Marcus Hutter is a senior research scientist at DeepMind and professor at Australian National University. Throughout his career of research, including with Jürgen Schmidhuber and Shane Legg, he has proposed a lot of interesting ideas in and around the field of artificial general intelligence, including the development of the AIXI model which is a mathematical approach to AGI that incorporates ideas of Kolmogorov complexity, Solomonoff induction, and reinforcement learning. EPISODE LINKS: Hutter Prize: http://prize.hutter1.net Marcus web: http://www.hutter1.net Books mentioned: - Universal AI: https://amzn.to/2waIAuw - AI: A Modern Approach: https://amzn.to/3camxnY - Reinforcement Learning: https://amzn.to/2PoANj9 - Theory of Knowledge: https://amzn.to/3a6Vp7x This conversation is part of the Artificial Intelligence podcast. 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. This episode is presented by Cash App. Download it (App Store, Google Play), use code "LexPodcast". 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: 00:00 - Introduction 03:32 - Universe as a computer 05:48 - Occam's razor 09:26 - Solomonoff induction 15:05 - Kolmogorov complexity 20:06 - Cellular automata 26:03 - What is intelligence? 35:26 - AIXI - Universal Artificial Intelligence 1:05:24 - Where do rewards come from? 1:12:14 - Reward function for human existence 1:13:32 - Bounded rationality 1:16:07 - Approximation in AIXI 1:18:01 - Godel machines 1:21:51 - Consciousness 1:27:15 - AGI community 1:32:36 - Book recommendations 1:36:07 - Two moments to relive (past and future)
AI researcher and co-founder of DeepMind, Shane Legg, discusses his proposed approach to artificial general intelligence, the A-I-X-I model, and the potential of benchmarks such as the Hutter Prize to drive progress in developing AGI systems. The transcript does not contain any advertisements.
Occam's razor suggests that when given multiple models that equally describe a phenomenon or data, one should choose the simpler one. However, humans have a natural bias towards finding patterns even if they are not present, which can affect the decision-making process.
The universe and all its different objects represent a range of Kolmogorov complexities. Without noise, the whole universe can potentially be described by the standard model plus generativity.
In this episode, the guest talks about the emergence of complexity in Game of Life, cellular automata, and the universe. The possibility of reverse-engineering the short program that generates fractals is also discussed.
In this episode, the speaker discusses the concept of general artificial intelligence (AGI) and whether it refers only to superhuman intelligence or extends to sub-human intelligence as well. The speaker explores if AGI systems can perform well in multiple environments and how they associate with rational intelligence.
Dr. Marcus Hutter discusses the challenges in formalizing artificial general intelligence (AGI) and how his mathematical framework, AIXI, attempts to solve them by creating an intelligent agent that can perform well in any environment it finds itself in.
The process of prediction in machine learning goes beyond passive observation; it involves taking action within an environment to learn a model. The goal is to obtain simple programs that accurately predict future observations based on past data and interactions.
The Bayesian framework involves assigning a priori probability to any given stochastic program and evaluating what policies or action sequences lead to the maximum reward sum in expectation by replacing the true distribution with a universal distribution. The reward signal is occasionally given to the agent to maximize the reward sum, while avoiding greedy approaches.
The podcast hosts discuss pushing the horizon back and extending experience in AI. They use chess as an example to explain the importance of time steps in training AI to become better at decision-making.
The idea that solving AGI will result in solving all other related problems is appealing to many people, leading to a comparison of AGI and the theory of everything. AGI is a beautiful mathematical framework, and some aspects of it could be derived to solve other problems.
The Bayesian framework with prior of possible worlds and long-term planning can provide optimal decision making with the right amount of exploration, important for simple problems such as the bandit problem.
The speaker discusses the applications of reinforcement learning to improve elevator efficiency, stating that reward systems based on wait time can be used to maximize user satisfaction.
This podcast episode discusses the idea of injecting exploration for its own sake in AI reward functions, and how it could lead to more intelligent systems.
The display of consciousness in an intelligent system like ICSI does not necessarily mean it is truly conscious, but if it behaves in a way close enough to humans or even dogs, the distinction of the hard problem of consciousness may not be interesting from an AGI perspective. The challenge now is to improve intelligence with computational resources, which has yet to be done satisfactorily.
The most promising approach towards building an AGI system is having an agent that interacts with a 3D simulated environment, like in many computer games, rather than formalizing intelligence. Building an AGI system for industrial or near-term applications may not require robotics at this stage.
Simulated 3D environments can be useful in AI development, especially for agents that need to interact with humans, but for abstract agents focused on mathematical reasoning, it may not be necessary. A recommended book on complex ideas and compression is mentioned.
Lex Friedman concludes his conversation with Marcus Hutter with a discussion of the meaning of life and shares words of wisdom from Albert Einstein. The episode is sponsored by Cash App and includes a promo code for listeners to support a STEM education organization.