these technologies to advance the field of AI.']}, {'end': 399.478, 'start': 192.14, 'title': 'The future of artificial general intelligence', 'summary': 'Discusses the potential emergence of ai with human-level intelligence by 2029 and ais surpassing human intelligence by 2045, the unique characteristics of ai minds, the theoretical approach to agi by ben, and the distinction between symbolic and sub-symbolic approaches.', 'duration': 207.338, 'highlights': ["AIs reaching human-level intelligence by 2029 and surpassing human intelligence by 2045 Ray Kurzweil's prediction of AIs reaching human-level intelligence by 2029 and surpassing human intelligence by 2045.", "Unique characteristics of AI minds AI's ability to swap brain matter, join into Borg minds, and subdivide, leading to a single big AI mind network.", "Ben's theoretical approach to AGI Ben's theoretical approach to AGI, inspired by human cognitive psychology and computer science, aiming to create artificial general intelligence.", 'Distinction between symbolic and sub-symbolic approaches The distinction between symbolic and sub-symbolic approaches, with Ben being in the rare camp of being sub-symbolic on a discrete knowledge graph.']}], 'duration': 389.389, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY10089.jpg', 'highlights': ['AIs reaching human-level intelligence by 2029 and surpassing human intelligence by 2045', 'The potential for creating intelligence beyond human capacity is acknowledged, indicating the awe and apprehension surrounding this advancement', 'The field of artificial intelligence was initially aimed at constructing thinking machines with human-like general intelligence, and it is essential to redirect efforts towards pursuing artificial general intelligence as a distinct endeavor from current narrow approaches', 'The speaker has been working on AI for three decades, providing a seasoned perspective on the topic', 'Ben Goertzel is a leading figure in the field of artificial general intelligence, spearheading influential projects like OpenCog, SingularityNet, and Sophia the Robot, and emphasizing the importance of organizing early stage AIs to be as beneficial as possible']}, {'end': 1813.324, 'segs': [{'end': 424.495, 'src': 'embed', 'start': 400.519, 'weight': 1, 'content': [{'end': 410.027, 'text': 'Now, MLST is really big on diversity of thought in AI, and we focused a lot on modern variants of go-fi or modern-day hybrid symbolism recently.', 'start': 400.519, 'duration': 9.508}, {'end': 412.83, 'text': 'As well as not liking the old-fashioned part.', 'start': 410.588, 'duration': 2.242}, {'end': 415.112, 'text': "let's face it, I mean all AI is old-fashioned right?", 'start': 412.83, 'duration': 2.282}, {'end': 424.495, 'text': 'Ben associated the term with the idea that knowledge should be handcrafted by humans, and his mind went straight to the disastrous PSYCH project,', 'start': 415.732, 'duration': 8.763}], 'summary': 'Mlst emphasizes diversity in ai and modern hybrid symbolism, rejecting old-fashioned ai concepts.', 'duration': 23.976, 'max_score': 400.519, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY400519.jpg'}, {'end': 542.572, 'src': 'embed', 'start': 511.479, 'weight': 0, 'content': [{'end': 516.245, 'text': 'formal logic and the belief that mind should be a hybrid of symbolic and sub-symbolic.', 'start': 511.479, 'duration': 4.766}, {'end': 526.518, 'text': 'Ben also thinks that cognitive architecture should be handcrafted with designs inspired from our knowledge of human cognition and also they should be strongly modular.', 'start': 516.966, 'duration': 9.552}, {'end': 534.627, 'text': 'He thinks that the underlying knowledge representation should be multimodal and distributed, composable and captured in a discrete graph.', 'start': 526.578, 'duration': 8.049}, {'end': 542.572, 'text': 'So Ben thinks that we should seed the knowledge graph to some extent, but it should be able to acquire most of its knowledge from reasoning,', 'start': 535.288, 'duration': 7.284}], 'summary': 'Ben advocates for a hybrid cognitive architecture with multimodal knowledge representation and reasoning capabilities.', 'duration': 31.093, 'max_score': 511.479, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY511479.jpg'}, {'end': 673.717, 'src': 'embed', 'start': 649.244, 'weight': 2, 'content': [{'end': 658.511, 'text': 'State of the Art and Future Prospects where he boiled down the spectrum of AGI designs into four even simpler categories right?', 'start': 649.244, 'duration': 9.267}, {'end': 664.355, 'text': 'So symbolic, emergentist, hybrid and universalist.', 'start': 658.571, 'duration': 5.784}, {'end': 673.717, 'text': 'AGI theory is a patchwork of overlapping concepts and frameworks and hypotheses, often synergistic, but sometimes completely contradictory.', 'start': 665.195, 'duration': 8.522}], 'summary': 'The agi designs are categorized into symbolic, emergentist, hybrid, and universalist, forming a patchwork of concepts and frameworks.', 'duration': 24.473, 'max_score': 649.244, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY649244.jpg'}, {'end': 1396.748, 'src': 'embed', 'start': 1369.292, 'weight': 4, 'content': [{'end': 1373.715, 'text': 'OpenCog is a project that aims to build an open-source artificial intelligence framework.', 'start': 1369.292, 'duration': 4.423}, {'end': 1383.4, 'text': 'It defines a set of interacting components designed to give rise to human-equivalent artificial general intelligence as an emergent phenomenon of the whole system.', 'start': 1374.115, 'duration': 9.285}, {'end': 1393.346, 'text': 'OpenCog is a software framework which hosts multimodal cognitive processes on top of a common representational substrate or, in plain English,', 'start': 1384.441, 'duration': 8.905}, {'end': 1396.748, 'text': 'a common knowledge graph, which has been implemented as a graph database.', 'start': 1393.346, 'duration': 3.402}], 'summary': 'Opencog aims to build open-source ai framework for human-equivalent general intelligence.', 'duration': 27.456, 'max_score': 1369.292, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY1369292.jpg'}], 'start': 400.519, 'title': 'Ai knowledge acquisition and agi', 'summary': "Discusses mlst's view on ai knowledge acquisition, emphasizing diversity of thought and hybrid cognitive architecture. it also explores general intelligence, the agi hypothesis, approaches to achieving agi, and the evolution of ai with a focus on the neuro-symbolic third wave and opencog framework.", 'chapters': [{'end': 673.717, 'start': 400.519, 'title': "Mlst's view on ai knowledge acquisition", 'summary': "Discusses mlst's perspective on ai knowledge acquisition, emphasizing the importance of diversity of thought and the contrast between handcrafted and emergent knowledge representation, with ben proposing a hybrid cognitive architecture influenced by cognitive psychology.", 'duration': 273.198, 'highlights': ['Ben proposes a hybrid cognitive architecture influenced by cognitive psychology, emphasizing the importance of multimodal and distributed knowledge representation, and the acquisition of knowledge through reasoning, imagination, and experience.', 'MLST focuses on the diversity of thought in AI, with a particular emphasis on modern variants of go-fi and hybrid symbolism, while questioning the old-fashioned approach to AI knowledge acquisition.', 'Ben categorizes AGI designs into four categories: symbolic, emergentist, hybrid, and universalist, providing a simplified framework for understanding different approaches to AGI.', 'MLST discusses the failure of the PSYCH project, the largest ever knowledge graph created by humans, highlighting the knowledge acquisition bottleneck as a significant failure of expert systems in the 1980s.', "Ben's mental framework for AI research interests includes neuroscience, biology, general math theory of intelligence, formal logic, cognitive science, philosophy of mind, and applied empirical work, demonstrating a broad and multidisciplinary approach to AI.", 'MLST mentions the disagreement between Ben and Rich Sutton regarding the design of the modular cognitive architecture of an AI system, with Sutton advocating for a blank slate approach, while Ben suggests designing it based on knowledge gleaned from cognitive psychology.', 'MLST emphasizes the concept of a mental map of AI researchers and their research interests, acknowledging that it may not be entirely accurate but highlighting the use of internal mental maps for understanding the AI research landscape.', 'Ben challenges the conventional belief that all AI is old-fashioned, invoking the idea that knowledge should be acquired through reasoning, deduction, instruction, and learning, rather than handcrafted by humans.']}, {'end': 849.82, 'start': 674.278, 'title': 'General intelligence and agi hypothesis', 'summary': 'Explores the nature of general intelligence, discussing its key characteristics such as achieving various goals, transferring knowledge, and efficiency in tasks, while also presenting the agi hypothesis and different characterizations of general intelligence.', 'duration': 175.542, 'highlights': ['Ben discussed the key characteristics of general intelligence, including achieving a variety of goals and tasks in different contexts, handling novel problems, and transferring knowledge between contexts. Achieving a variety of goals and tasks, handling novel problems, transferring knowledge', 'Ben introduced his core AGI hypothesis, emphasizing that the creation of general intelligence is an entirely different category from narrow intelligences. Core AGI hypothesis, differentiation from narrow intelligences', 'Three fundamental ways of characterizing general intelligence were discussed: the mathematical approach, the adaptionist approach, and the embodied approach, each with their own proponents and perspectives. Characterization of general intelligence: mathematical, adaptionist, embodied approaches', 'The mathematical approach, as defined by Legg and Hutter, involves a computer program executed by an agent achieving the best average reward calculated over all possible environments, showing the contrast between human and AI intelligence. Mathematical approach by Legg and Hutter, comparison of human and AI intelligence', 'The adaptionist approach, as defined by Pei Wang, focuses on adaptation to the environment given limited resources and the complex compromises needed to achieve generality using limited computational resources. Adaptionist approach by Pei Wang, emphasis on adaptation and limited computational resources']}, {'end': 1168.695, 'start': 850.46, 'title': 'Approaches to achieving agi', 'summary': 'Discusses four categories of agi research: symbolic, emergentist, hybrid, and universalist approaches, emphasizing their distinct characteristics and implications for achieving human-like general intelligence.', 'duration': 318.235, 'highlights': ['The chapter discusses four categories of AGI research: symbolic, emergentist, hybrid, and universalist approaches. The chapter provides a comprehensive overview of the four main categories of AGI research, establishing the foundation for further analysis.', 'Symbolic AGI advocates argue that symbolic thought is crucial for human general intelligence, but tends to be weak in learning, creativity, and procedure learning. The symbolic approach emphasizes the manipulation of symbols representing the world, however, it is limited in learning, creativity, and memory capabilities.', 'Emergentist AGI expects abstract symbol processing to emerge from low-level, sub-symbolic dynamics, such as neural networks or brain-inspired models. The emergentist approach focuses on the emergence of abstract symbol processing from sub-symbolic dynamics, potentially through neural networks or brain-inspired algorithms.', 'Hybrid AGI architectures aim to leverage the strengths of different components to work synergistically, but may face challenges in adapting due to the qualitative differences of the modules. Hybrid architectures emphasize the synergy between modules while acknowledging potential challenges in adaptation due to the diverse nature of the components.', 'Universalist approaches to AGI propose that supplying infinite computing power to AGI algorithms would yield incredibly powerful general intelligence. Universalist approaches advocate for starting at a high level of abstraction and working downwards, believing that infinite computing power can lead to powerful general intelligence.']}, {'end': 1327.684, 'start': 1168.715, 'title': 'Evolution of ai: neuro-symbolic third wave', 'summary': 'Discusses the evolving ideas in ai, particularly focusing on the neuro-symbolic or the third wave approach, advocating for a hybrid model, modular ai systems, and the possibility of emergent intelligence from a network of ais.', 'duration': 158.969, 'highlights': ['The evolving ideas in AI, particularly the neuro-symbolic or the third wave approach, advocate for a hybrid model, modular AI systems, and the possibility of emergent intelligence from a network of AIs.', 'The belief that AI systems should be strongly modular, with knowledge represented explicitly and discretely, usually in some kind of graph, is a key characteristic of the discussed mindset.', "Advocates for the primacy of logic-based reasoning for knowledge acquisition are prevalent in the modern-day symbolists' approach to AI.", 'The possibility of emergent intelligence from a network of different AIs, each specializing in certain kinds of problems and cooperating to form emergent intelligence, is suggested as a potential breakthrough in achieving general intelligence and super intelligence.', "The concept of the 'society of minds' and the idea of creating synthetic intelligences with tremendously greater general intelligence and practical capability than human beings are also discussed in the chapter."]}, {'end': 1813.324, 'start': 1327.684, 'title': 'Opencog: unifying ai framework', 'summary': "Discusses opencog, an open-source ai framework aiming to achieve artificial general intelligence through a neural symbolic knowledge graph, with a focus on decentralized agents, and the creator's perspective on general intelligence and the singularitynet marketplace.", 'duration': 485.64, 'highlights': ['OpenCog is a project aiming to build an open-source artificial intelligence framework for achieving artificial general intelligence. OpenCog aims to create human-equivalent artificial general intelligence through a neural symbolic knowledge graph and a software framework hosting multimodal cognitive processes.', 'The framework allows for decentralized agents to operate on a common knowledge platform. OpenCog enables decentralized agents to operate on a common knowledge platform, offering the potential for encoding and evolving knowledge and algorithms.', 'The creator also discusses the SingularityNet marketplace, a global AI marketplace built on blockchain, aiming to enable anyone to monetize AI. The creator introduces SingularityNet, a global AI marketplace built on blockchain, allowing anyone to monetize AI by providing a platform for APIs to call other APIs and pay for services.', "The creator's perspective on general intelligence emphasizes the role of patterns and organization in achieving complex goals in complex environments. The creator's philosophy of mind, termed 'patternism,' emphasizes the role of patterns and organization in achieving complex goals in complex environments, defining intelligence as the ability to recognize patterns and achieve goals."]}], 'duration': 1412.805, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY400519.jpg', 'highlights': ['Ben proposes a hybrid cognitive architecture influenced by cognitive psychology, emphasizing multimodal and distributed knowledge representation.', 'MLST focuses on the diversity of thought in AI, with a particular emphasis on modern variants of go-fi and hybrid symbolism.', 'Ben categorizes AGI designs into four categories: symbolic, emergentist, hybrid, and universalist, providing a simplified framework for understanding different approaches to AGI.', 'The chapter discusses four categories of AGI research: symbolic, emergentist, hybrid, and universalist approaches.', 'OpenCog aims to create human-equivalent artificial general intelligence through a neural symbolic knowledge graph and a software framework hosting multimodal cognitive processes.']}, {'end': 2414.338, 'segs': [{'end': 2085.973, 'src': 'embed', 'start': 2063.742, 'weight': 0, 'content': [{'end': 2072.667, 'text': 'and in in that sense i felt the transition from the expert system methodology, which was a key part of go-fi,', 'start': 2063.742, 'duration': 8.925}, {'end': 2075.569, 'text': 'back in the actual good old-fashioned days.', 'start': 2072.667, 'duration': 2.902}, {'end': 2085.973, 'text': 'the step from that to supervised learning was a smaller step than was commonly understood,', 'start': 2075.569, 'duration': 10.404}], 'summary': 'Transition from expert system to supervised learning was smaller than commonly understood', 'duration': 22.231, 'max_score': 2063.742, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2063742.jpg'}, {'end': 2178.962, 'src': 'embed', 'start': 2144.034, 'weight': 1, 'content': [{'end': 2160.537, 'text': 'This brings us to cognitive synergy, in that the need for cognitive synergy in an AGI system really comes out of the need to do experiential,', 'start': 2144.034, 'duration': 16.503}, {'end': 2172.938, 'text': 'interactive learning, which involves achieving complex goals in a complex environment under fairly strictly limited compute compute resources.', 'start': 2160.537, 'duration': 12.401}, {'end': 2178.962, 'text': "so if you you know, if you don't have limitations on compute resources, you can argue the point in various ways.", 'start': 2172.938, 'duration': 6.024}], 'summary': 'Agi system requires cognitive synergy for experiential learning and goal achievement with limited compute resources.', 'duration': 34.928, 'max_score': 2144.034, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2144034.jpg'}, {'end': 2275.078, 'src': 'embed', 'start': 2249.739, 'weight': 2, 'content': [{'end': 2254.961, 'text': 'That very, very simple line of reasoning is the crux of cognitive synergy.', 'start': 2249.739, 'duration': 5.222}, {'end': 2265.073, 'text': 'But then you need to get into what are the types of knowledge and learning involved in making an human-like AGI system, or, for that matter,', 'start': 2255.001, 'duration': 10.072}, {'end': 2272.076, 'text': 'a transhuman AGI system, which one would suppose would include even more kinds of knowledge and learning than a human-like AGI system.', 'start': 2265.073, 'duration': 7.003}, {'end': 2275.078, 'text': 'And there you get into cognitive science.', 'start': 2272.677, 'duration': 2.401}], 'summary': 'Cognitive synergy is crucial for developing human-like and transhuman agi systems, involving various types of knowledge and learning.', 'duration': 25.339, 'max_score': 2249.739, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2249739.jpg'}, {'end': 2325.486, 'src': 'embed', 'start': 2296.781, 'weight': 4, 'content': [{'end': 2307.769, 'text': 'Because if you look at sensory or perceptual knowledge as one of the kinds of knowledge that the human brain evolved to deal with for the kinds of tasks and environments it had to deal with,', 'start': 2296.781, 'duration': 10.988}, {'end': 2315.174, 'text': "it's clear that modern deep neural nets are one type of system that's really good at pattern recognition and huge amounts of sensory knowledge.", 'start': 2307.769, 'duration': 7.405}, {'end': 2325.486, 'text': 'If you look at declarative knowledge, like facts and beliefs, including more abstract ones, such as have been explored in the logic field,', 'start': 2316.181, 'duration': 9.305}], 'summary': 'Modern deep neural nets excel in sensory knowledge and pattern recognition.', 'duration': 28.705, 'max_score': 2296.781, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2296781.jpg'}], 'start': 1813.384, 'title': 'Agi and cognitive synergy', 'summary': 'Explores the importance of cognitive synergy in achieving human-like intelligence in agi systems, emphasizing the transition from expert system methodology to experiential and interactive learning, and the need for complex goals achievement in a limited compute environment. it also discusses the concept of cognitive synergy in agi systems, exploring the types of knowledge and learning involved, such as sensory processing with neural nets and declarative knowledge processing with logic engines, emphasizing the importance of coupling them together for perception, cognition, and feedback.', 'chapters': [{'end': 2249.238, 'start': 1813.384, 'title': 'The future of agi and cognitive synergy', 'summary': 'Delves into the importance of cognitive synergy in achieving human-like intelligence in agi systems, emphasizing the transition from expert system methodology to experiential and interactive learning, and the need for complex goals achievement in a limited compute environment.', 'duration': 435.854, 'highlights': ['The transition from expert system methodology to supervised learning was a smaller step than commonly understood, emphasizing the shift towards experiential and interactive learning. Transition from hand coding rules to hand labeling data, emphasizing the importance of unstructured experiential learning.', 'The need for cognitive synergy in an AGI system arises from the requirement to achieve complex goals in a complex environment under limited compute resources. Emphasizing the importance of achieving complex goals in a limited compute environment.', 'The mainstream of AI is getting closer to experiential and interactive learning but is still not fully there, akin to the situation with Go-Fi back in the 70s or 80s. Highlighting the gradual shift towards experiential and interactive learning in the mainstream AI community.']}, {'end': 2414.338, 'start': 2249.739, 'title': 'Understanding cognitive synergy in agi systems', 'summary': 'Discusses the concept of cognitive synergy in agi systems, exploring the types of knowledge and learning involved, such as sensory processing with neural nets and declarative knowledge processing with logic engines, emphasizing the importance of coupling them together for perception, cognition, and feedback.', 'duration': 164.599, 'highlights': ['The chapter explores the types of knowledge and learning involved in making an human-like AGI system and transhuman AGI system, emphasizing the manifestation of cognitive synergy in AGI design like OpenCog, which includes sensory processing with neural nets and declarative knowledge processing with logic engines.', "It emphasizes the importance of coupling sensory processing with neural nets and declarative knowledge processing with logic engines in AGI systems to enable perception, cognition, and feedback, stressing the need for them to be sensitive to each other's intermediate states while going through competition.", 'The chapter discusses the importance of understanding the kinds of knowledge that the human mind has and the associated learning, highlighting the effectiveness of modern deep neural nets in pattern recognition and processing of sensory knowledge, as well as uncertain reasoning engines and logic formalisms for dealing with abstract knowledge.']}], 'duration': 600.954, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY1813384.jpg', 'highlights': ['The transition from expert system methodology to supervised learning was a smaller step than commonly understood, emphasizing the shift towards experiential and interactive learning.', 'The need for cognitive synergy in an AGI system arises from the requirement to achieve complex goals in a complex environment under limited compute resources, emphasizing the importance of achieving complex goals in a limited compute environment.', 'The chapter explores the types of knowledge and learning involved in making a human-like AGI system and transhuman AGI system, emphasizing the manifestation of cognitive synergy in AGI design like OpenCog, which includes sensory processing with neural nets and declarative knowledge processing with logic engines.', "It emphasizes the importance of coupling sensory processing with neural nets and declarative knowledge processing with logic engines in AGI systems to enable perception, cognition, and feedback, stressing the need for them to be sensitive to each other's intermediate states while going through competition.", 'The chapter discusses the importance of understanding the kinds of knowledge that the human mind has and the associated learning, highlighting the effectiveness of modern deep neural nets in pattern recognition and processing of sensory knowledge, as well as uncertain reasoning engines and logic formalisms for dealing with abstract knowledge.']}, {'end': 3295.813, 'segs': [{'end': 2444.837, 'src': 'embed', 'start': 2414.678, 'weight': 0, 'content': [{'end': 2419.339, 'text': 'We interviewed loads and loads of AI researchers and I project them into a kind of vector space in my mind.', 'start': 2414.678, 'duration': 4.661}, {'end': 2425.841, 'text': "You know, Pedro Domingo says that there's five tribes in AI and you see evolutionary people and kind of logic based people,", 'start': 2419.379, 'duration': 6.462}, {'end': 2427.701, 'text': "and it's kind of empirical people.", 'start': 2425.841, 'duration': 1.86}, {'end': 2434.627, 'text': 'so the reason why i placed you there in the vector space is you believe in the primacy of modularization and the society of mind,', 'start': 2428.081, 'duration': 6.546}, {'end': 2438.051, 'text': 'the primacy of knowledge, and also the system should be discrete.', 'start': 2434.627, 'duration': 3.424}, {'end': 2444.837, 'text': "first. i agree with you 100 that knowledge shouldn't be handcrafted, by the way, and rich sutton, i mean we're completely aligned on that.", 'start': 2438.051, 'duration': 6.786}], 'summary': 'Ai researchers categorized into tribes based on beliefs, emphasizing modularization and knowledge, aligning with rich sutton.', 'duration': 30.159, 'max_score': 2414.678, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2414678.jpg'}, {'end': 2577.874, 'src': 'embed', 'start': 2544.9, 'weight': 2, 'content': [{'end': 2548.841, 'text': 'And could you maybe go into that a little bit?', 'start': 2544.9, 'duration': 3.941}, {'end': 2555.563, 'text': 'Why do you think that graphs, specifically, are good representations,', 'start': 2548.921, 'duration': 6.642}, {'end': 2564.547, 'text': 'or why should graphs be sort of the fundamental base where these different processes interconnect, if I understood you correctly in that?', 'start': 2555.563, 'duration': 8.984}, {'end': 2577.874, 'text': "Yeah, I've been an advocate of graph hypergraph and metagraph representations for AGI systems and narrow AI systems for a long time,", 'start': 2565.727, 'duration': 12.147}], 'summary': 'Graphs are fundamental representations for agi and narrow ai systems.', 'duration': 32.974, 'max_score': 2544.9, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2544900.jpg'}, {'end': 2672.297, 'src': 'embed', 'start': 2645.057, 'weight': 1, 'content': [{'end': 2661.273, 'text': "So I think that I don't believe that these sorts of linear operations on high-dimensional vector spaces are the right algebra for most of what general intelligence needs to do.", 'start': 2645.057, 'duration': 16.216}, {'end': 2672.297, 'text': 'I used to teach computer graphics when I was in a much earlier phase of my career, which is where the processing tools now used for AI come from.', 'start': 2662.174, 'duration': 10.123}], 'summary': 'Linear operations in high-dimensional vector spaces may not be suitable for general intelligence needs; ai processing tools originate from computer graphics.', 'duration': 27.24, 'max_score': 2645.057, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2645057.jpg'}, {'end': 2860.05, 'src': 'embed', 'start': 2803.895, 'weight': 3, 'content': [{'end': 2819.356, 'text': "And it seems like this nonlinear dynamical aspect has not yet been relearned or regained within the neural net paradigm as it's become mainstream.", 'start': 2803.895, 'duration': 15.461}, {'end': 2824.2, 'text': 'but it ties in with why graphs are useful,', 'start': 2819.356, 'duration': 4.844}, {'end': 2832.647, 'text': 'because if you have a complex body of knowledge and you have a bunch of nonlinear transformations that you want to carry out,', 'start': 2824.2, 'duration': 8.447}, {'end': 2842.256, 'text': 'perhaps in an asynchronous, parallel way on this large body of knowledge, a graph, or better yet, a hypergraph or a metagraph,', 'start': 2832.647, 'duration': 9.609}, {'end': 2854.265, 'text': 'are a relatively concise and efficient way to organize knowledge that will allow you to efficiently execute a variety of these nonlinear operations.', 'start': 2842.256, 'duration': 12.009}, {'end': 2860.05, 'text': "It's quite different than wanting to multiply a matrix by a vector,", 'start': 2854.686, 'duration': 5.364}], 'summary': 'Nonlinear dynamics not integrated in neural net paradigm; graphs efficient for complex knowledge and nonlinear transformations.', 'duration': 56.155, 'max_score': 2803.895, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2803895.jpg'}, {'end': 3303.717, 'src': 'embed', 'start': 3274.312, 'weight': 5, 'content': [{'end': 3281.614, 'text': 'What Gerald Edelman explored in his work on neural Darwinism and he was a collaborator later of Eugene Izikovich,', 'start': 3274.312, 'duration': 7.302}, {'end': 3284.895, 'text': 'whose work I mentioned on the nonlinear dynamics of the neuron.', 'start': 3281.614, 'duration': 3.281}, {'end': 3291.437, 'text': 'neural Darwinism was a theory of the evolution by a sort of natural selection of neural sub-assemblies.', 'start': 3284.895, 'duration': 6.542}, {'end': 3295.813, 'text': "I mean, we're we're not capturing that in modern neural architectures at all.", 'start': 3291.617, 'duration': 4.196}, {'end': 3303.717, 'text': 'And simplistic attempts to use CMAES or other evolutionary algorithms to evolve neural nets are not the same right?', 'start': 3295.913, 'duration': 7.804}], 'summary': 'Gerald edelman explored neural darwinism theory for evolution of neural sub-assemblies, not captured in modern neural architectures.', 'duration': 29.405, 'max_score': 3274.312, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY3274312.jpg'}], 'start': 2414.678, 'title': 'Ai research and nonlinear dynamics in ai', 'summary': 'Covers ai research categorization, modularization and knowledge, inefficiency of linear operations, and the effectiveness of graph representations. it also discusses the importance of nonlinear dynamics in ai, limitations of modern neural networks, and the potential of hypergraphs.', 'chapters': [{'end': 2729.02, 'start': 2414.678, 'title': 'Ai research and knowledge representation', 'summary': 'Discusses the categorization of ai researchers into different tribes, the primacy of modularization and knowledge, and the inefficiency of linear operations on high-dimensional vector spaces in ai systems. it also explores the effectiveness of graph representations for interconnecting different systems.', 'duration': 314.342, 'highlights': ["The chapter discusses the categorization of AI researchers into different tribes, including evolutionary, logic-based, and empirical people. Pedro Domingo's classification of AI into five tribes is mentioned, revealing the diversity in AI research approaches.", 'The inefficiency of linear operations on high-dimensional vector spaces in AI systems is emphasized, suggesting that these operations are not the right algebra for most general intelligence tasks. The discussion points out that most neural net algorithms in common commercial use today are based on matrix math and linear operations, which may not be suitable for general intelligence tasks.', 'The effectiveness of graph representations for interconnecting different systems in AI is highlighted, emphasizing the value of graph, hypergraph, and metagraph representations for AGI and narrow AI systems. The advocate of graph, hypergraph, and metagraph representations for AGI and narrow AI systems is discussed, noting the inefficiency of linear operations on high-dimensional vector spaces for general intelligence tasks.']}, {'end': 3295.813, 'start': 2729.02, 'title': 'Nonlinear dynamics in ai', 'summary': 'Discusses the importance of nonlinear dynamics in ai, emphasizing the limitations of modern neural networks in capturing complex, nonlinear temporal dynamics and attractor formation, while highlighting the potential of hypergraphs in organizing knowledge for efficient execution of nonlinear operations.', 'duration': 566.793, 'highlights': ["Modern neural networks lack the ability to capture complex, nonlinear temporal dynamics and attractor formation, as emphasized by Walter Freeman's work on rabbit brains and Gary Lynch's research on the evolution of cognitive cortex.", 'The potential of hypergraphs in organizing knowledge for efficient execution of nonlinear operations is highlighted, suggesting their use for asynchronous, parallel execution of complex nonlinear transformations on a large body of knowledge.', 'Neural Darwinism, a theory of the evolution of neural sub-assemblies through natural selection, and its importance are discussed, pointing out the absence of its incorporation in modern neural architectures.']}], 'duration': 881.135, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY2414678.jpg', 'highlights': ["Pedro Domingo's classification reveals diversity in AI research approaches.", 'Linear operations are inefficient for general intelligence tasks in AI systems.', 'Graph representations are effective for interconnecting different systems in AI.', 'Modern neural networks lack the ability to capture complex, nonlinear temporal dynamics.', 'Hypergraphs have potential in organizing knowledge for efficient execution of nonlinear operations.', "Neural Darwinism theory's importance is discussed, highlighting its absence in modern neural architectures."]}, {'end': 4528.175, 'segs': [{'end': 3343.601, 'src': 'embed', 'start': 3295.913, 'weight': 0, 'content': [{'end': 3303.717, 'text': 'And simplistic attempts to use CMAES or other evolutionary algorithms to evolve neural nets are not the same right?', 'start': 3295.913, 'duration': 7.804}, {'end': 3308.98, 'text': "And then we're not getting what my friend Weaver, aka David Weinbaum,", 'start': 3304.278, 'duration': 4.702}, {'end': 3314.803, 'text': 'called open-ended intelligence in his PhD thesis at the Free University of Brussels,', 'start': 3308.98, 'duration': 5.823}, {'end': 3322.362, 'text': 'which is the ability that the brain the the radical level of plasticity that that the brain can show.', 'start': 3314.803, 'duration': 7.559}, {'end': 3328.766, 'text': "like my, my mother's partner lost 30 of her brain in a car crash literally the first year after that.", 'start': 3322.362, 'duration': 6.404}, {'end': 3330.348, 'text': 'she thought people were chickens.', 'start': 3328.766, 'duration': 1.582}, {'end': 3333.91, 'text': "i mean she didn't know idea what the was going on.", 'start': 3330.348, 'duration': 3.562}, {'end': 3338.873, 'text': 'she regained essentially all functions but lost some traumatic memories from her youth.', 'start': 3333.91, 'duration': 4.963}, {'end': 3343.601, 'text': 'but there was like radical rebuilding in in in the brain.', 'start': 3338.873, 'duration': 4.728}], 'summary': "Evolutionary algorithms fail to achieve open-ended intelligence; brain's radical plasticity after trauma, like regaining lost functions.", 'duration': 47.688, 'max_score': 3295.913, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY3295913.jpg'}, {'end': 3554.64, 'src': 'embed', 'start': 3524.075, 'weight': 3, 'content': [{'end': 3531.557, 'text': "And I mean, I think there's top-down dynamics and there's bottom-up dynamics going on.", 'start': 3524.075, 'duration': 7.482}, {'end': 3537.126, 'text': 'In terms of the role of sensory processing.', 'start': 3532.538, 'duration': 4.588}, {'end': 3539.848, 'text': "indeed, it it's.", 'start': 3537.126, 'duration': 2.722}, {'end': 3541.87, 'text': "it's complex and goes both ways right.", 'start': 3539.848, 'duration': 2.022}, {'end': 3549.155, 'text': 'so data comes through sensory processing and then, in cases where sensory processing is too ambiguous,', 'start': 3541.87, 'duration': 7.285}, {'end': 3551.597, 'text': 'cognitive processing is used for disambiguation.', 'start': 3549.155, 'duration': 2.442}, {'end': 3554.64, 'text': 'you can have multiple cycles of feedback.', 'start': 3551.597, 'duration': 3.043}], 'summary': 'Sensory and cognitive processing interact bidirectionally, leading to multiple feedback cycles.', 'duration': 30.565, 'max_score': 3524.075, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY3524075.jpg'}, {'end': 3609.691, 'src': 'embed', 'start': 3580.986, 'weight': 2, 'content': [{'end': 3587.472, 'text': "but there's some at least topological, if not geometric image in my, in my mind, there right.", 'start': 3580.986, 'duration': 6.486}, {'end': 3598.821, 'text': 'so clearly my brain must be projecting stuff from cognitive cortex into visual cortex and using sensory manipulate,', 'start': 3587.472, 'duration': 11.349}, {'end': 3609.691, 'text': 'using matrices and rotations and like sensory knowledge manipulations as a crutch to help it sort through some other type of complex cognitive reasoning.', 'start': 3598.821, 'duration': 10.87}], 'summary': 'Brain projects images using cognitive cortex into visual cortex for complex reasoning.', 'duration': 28.705, 'max_score': 3580.986, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY3580986.jpg'}, {'end': 4085.702, 'src': 'embed', 'start': 4060.182, 'weight': 4, 'content': [{'end': 4071.331, 'text': 'did a bunch of work connecting the current version of OpenCog pre-Hyperon with neural models in Torch for vision and language processing,', 'start': 4060.182, 'duration': 11.149}, {'end': 4082.055, 'text': "where you're sort of doing symbolic reasoning on graph nodes that represent layers or other portions of the neural net.", 'start': 4071.331, 'duration': 10.724}, {'end': 4085.702, 'text': "And so I think I'm not actually trying to do everything inside the graph.", 'start': 4082.216, 'duration': 3.486}], 'summary': 'Connected opencog pre-hyperon with neural models in torch for vision and language processing, implementing symbolic reasoning on graph nodes.', 'duration': 25.52, 'max_score': 4060.182, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY4060182.jpg'}], 'start': 3295.913, 'title': 'Brain plasticity and cognitive processes', 'summary': "Explores brain plasticity after a car crash, the brain's modularity and perception, cognitive synergy, and the integration of opencog with neural models in torch for vision and language processing.", 'chapters': [{'end': 3343.601, 'start': 3295.913, 'title': 'Evolutionary algorithms and brain plasticity', 'summary': 'Discusses the limitations of using evolutionary algorithms for evolving neural nets and emphasizes the concept of open-ended intelligence, as demonstrated by a personal anecdote of brain plasticity after a car crash resulting in significant rebuilding and recovery.', 'duration': 47.688, 'highlights': ["The brain's radical level of plasticity and rebuilding after a car crash, resulting in the recovery of essential functions and loss of traumatic memories.", "The concept of open-ended intelligence, as described in David Weinbaum's PhD thesis, emphasizing the limitations of evolutionary algorithms for evolving neural nets."]}, {'end': 3679.793, 'start': 3343.601, 'title': "The brain's modularity and perception", 'summary': "Discusses the brain's modularity, the interplay between sensory and cognitive processing, and the potential impact on reasoning, highlighting the complexity and interconnectedness of these processes.", 'duration': 336.192, 'highlights': ["The brain's ability to rebuild its own goals as part of its evolutionary, non-linear dynamic is discussed, emphasizing the non-comprehensible nature of many activities driven by the brain's modularity, with a focus on the cortex hippocampus interaction.", "The interplay between sensory and cognitive processing is highlighted, showcasing the brain's projection of abstract mathematical reasoning into the visual and auditory domains, indicating a strong linkage between cognition and sensory domains.", "The discussion on different schools of thought in cognitive architectures, including the surprising effectiveness of trivial systems like deep reinforcement learning, emphasizes the interconnectedness of reasoning, perception, and general intelligence, showcasing the brain's adaptability and repurposing capabilities in response to events like a stroke.", 'The potential impact of sensory processing on cognitive reasoning and the role of perception in creating representations for downstream reasoning is examined, showcasing the complexity and bidirectional nature of these processes, with a focus on the role of sensory processing in disambiguation and creative imagination.']}, {'end': 4034.902, 'start': 3681.437, 'title': 'Cognitive synergy and graph operations', 'summary': "Discusses the relationship between sensory knowledge and cognitive domain, the continuous nature of the brain's communication, and the efficient algebraic operations supported by knowledge hypergraphs in the opencog system.", 'duration': 353.465, 'highlights': ["The brain's communication occurs in the form of attractors of continuous variables, and the communication is primarily based on continuous variables rather than discrete variables. The brain's communication occurs in the form of attractors of continuous variables, and the communication is primarily based on continuous variables rather than discrete variables.", 'Efficient algebraic operations supported by knowledge hypergraphs in the OpenCog system include pattern matching against graphs or hypergraphs and fold operations, such as catamorphisms and anamorphisms, familiar to functional programmers. Efficient algebraic operations supported by knowledge hypergraphs in the OpenCog system include pattern matching against graphs or hypergraphs and fold operations, such as catamorphisms and anamorphisms, familiar to functional programmers.', 'The chapter also touches on the debate between continuous and discrete variables in building human-level general intelligence in a digital computer. The chapter also touches on the debate between continuous and discrete variables in building human-level general intelligence in a digital computer.']}, {'end': 4528.175, 'start': 4034.902, 'title': 'Neural networks and knowledge graphs', 'summary': 'Discusses the integration of opencog with neural models in torch for vision and language processing, the importance of forgetting in ai systems, and the debate surrounding analog computing and quantum gravity in relation to human cognition.', 'duration': 493.273, 'highlights': ['The integration of OpenCog with neural models in Torch for vision and language processing, allowing symbolic reasoning on graph nodes, is discussed, highlighting the efficiency and transparency of Torch compute graph. (Relevance Score: 5)', 'The importance of forgetting in AI systems is emphasized, with reference to the aphorisms of Pei Wang, emphasizing the need for knowledge graphs to not contain everything. (Relevance Score: 4)', 'The debate surrounding analog computing and quantum gravity in relation to human cognition, including the hypotheses made by Penrose and the potential for transferring cosmic inspiration in human cognition, is explored. (Relevance Score: 3)']}], 'duration': 1232.262, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY3295913.jpg', 'highlights': ["The brain's radical level of plasticity and rebuilding after a car crash, resulting in the recovery of essential functions and loss of traumatic memories.", "The concept of open-ended intelligence, as described in David Weinbaum's PhD thesis, emphasizing the limitations of evolutionary algorithms for evolving neural nets.", "The interplay between sensory and cognitive processing, showcasing the brain's projection of abstract mathematical reasoning into the visual and auditory domains, indicating a strong linkage between cognition and sensory domains.", 'The potential impact of sensory processing on cognitive reasoning and the role of perception in creating representations for downstream reasoning, showcasing the complexity and bidirectional nature of these processes, with a focus on the role of sensory processing in disambiguation and creative imagination.', 'The integration of OpenCog with neural models in Torch for vision and language processing, allowing symbolic reasoning on graph nodes, is discussed, highlighting the efficiency and transparency of Torch compute graph.']}, {'end': 5077.275, 'segs': [{'end': 4604.366, 'src': 'embed', 'start': 4560.623, 'weight': 0, 'content': [{'end': 4567.286, 'text': "I mean, Gödel's second incompleteness theorem is like you can't be reasonably powerful, consistent, and complete.", 'start': 4560.623, 'duration': 6.663}, {'end': 4577.271, 'text': 'But for better or worse, we bypass that by being like wildly logically inconsistent dynamical systems in the first place.', 'start': 4567.366, 'duration': 9.905}, {'end': 4582.274, 'text': 'But the other thing is your digital computer systems are not sitting in a vat.', 'start': 4577.632, 'duration': 4.642}, {'end': 4586.541, 'text': "They're engaged in practice with the outside world,", 'start': 4582.334, 'duration': 4.207}, {'end': 4595.123, 'text': 'which is largely incomprehensible to us and presumably has massively greater algorithmic information than than any of us.', 'start': 4586.541, 'duration': 8.582}, {'end': 4596.144, 'text': 'so you really,', 'start': 4595.123, 'duration': 1.021}, {'end': 4603.986, 'text': "you have to look at both humans and digital computers as somehow modulating the self-organizing complexity of the world they're interacting with.", 'start': 4596.144, 'duration': 7.842}, {'end': 4604.366, 'text': "right it's?", 'start': 4603.986, 'duration': 0.38}], 'summary': "Gödel's theorem: systems are logically inconsistent. digital systems interact with complex world.", 'duration': 43.743, 'max_score': 4560.623, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY4560623.jpg'}, {'end': 4728.961, 'src': 'embed', 'start': 4705.304, 'weight': 2, 'content': [{'end': 4714.731, 'text': "So you've said that Hutter's universal AI theory, AIXI the main article of faith at DeepMind, by the way serves as a credible, although debatable,", 'start': 4705.304, 'duration': 9.427}, {'end': 4719.595, 'text': 'theoretic approach to AGI, and that several proto-AGI systems are emerging from it already.', 'start': 4714.731, 'duration': 4.864}, {'end': 4721.236, 'text': "I'm not sure which ones you meant, but we'll get to that in a sec.", 'start': 4719.615, 'duration': 1.621}, {'end': 4728.961, 'text': 'But your core belief here is that intelligence is closely tied to the creation of procedures that achieve goals and environments in the simplest possible way.', 'start': 4721.776, 'duration': 7.185}], 'summary': "Hutter's aixi theory serves as a credible approach to agi, giving rise to proto-agi systems already.", 'duration': 23.657, 'max_score': 4705.304, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY4705304.jpg'}, {'end': 4773.79, 'src': 'embed', 'start': 4742.73, 'weight': 3, 'content': [{'end': 4743.49, 'text': 'I mean, oh man.', 'start': 4742.73, 'duration': 0.76}, {'end': 4747.073, 'text': 'Well, that may well be the case.', 'start': 4744.391, 'duration': 2.682}, {'end': 4752.737, 'text': 'But for example, if it were possible to find an optimal agent, you might need to have an AGI already.', 'start': 4747.113, 'duration': 5.624}, {'end': 4756.159, 'text': 'But anyway, AIXI also assumes that the environment is computable.', 'start': 4752.797, 'duration': 3.362}, {'end': 4762.584, 'text': "And I think there's a huge gap between some of the very theoretical notions we talk about in AGI and reality.", 'start': 4756.6, 'duration': 5.984}, {'end': 4767.988, 'text': 'AGI researchers cite apparently incontrovertible and mathematically rigorous articles of faith,', 'start': 4763.044, 'duration': 4.944}, {'end': 4773.79, 'text': "whether it's AIXI or universal function approximation or the so-called Turing completeness of neural networks.", 'start': 4767.988, 'duration': 5.802}], 'summary': 'Agi researchers discuss theoretical concepts like aixi, which assumes computable environment, but note a gap with reality.', 'duration': 31.06, 'max_score': 4742.73, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY4742730.jpg'}], 'start': 4528.575, 'title': "Gödel's theorem, creativity, and agi evolution", 'summary': "Discusses gödel's theorem's impact on human and digital computer creativity, emphasizing inconsistency and engagement with the outside world. it also covers the evolution of agi, aixi theory application, limitations of expected reward maximization, and the importance of resource constraints and individuation-transcendence contradiction in building practical agi systems.", 'chapters': [{'end': 4634.612, 'start': 4528.575, 'title': "Gödel's theorem and creativity", 'summary': "Discusses the limitations of human and digital computer creativity due to gödel's theorem, emphasizing the inconsistency of humans and the engagement of digital computers with the incomprehensible outside world.", 'duration': 106.037, 'highlights': ["Humans bypass Gödel's theorem by being wildly logically inconsistent dynamical systems, unlike digital computers that blow up to obey the given axiom system.", "Digital computer systems engage in practice with the outside world, which has massively greater algorithmic information than any of us, modulating the self-organizing complexity of the world they're interacting with.", "The idea that Gödel's theorem limits digital computers from creativity makes no sense due to the wildly inconsistent nature of humans and the engagement of digital computers with the incomprehensible outside world."]}, {'end': 5077.275, 'start': 4634.612, 'title': 'Evolution of agi and limitations', 'summary': 'Discusses the evolution of agi, the shift in motivations of people entering the field, the application of aixi theory in agi, and the limitations of expected reward maximization in defining general intelligence, emphasizing the importance of resource constraints and the contradiction between individuation and transcendence in building practical agi systems.', 'duration': 442.663, 'highlights': ['The shift in motivations of people entering the AGI field from profound thinking about the mind to seeking high salaries and assuming Google has already figured it out is highlighted, reflecting a remarkable change in the orientation of individuals in the field.', 'The application of AIXI theory as a credible, although debatable, theoretic approach to AGI and the emergence of proto-AGI systems from it are mentioned, suggesting the significance of this theory in the AGI landscape.', 'The limitations of AIXI, particularly its requirement of infinite computing power and its focus on optimizing expected reward within a specific context, are discussed, highlighting the practical constraints in achieving open-ended intelligence in AGI systems.', 'The discussion of the contradiction between individuation and transcendence in building practical AGI systems, as well as the emphasis on resource constraints and the limitations of maximizing expected reward, provides insights into the challenges and considerations in developing practical AGI systems.']}], 'duration': 548.7, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY4528575.jpg', 'highlights': ['Digital computers engage with the outside world, modulating its complexity.', "Humans bypass Gödel's theorem by being wildly logically inconsistent.", 'The application of AIXI theory as a credible theoretic approach to AGI is mentioned.', 'The discussion of the contradiction between individuation and transcendence in building practical AGI systems provides insights into the challenges and considerations in developing practical AGI systems.']}, {'end': 6787.527, 'segs': [{'end': 5153.786, 'src': 'embed', 'start': 5124.774, 'weight': 0, 'content': [{'end': 5126.935, 'text': "Because I'm really trying to understand this.", 'start': 5124.774, 'duration': 2.161}, {'end': 5128.637, 'text': 'This reward function?', 'start': 5127.616, 'duration': 1.021}, {'end': 5131.96, 'text': 'it has to be a function of the environment, but where does it come from?', 'start': 5128.637, 'duration': 3.323}, {'end': 5133.661, 'text': "Why isn't it deceptive??", 'start': 5131.98, 'duration': 1.681}, {'end': 5140.067, 'text': 'Why is it a smart idea to pluck a reward function out of thin air and optimize on it and then say', 'start': 5133.982, 'duration': 6.085}, {'end': 5147.2, 'text': "Well, it probably isn't, but Marcus Söder didn't make that up, right? I mean, he's just putting statistical..", 'start': 5140.267, 'duration': 6.933}, {'end': 5153.786, 'text': 'decision theory together with algorithmic, algorithmic information theory.', 'start': 5148.922, 'duration': 4.864}], 'summary': 'Analyzing the origin and reliability of reward functions in decision-making algorithms, integrating statistical decision theory and algorithmic information theory.', 'duration': 29.012, 'max_score': 5124.774, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY5124774.jpg'}, {'end': 5318.214, 'src': 'embed', 'start': 5293.801, 'weight': 1, 'content': [{'end': 5300.124, 'text': 'But no, anyway, he came along and he said, cause obviously another conception of intelligence is about flexibility.', 'start': 5293.801, 'duration': 6.323}, {'end': 5307.167, 'text': 'And he says that intelligence is about the ability to adapt to novelty using knowledge you already have as efficiently as possible.', 'start': 5300.544, 'duration': 6.623}, {'end': 5310.869, 'text': "And essentially it's about being able to create a skill program.", 'start': 5307.588, 'duration': 3.281}, {'end': 5318.214, 'text': "but in his formalism of intelligence, That's basically the same as Pei Wang's adapting to the environment.", 'start': 5310.869, 'duration': 7.345}], 'summary': 'Intelligence is the efficient adaptation to novelty with existing knowledge.', 'duration': 24.413, 'max_score': 5293.801, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY5293801.jpg'}, {'end': 5913.71, 'src': 'embed', 'start': 5885.297, 'weight': 2, 'content': [{'end': 5891.622, 'text': 'So it was just very inspirational to me to see how these, like actual professors and serious thinkers,', 'start': 5885.297, 'duration': 6.325}, {'end': 5895.343, 'text': 'were thinking about how to really build ai systems.', 'start': 5891.622, 'duration': 3.721}, {'end': 5899.725, 'text': "but uh, i've diverged from douglas in various ways since then.", 'start': 5895.343, 'duration': 4.382}, {'end': 5900.425, 'text': 'i mean he.', 'start': 5899.725, 'duration': 0.7}, {'end': 5903.346, 'text': 'he thinks the singularity is kind of a bunch of bunk,', 'start': 5900.425, 'duration': 2.921}, {'end': 5913.71, 'text': "but he he also thinks current deep neural systems are completely unrelated to general intelligence and there's almost nothing to be learned from them except what not to do.", 'start': 5903.346, 'duration': 10.364}], 'summary': 'Professors and serious thinkers discuss ai systems; some divergence in opinions on singularity and deep neural systems.', 'duration': 28.413, 'max_score': 5885.297, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY5885297.jpg'}, {'end': 6397.539, 'src': 'embed', 'start': 6370.235, 'weight': 3, 'content': [{'end': 6382.963, 'text': "And so, together with IOHK, the company that works on the Cardano blockchain, we're working on what we're calling an AIDSL,", 'start': 6370.235, 'duration': 12.728}, {'end': 6386.826, 'text': 'which is a very rich sort of dependently typed functional language.', 'start': 6382.963, 'duration': 3.863}, {'end': 6392.577, 'text': 'for communication between different AI processes in this network.', 'start': 6387.574, 'duration': 5.003}, {'end': 6397.539, 'text': "So I mean, that's the vision and aspiration with SingularityNet.", 'start': 6392.677, 'duration': 4.862}], 'summary': 'Singularitynet is collaborating with iohk on an aidsl for ai communication in the cardano blockchain network.', 'duration': 27.304, 'max_score': 6370.235, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY6370235.jpg'}], 'start': 5077.775, 'title': 'Ai reward functions and general intelligence', 'summary': 'Examines the limitations of reward functions in ai, the concept of intelligence as adaptation and development, and the potential of singularitynet and ai dsl for achieving general intelligence, emphasizing the importance of fundamental development and the skepticism towards reward functions.', 'chapters': [{'end': 5277.608, 'start': 5077.775, 'title': 'Reward functions and agi limitations', 'summary': 'Discusses the limitations of using reward functions in ai, highlighting the changing nature of reward functions and the skepticism towards their effectiveness, along with the argument that almost any learning approach can lead to agi given enough resources and exploration of the search space.', 'duration': 199.833, 'highlights': ['The limitations of using reward functions in AI The speaker questions the origin and reliability of reward functions and expresses skepticism towards plucking a reward function out of thin air and optimizing it.', 'Changing nature of reward functions The speaker mentions that the reward functions he approximates have changed over the course of his life, indicating the dynamic nature of reward functions.', 'Skepticism towards the effectiveness of reward functions in achieving AGI The speaker argues that almost any learning approach can lead to AGI given enough resources and exploration of the search space, casting doubt on the significance of reinforcement learning in real-life scenarios.']}, {'end': 5863.78, 'start': 5277.628, 'title': 'Intelligence: adaptation and development', 'summary': 'Discusses the concept of intelligence as the ability to adapt to novelty using existing knowledge efficiently, taking into account experience, priors, and knowledge, and the importance of fundamental development in general intelligence, while also considering the limitations imposed by the laws of physics on general intelligence.', 'duration': 586.152, 'highlights': ["The concept of intelligence as the ability to adapt to novelty using existing knowledge efficiently, taking into account experience, priors, and knowledge. Intelligence is defined as the ability to adapt to novelty using existing knowledge efficiently, considering experience, priors, and knowledge. Francois Chollet's paper on the measure of intelligence is referenced.", 'Importance of fundamental development in general intelligence and the limitations imposed by the laws of physics on general intelligence. The discussion emphasizes the importance of fundamental development in general intelligence, particularly the aspect of learning and development in conjunction with the human analogy. It also considers the limitations imposed by the laws of physics on general intelligence.']}, {'end': 6370.094, 'start': 5863.8, 'title': 'Ai, singularitynet, and general intelligence', 'summary': 'Discusses the diverging views of douglas hofstadter and the speaker on ai, the singularity, and the singularitynet project, emphasizing the potential of deep neural models and the decentralized network approach for achieving general intelligence.', 'duration': 506.294, 'highlights': ["The speaker's diverging views from Douglas Hofstadter on AI and the Singularity, emphasizing the potential of deep neural models and the SingularityNet project. The speaker outlines his differing perspective from Douglas Hofstadter on AI and the Singularity, highlighting the potential of deep neural models and the SingularityNet project for achieving general intelligence.", 'The concept of a decentralized network approach for achieving general intelligence through different AI components communicating and combining on the internet. The chapter explores the concept of a decentralized network approach for achieving general intelligence, where different AI components communicate and combine on the internet, potentially leading to the emergence of general intelligence.', 'The potential integration of OpenCog Hyperon with SingularityNet, creating a network of interacting agents for achieving general intelligence. The speaker discusses the potential integration of OpenCog Hyperon with SingularityNet, envisioning a network of interacting agents that could lead to the achievement of general intelligence.']}, {'end': 6787.527, 'start': 6370.235, 'title': 'Singularitynet and aidsl', 'summary': "Discusses singularitynet's collaboration with iohk to develop an aidsl for communication between ai processes, the use of functional programming languages to automate ai tasks, and the potential of a meta reflective agi system.", 'duration': 417.292, 'highlights': ["SingularityNet's collaboration with IOHK to develop an AIDSL for communication between AI processes The collaboration aims to create a rich dependently typed functional language for communication between AI processes in the network.", 'The use of functional programming languages to automate AI tasks The discussion emphasizes using languages like Haskell or OCaml to enable automated reasoning and program transformation, reducing the need for human intervention in AI algorithm operations.', 'The potential of a meta reflective AGI system Exploring the path towards a meta reflective AGI system, leveraging cognitive architectures inspired by the human mind while embracing current computing infrastructure and programming languages.']}], 'duration': 1709.752, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY5077775.jpg', 'highlights': ['The limitations of using reward functions in AI The speaker questions the origin and reliability of reward functions and expresses skepticism towards plucking a reward function out of thin air and optimizing it.', "The concept of intelligence as the ability to adapt to novelty using existing knowledge efficiently, taking into account experience, priors, and knowledge. Intelligence is defined as the ability to adapt to novelty using existing knowledge efficiently, considering experience, priors, and knowledge. Francois Chollet's paper on the measure of intelligence is referenced.", "The speaker's diverging views from Douglas Hofstadter on AI and the Singularity, emphasizing the potential of deep neural models and the SingularityNet project. The speaker outlines his differing perspective from Douglas Hofstadter on AI and the Singularity, highlighting the potential of deep neural models and the SingularityNet project for achieving general intelligence.", "SingularityNet's collaboration with IOHK to develop an AIDSL for communication between AI processes The collaboration aims to create a rich dependently typed functional language for communication between AI processes in the network."]}, {'end': 7548.249, 'segs': [{'end': 7087.534, 'src': 'embed', 'start': 7048.925, 'weight': 2, 'content': [{'end': 7049.886, 'text': 'Yeah, I agree with you.', 'start': 7048.925, 'duration': 0.961}, {'end': 7054.83, 'text': "I think it's, I mean, it seems clear to me to just the magnitude of his knowledge.", 'start': 7049.906, 'duration': 4.924}, {'end': 7057.972, 'text': 'And also, by the way, pragmatic experience.', 'start': 7055.85, 'duration': 2.122}, {'end': 7065.418, 'text': 'I mean, though he has this huge background of theoretical knowledge and study, he also has been involved in building systems.', 'start': 7058.052, 'duration': 7.366}, {'end': 7072.123, 'text': "And in my opinion that's a huge benefit to anyone's learning is actually going out there and trying to practice, trying to build something,", 'start': 7065.498, 'duration': 6.625}, {'end': 7073.564, 'text': 'trying to do something in reality.', 'start': 7072.123, 'duration': 1.441}, {'end': 7074.625, 'text': "He's done that.", 'start': 7074.005, 'duration': 0.62}, {'end': 7076.066, 'text': "And so he's just got this huge..", 'start': 7074.905, 'duration': 1.161}, {'end': 7087.534, 'text': "mountain of knowledge here and when you ask them a question and we're trying to get to point Z and we're starting at a there's a long path there and I'm sure all of us have been in this situation when say,", 'start': 7077.407, 'duration': 10.127}], 'summary': 'Huge theoretical knowledge and pragmatic experience make him valuable.', 'duration': 38.609, 'max_score': 7048.925, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7048925.jpg'}, {'end': 7163.794, 'src': 'embed', 'start': 7135.243, 'weight': 1, 'content': [{'end': 7140.925, 'text': 'But there were a couple of things that I noticed, though, because I kind of got the impression with Ben at the beginning that,', 'start': 7135.243, 'duration': 5.682}, {'end': 7147.507, 'text': 'a little bit like Arai who we had on the podcast, he almost thinks that AI should just be this confection of everything but the kitchen sink.', 'start': 7140.925, 'duration': 6.582}, {'end': 7149.508, 'text': "Let's throw neural networks in there.", 'start': 7147.787, 'duration': 1.721}, {'end': 7152.069, 'text': "Let's throw logic systems in there.", 'start': 7149.548, 'duration': 2.521}, {'end': 7153.909, 'text': "Let's do evolutionary programming.", 'start': 7152.089, 'duration': 1.82}, {'end': 7155.63, 'text': 'Episodic memory, no problem.', 'start': 7154.229, 'duration': 1.401}, {'end': 7156.75, 'text': "That's a simulation.", 'start': 7155.65, 'duration': 1.1}, {'end': 7158.591, 'text': "Let's just kind of conflate everything together.", 'start': 7156.77, 'duration': 1.821}, {'end': 7163.794, 'text': 'and believe strongly in the emergence of some kind of AI on top of that.', 'start': 7159.191, 'duration': 4.603}], 'summary': "Ben's approach to ai includes various techniques like neural networks, logic systems, and evolutionary programming, aiming for the emergence of advanced ai.", 'duration': 28.551, 'max_score': 7135.243, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7135243.jpg'}, {'end': 7211.468, 'src': 'embed', 'start': 7174.78, 'weight': 0, 'content': [{'end': 7176.681, 'text': 'And so I think that, in my opinion,', 'start': 7174.78, 'duration': 1.901}, {'end': 7193.132, 'text': "I'm guessing that that leads him to just have a tremendous amount of understanding of the value in what's already been built by evolution and kind of the human mind that we need to utilize that if we want to make more rapid progress in AI,", 'start': 7176.681, 'duration': 16.451}, {'end': 7197.276, 'text': 'rather than just rediscovering all this stuff, you know, 500 years from now.', 'start': 7193.132, 'duration': 4.144}, {'end': 7201.019, 'text': 'On SingularityNet, Yannick made a video about SingularityNet.', 'start': 7198.377, 'duration': 2.642}, {'end': 7206.865, 'text': 'I assume that we all agree that something like SingularityNet is a core substrate for any future artificial general intelligence,', 'start': 7201.079, 'duration': 5.786}, {'end': 7208.687, 'text': 'with all of the caveats that we spoke about.', 'start': 7206.865, 'duration': 1.822}, {'end': 7210.668, 'text': 'you know is Is it really decentralized??', 'start': 7208.687, 'duration': 1.981}, {'end': 7211.468, 'text': "I'm not sure it is.", 'start': 7210.788, 'duration': 0.68}], 'summary': 'Utilize human understanding to make rapid progress in ai, singularitynet is core for future agi, with caveats.', 'duration': 36.688, 'max_score': 7174.78, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7174780.jpg'}], 'start': 6787.708, 'title': "Dr. ben gertzl's polymathic insights", 'summary': "Delves into dr. ben gertzl's polymathic nature, his insights on ai, emphasizing personal interaction, marketing unconventional projects, skepticism towards singularitynet, and the value of practical experience.", 'chapters': [{'end': 6985.921, 'start': 6787.708, 'title': "Dr. ben gertzl's insightful conversation", 'summary': 'Highlights the engaging conversation with dr. ben gertzl, emphasizing the significance of personal interaction in understanding complex concepts in ai, and his adeptness at marketing unconventional ai projects.', 'duration': 198.213, 'highlights': ['Dr. Ben Gertzl emphasized the importance of personal interaction in understanding complex concepts in AI, noting that direct communication with experts often leads to better comprehension and agreement, as opposed to solely relying on reading papers and watching talks.', "Dr. Ben Gertzl's adeptness at marketing unconventional AI projects, such as SingularityNET, Sophia, and OpenCog, was highlighted, with a specific mention of the attention garnered by Sophia due to its human-like appearance, underscoring the significance of emotional appeal in AI projects.", 'The conversation with Dr. Ben Gertzl showcased the value of engaging directly with experts, as it led to a deeper understanding of his insights, emphasizing the importance of personal interaction in comprehending complex AI concepts.']}, {'end': 7548.249, 'start': 6986.181, 'title': 'Ben: a unique polymath', 'summary': 'Discusses the polymathic nature of ben, his vast knowledge in various fields, involvement in building systems, skepticism towards singularitynet, and the challenges in describing and comparing ai models, while emphasizing the importance of practical experience in learning.', 'duration': 562.068, 'highlights': ["Ben's polymathic nature and vast knowledge in various fields Ben is recognized as a polymath with extensive theoretical knowledge and practical experience in building systems, garnering praise for his unique character and rare breed status.", 'Skepticism towards SingularityNet There is skepticism expressed about SingularityNet, questioning its decentralization, efficiency, and the challenges in describing and comparing AI models, indicating the need for improvement in these areas.', "The importance of practical experience in learning The significance of practical experience in learning is highlighted, emphasizing the value of going out and trying to build something in reality, as demonstrated by Ben's pragmatic experience."]}], 'duration': 760.541, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY6787708.jpg', 'highlights': ['Dr. Ben Gertzl emphasized the importance of personal interaction in understanding complex concepts in AI.', "Dr. Ben Gertzl's adeptness at marketing unconventional AI projects was highlighted, underscoring the significance of emotional appeal in AI projects.", 'Ben is recognized as a polymath with extensive theoretical knowledge and practical experience in building systems.', 'Skepticism expressed about SingularityNet, questioning its decentralization, efficiency, and the challenges in describing and comparing AI models.', 'The significance of practical experience in learning is highlighted, emphasizing the value of going out and trying to build something in reality.']}, {'end': 8158.762, 'segs': [{'end': 7595.601, 'src': 'embed', 'start': 7548.249, 'weight': 0, 'content': [{'end': 7554.939, 'text': "we're not going to use a singularity net in the future, for you know, doing our Excel spreadsheets.", 'start': 7548.249, 'duration': 6.69}, {'end': 7562.841, 'text': 'Like all of the things we do today, will still be handled largely by, you know, classic software.', 'start': 7555.299, 'duration': 7.542}, {'end': 7567.742, 'text': "I think we're going to have new modes of getting serviced by the internet,", 'start': 7562.841, 'duration': 4.901}, {'end': 7573.544, 'text': 'with some of the stuff we want to do where it is just going to be an acceptable mode of.', 'start': 7567.742, 'duration': 5.802}, {'end': 7573.964, 'text': 'you know this.', 'start': 7573.544, 'duration': 0.42}, {'end': 7577.667, 'text': "This is kind of what the network it's more like.", 'start': 7574.764, 'duration': 2.903}, {'end': 7582.591, 'text': 'if I go to 4chan and ask around to get help with a problem, right?', 'start': 7577.667, 'duration': 4.924}, {'end': 7590.077, 'text': "Sometimes I'm going to get a good, unexpected answer better than I could have come up with myself.", 'start': 7584.252, 'duration': 5.825}, {'end': 7595.601, 'text': "Sometimes I'm going to get some trolling, sometimes You know, like, it's just going to be different.", 'start': 7590.358, 'duration': 5.243}], 'summary': 'Future usage of singularity net not for excel, new internet modes emerging.', 'duration': 47.352, 'max_score': 7548.249, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7548249.jpg'}, {'end': 7634.857, 'src': 'embed', 'start': 7607.563, 'weight': 2, 'content': [{'end': 7614.545, 'text': "And I think we can't, right? Because I want to be able to test things and I want to understand its behavior and understand its inner workings.", 'start': 7607.563, 'duration': 6.982}, {'end': 7620.727, 'text': "Whereas another mindset is kind of like, well, you know, let's just sample this thing and, you know, it might work, it might not work.", 'start': 7614.825, 'duration': 5.902}, {'end': 7627.911, 'text': "Do you know what I mean? Sure, but if you don't want to use a product based on GPT-3, then don't.", 'start': 7620.787, 'duration': 7.124}, {'end': 7630.113, 'text': "Then don't use it.", 'start': 7628.452, 'duration': 1.661}, {'end': 7634.857, 'text': 'The question is would anyone want to use it?', 'start': 7632.235, 'duration': 2.622}], 'summary': 'Debating the potential use of gpt-3 and its value for testing and understanding behavior.', 'duration': 27.294, 'max_score': 7607.563, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7607563.jpg'}, {'end': 7896.575, 'src': 'embed', 'start': 7866.789, 'weight': 4, 'content': [{'end': 7872.553, 'text': 'diverse AI algorithms and get them to talk with each other using common representations.', 'start': 7866.789, 'duration': 5.764}, {'end': 7874.475, 'text': "That's exactly what they've been talking about doing.", 'start': 7872.914, 'duration': 1.561}, {'end': 7881.139, 'text': 'Yeah, but you got to pay careful attention to those little qualifying words in there like if, could, et cetera.', 'start': 7875.075, 'duration': 6.064}, {'end': 7882.8, 'text': "Some of it's done, some of it's not done.", 'start': 7881.239, 'duration': 1.561}, {'end': 7888.386, 'text': "I think the limiting factor is not whether it can be done or not, it's just whether it will work.", 'start': 7883.721, 'duration': 4.665}, {'end': 7896.575, 'text': "I think he said that the existing system was written some time ago, so there's a computational problem with scalability.", 'start': 7889.568, 'duration': 7.007}], 'summary': 'Challenges in integrating diverse ai algorithms for communication and scalability.', 'duration': 29.786, 'max_score': 7866.789, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7866789.jpg'}, {'end': 8074.956, 'src': 'embed', 'start': 8045.493, 'weight': 5, 'content': [{'end': 8048.636, 'text': "So whether it's an intelligence or not, I think we could debate that.", 'start': 8045.493, 'duration': 3.143}, {'end': 8051.278, 'text': 'But it is a complex dynamic system.', 'start': 8048.756, 'duration': 2.522}, {'end': 8058.803, 'text': 'And maybe you could think quite generally, as Kenneth Stanley does, that intelligence is a kind of information accumulation right?', 'start': 8051.838, 'duration': 6.965}, {'end': 8063.927, 'text': 'But I think the problem is brittleness, because when the primitives are,', 'start': 8059.103, 'duration': 4.824}, {'end': 8069.291, 'text': 'because he has this kind of type theory system of how the agents talk to each other.', 'start': 8063.927, 'duration': 5.364}, {'end': 8071.373, 'text': "And it's so specific.", 'start': 8069.791, 'duration': 1.582}, {'end': 8074.956, 'text': "The reason why you get emergence in the brain is because there's no brittleness.", 'start': 8071.813, 'duration': 3.143}], 'summary': 'Debate over intelligence as a complex system, with emphasis on brittleness and emergence.', 'duration': 29.463, 'max_score': 8045.493, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY8045493.jpg'}], 'start': 7548.249, 'title': 'Future internet services and gpt-3 limitations', 'summary': 'Discusses the future of internet services, predicting new modes of internet servicing and limitations of gpt-3 in business settings, highlighting its unsuitability for non-interactive, safety-critical tasks, and the potential for building better ai systems.', 'chapters': [{'end': 7595.601, 'start': 7548.249, 'title': 'Future of internet services', 'summary': 'Discusses the future of internet services, predicting that while traditional software will still handle most tasks, new modes of internet servicing will emerge, offering unexpected and potentially superior assistance in problem-solving.', 'duration': 47.352, 'highlights': ['The future of internet services will see traditional software still handling most tasks, with new modes of internet servicing emerging (e.g. Singularity net) for more personalized and potentially superior assistance in problem-solving.', 'Comparing the future internet services to 4chan, where seeking help can result in unexpected and potentially superior answers, as well as trolling and varying responses.']}, {'end': 8158.762, 'start': 7595.981, 'title': 'The limitations of gpt-3 in business settings', 'summary': "Discusses the limitations of using gpt-3 in a business setting, highlighting its unsuitability for non-interactive, safety-critical tasks, and the lack of certainty in the service's output, while exploring the potential for building better ai systems and the emergence of intelligence in complex dynamic systems.", 'duration': 562.781, 'highlights': ["The limitations of using GPT-3 in a business setting The chapter delves into the unsuitability of GPT-3 for business settings, emphasizing the need to test and understand its behavior and inner workings, and the uncertainty in the service's output, questioning its usability and desirability in such settings.", 'Potential for building better AI systems The discussion explores the potential for building better AI systems, emphasizing the need for extensive testing and the application of existing engineering paradigms to ensure the reliability and safety of AI systems.', 'Emergence of intelligence in complex dynamic systems The conversation touches on the emergence of intelligence in complex dynamic systems, raising philosophical questions about the emergence of intelligence and the brittleness of AI systems, while considering the free energy principle and uncertain processes as catalysts for intelligence.']}], 'duration': 610.513, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY7548249.jpg', 'highlights': ['New modes of internet servicing emerging for more personalized assistance in problem-solving.', 'Comparing future internet services to 4chan, seeking help can result in unexpected and potentially superior answers.', "The need to test and understand GPT-3's behavior and inner workings in business settings.", "Emphasizing the uncertainty in GPT-3's output, questioning its usability and desirability in business settings.", 'Exploring the potential for building better AI systems through extensive testing and application of existing engineering paradigms.', 'Raising philosophical questions about the emergence of intelligence in complex dynamic systems and the brittleness of AI systems.']}, {'end': 8891.073, 'segs': [{'end': 8272.023, 'src': 'embed', 'start': 8243.391, 'weight': 0, 'content': [{'end': 8249.253, 'text': "So I think where I kind of would challenge that is that I don't think prediction is binary zero or one.", 'start': 8243.391, 'duration': 5.862}, {'end': 8251.254, 'text': "I think there's kind of a continuum of prediction.", 'start': 8249.293, 'duration': 1.961}, {'end': 8259.337, 'text': 'And I would personally argue that bacteria as a bacterial system are predicting.', 'start': 8251.374, 'duration': 7.963}, {'end': 8264.659, 'text': 'They receive some inputs and they have some coding in there that tells them, I need to swim this way.', 'start': 8259.477, 'duration': 5.182}, {'end': 8272.023, 'text': "They're not consciously thinking, so I'm not saying they're consciously reasoning,", 'start': 8265.539, 'duration': 6.484}], 'summary': "Prediction is not binary; there's a continuum. bacteria can predict with inputs and coding.", 'duration': 28.632, 'max_score': 8243.391, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY8243391.jpg'}, {'end': 8341.62, 'src': 'embed', 'start': 8313.248, 'weight': 2, 'content': [{'end': 8314.949, 'text': "No, I don't think the Earth is predictive.", 'start': 8313.248, 'duration': 1.701}, {'end': 8317.51, 'text': "Like. I mean I don't think the Earth as a whole,", 'start': 8315.249, 'duration': 2.261}, {'end': 8326.353, 'text': "the giant ball like it's not an intelligence because it has no ability to control or react or predict its environment at all.", 'start': 8317.51, 'duration': 8.843}, {'end': 8328.475, 'text': "It's just following the simple physical laws.", 'start': 8326.374, 'duration': 2.101}, {'end': 8336.779, 'text': "You know, if an asteroid big enough hits it, it's going to melt and and and uh, end up with just a, you know, new volcanic surface or whatever,", 'start': 8328.514, 'duration': 8.265}, {'end': 8339.559, 'text': 'or you know, frozen lava surface.', 'start': 8336.779, 'duration': 2.78}, {'end': 8341.62, 'text': "But so my point is just, it's a spectrum.", 'start': 8339.62, 'duration': 2}], 'summary': 'The earth is not predictive, as it follows physical laws and can be impacted by external forces like asteroids.', 'duration': 28.372, 'max_score': 8313.248, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY8313248.jpg'}, {'end': 8379.013, 'src': 'embed', 'start': 8352.665, 'weight': 3, 'content': [{'end': 8364.128, 'text': "This is something that rightly or wrongly, I don't know why I'm skeptical about this, but AIXI depends on having a computable environment.", 'start': 8352.665, 'duration': 11.463}, {'end': 8368.69, 'text': "We've heard so many folks talk about intelligence from a different perspective,", 'start': 8364.528, 'duration': 4.162}, {'end': 8379.013, 'text': "whether it's Hawkins thinking about it as a general learning algorithm or people like Mark Bishop denying computationalism,", 'start': 8368.69, 'duration': 10.323}], 'summary': 'Aixi depends on a computable environment, skepticism about intelligence perspectives.', 'duration': 26.348, 'max_score': 8352.665, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY8352665.jpg'}], 'start': 8158.842, 'title': 'Predictive abilities and limits of intelligence', 'summary': 'Discusses the predictive abilities in bacterial systems, highlighting their limitations, and delves into the constraints of aixi in representing intelligence, the debate on computationalism, and the practical implications for building real-world intelligent systems.', 'chapters': [{'end': 8351.605, 'start': 8158.842, 'title': 'Predictive abilities in bacterial systems', 'summary': 'Discusses the continuum of prediction, arguing that bacterial systems exhibit minor predictive abilities, while highlighting the spectrum of prediction and the limitations of earth as a predictive entity.', 'duration': 192.763, 'highlights': ['Bacteria as predictive systems: Bacteria are discussed as predictive systems due to their ability to receive inputs and react based on programmed evolutionary coding, presenting a continuum of prediction from minor to complex forms.', 'Spectrum of prediction: The chapter emphasizes a spectrum of prediction, ranging from minor predictive abilities in bacterial systems to complex predictive abilities in humans, challenging the notion of prediction as a binary concept.', "Earth's limited predictive abilities: The Earth is described as lacking predictive abilities, being unable to control or predict its environment, and merely following physical laws, highlighting its limitations as a predictive entity."]}, {'end': 8891.073, 'start': 8352.665, 'title': 'Aixi and the limits of intelligence', 'summary': 'Discusses the limitations of aixi in representing intelligence as a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical ai approaches, such as aixi, for building real-world intelligent systems.', 'duration': 538.408, 'highlights': ['The limitations of AIXI in representing intelligence as a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems.', 'The concept that intelligence cannot be reduced to a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems.', 'The debate on computationalism and embodiment in defining intelligence, the limitations of AIXI in representing intelligence as a static program, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems.']}], 'duration': 732.231, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/sw8IE3MX1SY/pics/sw8IE3MX1SY8158842.jpg', 'highlights': ['Bacteria are discussed as predictive systems due to their ability to receive inputs and react based on programmed evolutionary coding, presenting a continuum of prediction from minor to complex forms.', 'The chapter emphasizes a spectrum of prediction, ranging from minor predictive abilities in bacterial systems to complex predictive abilities in humans, challenging the notion of prediction as a binary concept.', 'The Earth is described as lacking predictive abilities, being unable to control or predict its environment, and merely following physical laws, highlighting its limitations as a predictive entity.', 'The limitations of AIXI in representing intelligence as a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems.', 'The concept that intelligence cannot be reduced to a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems.']}], 'highlights': ['AIs reaching human-level intelligence by 2029 and surpassing human intelligence by 2045', 'The potential for creating intelligence beyond human capacity is acknowledged, indicating the awe and apprehension surrounding this advancement', 'The field of artificial intelligence was initially aimed at constructing thinking machines with human-like general intelligence, and it is essential to redirect efforts towards pursuing artificial general intelligence as a distinct endeavor from current narrow approaches', 'Ben Goertzel is a leading figure in the field of artificial general intelligence, spearheading influential projects like OpenCog, SingularityNet, and Sophia the Robot, and emphasizing the importance of organizing early stage AIs to be as beneficial as possible', 'Ben proposes a hybrid cognitive architecture influenced by cognitive psychology, emphasizing multimodal and distributed knowledge representation', 'The need for cognitive synergy in an AGI system arises from the requirement to achieve complex goals in a complex environment under limited compute resources, emphasizing the importance of achieving complex goals in a limited compute environment', 'The transition from expert system methodology to supervised learning was a smaller step than commonly understood, emphasizing the shift towards experiential and interactive learning', "The brain's radical level of plasticity and rebuilding after a car crash, resulting in the recovery of essential functions and loss of traumatic memories", 'The limitations of using reward functions in AI The speaker questions the origin and reliability of reward functions and expresses skepticism towards plucking a reward function out of thin air and optimizing it', 'The concept of intelligence as the ability to adapt to novelty using existing knowledge efficiently, taking into account experience, priors, and knowledge. Intelligence is defined as the ability to adapt to novelty using existing knowledge efficiently, considering experience, priors, and knowledge', 'Dr. Ben Gertzl emphasized the importance of personal interaction in understanding complex concepts in AI', 'New modes of internet servicing emerging for more personalized assistance in problem-solving', 'Bacteria are discussed as predictive systems due to their ability to receive inputs and react based on programmed evolutionary coding, presenting a continuum of prediction from minor to complex forms', 'The concept that intelligence cannot be reduced to a static program, the debate on computationalism and embodiment in defining intelligence, and the practical implications of theoretical AI approaches, such as AIXI, for building real-world intelligent systems']}