Episode
#115 – Dileep George: Brain-Inspired AI
2:10:33
Published: Fri Aug 14 2020
Description
Dileep George is a researcher at the intersection of neuroscience and artificial intelligence, co-founder of Vicarious, formerly co-founder of Numenta. From the early work on Hierarchical temporal memory to Recursive Cortical Networks to today, Dileep's always sought to engineer intelligence that is closely inspired by the human brain. Support this channel by supporting our sponsors. Click links, get discount: - Babbel: https://babbel.com and use code LEX - MasterClass: https://masterclass.com/lex - Raycon: https://buyraycon.com/lex If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter, LinkedIn, Facebook, Medium, or YouTube where you can watch the video versions of these conversations. If you enjoy the podcast, please rate it 5 stars on Apple Podcasts, follow on Spotify, or support it on Patreon. Here's the outline of the episode. On some podcast players you should be able to click the timestamp to jump to that time. OUTLINE: 0:00 - Introduction 4:50 - Building a model of the brain 17:11 - Visual cortex 27:50 - Probabilistic graphical models 31:35 - Encoding information in the brain 36:56 - Recursive Cortical Network 51:09 - Solving CAPTCHAs algorithmically 1:06:48 - Hype around brain-inspired AI 1:18:21 - How does the brain learn? 1:21:32 - Perception and cognition 1:25:43 - Open problems in brain-inspired AI 1:30:33 - GPT-3 1:40:41 - Memory 1:45:08 - Neuralink 1:51:32 - Consciousness 1:57:59 - Book recommendations 2:06:49 - Meaning of life
Chapters
Dilip George, a researcher at the intersection of neuroscience and artificial intelligence, discusses with Scott Phoenix and Dana Dubinsky the potential usefulness of understanding the fundamental principles of the human brain in engineering intelligence.
00:00 - 03:05 (03:05)
Summary
Dilip George, a researcher at the intersection of neuroscience and artificial intelligence, discusses with Scott Phoenix and Dana Dubinsky the potential usefulness of understanding the fundamental principles of the human brain in engineering intelligence.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The Blue Brain project aims to simulate the brain by putting details of neuroscience experiments into a giant simulation, without completely understanding its functions.
03:06 - 14:49 (11:42)
Summary
The Blue Brain project aims to simulate the brain by putting details of neuroscience experiments into a giant simulation, without completely understanding its functions. However, simulating the brain without understanding its complex functionality may not lead to a functioning brain simulation.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The complexity of studying the brain can be assisted with the use of AI models by determining how specific layers of the brain encode information and contribute to the final product.
14:49 - 24:21 (09:32)
Summary
The complexity of studying the brain can be assisted with the use of AI models by determining how specific layers of the brain encode information and contribute to the final product. This insight can be used in both neuroscience research and AI development.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The iterative procedure of projecting a model on evidence and taking evidence back into the model helps in constantly hallucinating the explanation of the world based on the model.
24:21 - 31:57 (07:35)
Summary
The iterative procedure of projecting a model on evidence and taking evidence back into the model helps in constantly hallucinating the explanation of the world based on the model. The feedback mechanism used helps explain the evidence using the world model.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The connections between cortical columns store knowledge of how different concepts connect to each other, implementing actual computations for inference through a structured cortical neural network.
31:57 - 38:47 (06:50)
Summary
The connections between cortical columns store knowledge of how different concepts connect to each other, implementing actual computations for inference through a structured cortical neural network.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The compatibility between perception and cognition systems is crucial in developing a system that understands conceptual models of the world.
38:47 - 47:28 (08:40)
Summary
The compatibility between perception and cognition systems is crucial in developing a system that understands conceptual models of the world. The RCA network is an example of how the background and foreground of an image are modeled separately, making the top-down controllability of the generative model apparent.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
This episode discusses the representation of prior knowledge in models for efficient and distributed inference, comparing it to convolutional neural networks and their feed forward processing cascade.
47:28 - 51:26 (03:57)
Summary
This episode discusses the representation of prior knowledge in models for efficient and distributed inference, comparing it to convolutional neural networks and their feed forward processing cascade.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
Captchas have evolved, but are still a good challenge problem for understanding how human perception works.
51:26 - 55:22 (03:55)
Summary
Captchas have evolved, but are still a good challenge problem for understanding how human perception works.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
Dynamic inference is an important aspect of reasoning with artificial neural networks and can be done through amortized inference to allow for flexibility and adaptability.
55:22 - 1:00:57 (05:35)
Summary
Dynamic inference is an important aspect of reasoning with artificial neural networks and can be done through amortized inference to allow for flexibility and adaptability. This allows for dynamic inference instead of solely relying on fixed combinations shown during training.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
This podcast episode discusses benchmarking deep learning algorithms with limited training data, including the inverted MNIST problem and the errors made by the network, which are similar to those made by humans in a similar scenario.
1:00:57 - 1:06:06 (05:08)
Summary
This podcast episode discusses benchmarking deep learning algorithms with limited training data, including the inverted MNIST problem and the errors made by the network, which are similar to those made by humans in a similar scenario.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The speaker discusses the importance of feedback connections, recursive inference, lateral connections, and coming to the best explanation of a scene as the problems to solve while also tackling recognition, segmentation, top-down attention, and higher level cognition.
1:06:06 - 1:12:58 (06:52)
Summary
The speaker discusses the importance of feedback connections, recursive inference, lateral connections, and coming to the best explanation of a scene as the problems to solve while also tackling recognition, segmentation, top-down attention, and higher level cognition.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The debate on the importance of biological plausibility in AI and neural networks is discussed with regards to the brain's mechanisms and their relationship to achieving efficiency.
1:12:58 - 1:18:50 (05:52)
Summary
The debate on the importance of biological plausibility in AI and neural networks is discussed with regards to the brain's mechanisms and their relationship to achieving efficiency.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The speaker discusses the importance of understanding the cognitive architecture and credit assignment in learning systems.
1:18:50 - 1:23:38 (04:47)
Summary
The speaker discusses the importance of understanding the cognitive architecture and credit assignment in learning systems. They consider several factors including how learning dynamics are spiking or looping, the role of regularization in injecting robustness, and the type of system (mechanical, electrical, or chemical) involved in the learning process.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The visual and motor systems are important prerequisites for any real language understanding, as they help in building the concept system.
1:23:38 - 1:30:49 (07:11)
Summary
The visual and motor systems are important prerequisites for any real language understanding, as they help in building the concept system. Perception system needs to be built in the right way to access the language-based knowledge stored in visual and motor systems.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
This episode explores the use of Markov chains and higher order models in text compression and modeling.
1:30:50 - 1:39:10 (08:19)
Summary
This episode explores the use of Markov chains and higher order models in text compression and modeling. The conversation touches on the limitations of models based on Markov chains, as well as the potential for more sophisticated models to capture key abstractions and common sense knowledge.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The memory reinstatement process plays a crucial role in building world models and interpreting current experiences in hierarchical ways.
1:39:10 - 1:44:39 (05:28)
Summary
The memory reinstatement process plays a crucial role in building world models and interpreting current experiences in hierarchical ways. The hippocampus enables us to replay past experiences and connect concepts, while the cortex helps in evaluating and forming memories.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
Brain-computer interfaces have direct applications for people with different conditions in the short-term, but the long-term possibilities include high bandwidth communication with AI systems.
1:44:39 - 1:48:38 (03:58)
Summary
Brain-computer interfaces have direct applications for people with different conditions in the short-term, but the long-term possibilities include high bandwidth communication with AI systems. Machine learning and the computer side will adjust and interact with the brain.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
Exploring the idea of connecting the brain to machines and the internet, and the potential difficulty in figuring out the basic communication protocols of the brain.
1:48:38 - 1:52:35 (03:56)
Summary
Exploring the idea of connecting the brain to machines and the internet, and the potential difficulty in figuring out the basic communication protocols of the brain.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The hard problem of consciousness proposes that this whole process feels like something and the existence proof suggests that death may not be a fundamental property of an intelligent system.
1:52:35 - 1:58:09 (05:34)
Summary
The hard problem of consciousness proposes that this whole process feels like something and the existence proof suggests that death may not be a fundamental property of an intelligent system.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
A conversation about impactful books leads to the recommendation of Judea Paul's "The Mind's Eye," a book that discusses the brain's thinking processes and resource management, which the speaker bought but hasn't read yet as an electronic version is not available.
1:58:09 - 2:03:06 (04:56)
Summary
A conversation about impactful books leads to the recommendation of Judea Paul's "The Mind's Eye," a book that discusses the brain's thinking processes and resource management, which the speaker bought but hasn't read yet as an electronic version is not available.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
The openness of cognitive neuroscience provides interesting opportunities to generate goals in understanding the brain and cognition.
2:03:06 - 2:08:51 (05:45)
Summary
The openness of cognitive neuroscience provides interesting opportunities to generate goals in understanding the brain and cognition. While experimental neuroscience might not be the best approach to build a brain, it can be a useful step towards pursuing computer science or neuroscience.
Episode#115 – Dileep George: Brain-Inspired AI
PodcastLex Fridman Podcast
In this podcast, Lex Friedman talks with Deleak George about entrepreneurship, mental health and personal development.
2:08:54 - 2:10:09 (01:14)
Summary
In this podcast, Lex Friedman talks with Deleak George about entrepreneurship, mental health and personal development. They also discuss the importance of finding one's own philosophy and values.