What Kind of Computation Is Cognition?

Yale University
15 Mar 202278:09

TLDRIn a talk at Yale, Professor Josh Tenenbaum explores the nature of cognition as computation. He discusses the reverse-engineering approach to understanding the human mind by building models akin to AI systems, focusing on intuitive physics and psychology. Tenenbaum emphasizes the importance of integrating neural network pattern recognition with symbolic languages and probabilistic inference to capture human common sense. He also highlights the challenges in cognitive science and AI, including the current limitations in self-driving cars and the promise of probabilistic programming in advancing our understanding of the mind.

Takeaways

  • ๐ŸŒŸ Professor Josh Tenenbaum, a computational cognitive scientist at MIT, discusses the nature of cognition and its computational aspects in a seminar at Yale.
  • ๐Ÿค” The central question of the talk is 'What kind of computation is cognition?', aiming to explore the computational processes underlying human thinking.
  • ๐Ÿ† Tenenbaum's work and genius were recognized by the MacArthur Foundation in 2019, highlighting his contributions to the field of cognitive science.
  • ๐Ÿ“Š Tenenbaum presents the 'reverse engineering approach' to understanding cognition, which involves characterizing the human mind using terms from intelligent machine engineering.
  • ๐Ÿค– He discusses the limitations of current AI technologies, emphasizing the gap between artificial intelligence and the flexible, general-purpose intelligence humans possess.
  • ๐Ÿš— The discussion includes the challenges in developing self-driving cars, illustrating the long tail of problems that arise even after significant investment and progress.
  • ๐Ÿง  Tenenbaum touches on the importance of understanding the brain's computations and how they might be represented in computational models.
  • ๐ŸŽฏ The goal is to build models that not only match human behavior qualitatively but also solve real-world problems, providing a quantitative account of human cognition.
  • ๐Ÿ“ˆ He introduces the concept of 'probabilistic programs' which combine neural network pattern recognition, symbolic languages for knowledge representation, and probabilistic inference.
  • ๐ŸŽฎ Tenenbaum also brings up the idea of a 'game engine in your head', suggesting that our minds use similar principles to game engines to simulate and predict outcomes.
  • ๐Ÿ“š The talk concludes with a call for further research into how these computational models can be instantiated in the brain and the implications for understanding human learning and intelligence.

Q & A

  • What is the main topic of Josh Tenenbaum's talk?

    -The main topic of Josh Tenenbaum's talk is the nature of cognition as a form of computation, specifically exploring the question, 'What kind of computation is cognition?'

  • How does Tenenbaum approach the study of cognition?

    -Tenenbaum approaches the study of cognition using the reverse-engineering approach, which involves characterizing how the human mind works in terms that would be used to engineer an intelligent machine, focusing on natural intelligence rather than artificial intelligence.

  • What does Tenenbaum mean by 'reverse engineering' in the context of cognitive science?

    -In the context of cognitive science, 'reverse engineering' refers to the process of trying to characterize how the human mind works using the same terms and principles that would be applied to engineer an intelligent machine, with the goal of building models that resemble AI systems but are focused on natural intelligence.

  • What are the key components of the 'game engine in your head' concept?

    -The 'game engine in your head' concept involves using tools from game engines, such as physics engines, to create a rich, immersive, interactive experience within the mind. This includes models of physical objects, agents with goals and beliefs, and the ability to simulate and predict outcomes based on these models.

  • How does Tenenbaum's research on intuitive physics and intuitive psychology relate to the study of metaphysics?

    -Tenenbaum's research on intuitive physics and intuitive psychology is deeply connected to the study of metaphysics because it involves understanding the basic concepts of physical objects, causal interactions, and goal-driven intentional agents, which are fundamental to our common sense and philosophical inquiry.

  • What are the limitations of current AI technologies like self-driving cars in terms of cognitive abilities?

    -Current AI technologies like self-driving cars are limited in their cognitive abilities because they lack the flexible, general-purpose kind of common sense intelligence that humans possess. They are good at specific tasks but struggle with the unpredicted cases and the need for robust, real-world problem-solving.

  • What is the significance of the Heider and Simmel movie in cognitive science?

    -The Heider and Simmel movie is significant in cognitive science because it demonstrates the human ability to attribute goals, emotions, and social interactions to simple moving shapes, highlighting our innate capacity for understanding the world in terms of physical objects and goal-driven agents.

  • How does Tenenbaum's work on probabilistic programming languages contribute to the understanding of cognition?

    -Tenenbaum's work on probabilistic programming languages contributes to the understanding of cognition by providing a computational paradigm that combines neural network pattern recognition, symbolic language representation, and probabilistic inference to model the common sense intuitive physics and psychology that underlie human thought and behavior.

  • What is the role of symbolic languages in cognitive science?

    -Symbolic languages play a crucial role in cognitive science as they are essential for abstract knowledge representation and reasoning. They form the basis of natural language, mathematics, and all of computing, including programming languages used for AI and machine learning.

  • How does Tenenbaum's research on the 'common sense core' relate to developmental psychology?

    -Tenenbaum's research on the 'common sense core' relates to developmental psychology by exploring the basic concepts that humans are built to understand from the earliest stages of development, such as intuitive physics and psychology, which are also central topics in the study of human brain development and cognitive growth.

Outlines

00:00

๐ŸŽค Welcoming Professor Josh Tenenbaum

The speaker warmly welcomes Professor Josh Tenenbaum, a computational cognitive scientist at MIT, to speak about the nature of cognition. His achievements and recognition by the MacArthur Foundation are highlighted, and the anticipation for his talk on the computation of cognition is expressed.

05:00

๐Ÿค” The Central Question of Cognition

Josh Tenenbaum introduces the central question of his talk: what kind of computation is cognition? He discusses the importance of understanding cognition in terms of computation and the challenges of finding meaningful answers. The talk aims to raise foundational questions about modeling the mind and the relationship between cognitive science and metaphysics.

10:02

๐Ÿง  The Gap Between AI and True Intelligence

Tenenbaum discusses the current state of artificial intelligence, noting its increasing utility but lack of true general intelligence. He uses the example of self-driving cars to illustrate the challenges and investments in AI, emphasizing the need for fundamental advances in AI to bridge the gap between current technologies and human-like intelligence.

15:02

๐Ÿš— The Progress and Limitations of Autonomous Driving

The speaker delves into the specifics of autonomous driving, highlighting the progress made and the underlying AI technologies like deep learning and neural networks. Despite advancements, he points out the limitations and the 'long tail' of problems that remain unsolved, indicating that AI still has a long way to go to achieve true autonomy.

20:02

๐Ÿง ๐Ÿค– The Computational Aspect of Intelligence

Tenenbaum explores the computational aspects of intelligence, focusing on the 'common sense core' of human cognition. He discusses the intuitive understanding of physics and psychology that humans possess from a young age and the challenge of capturing these aspects computationally.

25:03

๐Ÿ‘ถ Intuitive Physics and Psychology in Infants

The speaker uses examples of infants engaging with objects to illustrate intuitive physics and psychology. He discusses how young children demonstrate an understanding of physical objects and goal-driven agents through their actions and reactions, highlighting the innate human capacity for modeling the world.

30:03

๐ŸŽฅ Heider and Simmel: A Cognitive Science Classic

Tenenbaum presents the famous Heider and Simmel animation as a key example of how humans naturally interpret interactions between agents, even from simple shapes. He discusses the rich understanding of physical dynamics, goals, and emotions that humans exhibit when viewing such animations, emphasizing the computational challenge of replicating this understanding in AI.

35:06

๐Ÿ’ก Probabilistic Programs and the Future of AI

The speaker introduces the concept of probabilistic programming as a way to integrate neural networks, symbolic languages, and probabilistic inference to model common sense. He discusses the potential of this computational paradigm to advance AI and our understanding of cognition, including the role of game engines in simulating physical and social interactions.

40:09

๐Ÿง  Intuitive Physics and Causal Responsibility

Tenenbaum discusses research on intuitive physics and how people judge causal responsibility. He describes experiments that use eye-tracking to understand how people's gaze patterns change based on whether they are making predictions or causal judgments, revealing the computational processes underlying human intuition about physical interactions.

45:10

๐Ÿ› ๏ธ Problem-Solving with Probabilistic Intuitive Physics

The speaker talks about recent work on using probabilistic intuitive physics for problem-solving, particularly in creative tool use. He describes a virtual tools game that demonstrates how people use trial and error to solve problems and how a probabilistic simulation-based model can capture this process, including learning dynamics.

50:10

๐Ÿค– AI Models of Multi-Agent Interactions

Tenenbaum presents a model called 'Flatland' that simulates multi-agent interactions with physical objects, capturing complex behaviors like collaboration and competition. He emphasizes the potential of these models to understand and predict social interactions and causal relationships, highlighting their success in mimicking human-like interactions.

55:11

๐Ÿง  The Brain and Learning in Computational Terms

In the final paragraph, Tenenbaum touches on the relationship between these computational models and the brain, and the potential for understanding how humans learn. He suggests that future learning algorithms may need to be 'program learning programs,' capable of writing their own algorithms, much like how humans construct knowledge.

00:12

๐ŸŽ“ Yale and the Future of Cognitive Science

Tenenbaum concludes his talk by reflecting on his educational experience at Yale and the future of cognitive science. He discusses the potential of algorithms that write algorithms, and the importance of understanding the brain's computational processes. He ends with a call to explore the synthesis of symbolic representations, probabilistic inference, and learning in understanding cognition.

Mindmap

Keywords

๐Ÿ’กComputational Cognitive Science

Computational Cognitive Science is an interdisciplinary field that uses computational models and simulations to study the nature of the human mind and its processes. In the video, Professor Josh Tenenbaum, a computational cognitive scientist, discusses the use of computational methods to understand cognition, emphasizing the importance of creating models that mimic human-like intelligence.

๐Ÿ’กCognition

Cognition refers to the mental processes of understanding, learning, remembering, and reasoning that humans use to interpret and respond to the world around them. In the context of the video, the speaker is exploring the nature of computation that underlies these cognitive processes, aiming to understand how the human mind works in computational terms.

๐Ÿ’กMetaphysics

Metaphysics is a branch of philosophy that deals with the fundamental nature of reality, including the relationship between mind and matter, substance and attribute, potentiality and actuality. In the video, metaphysics is discussed in relation to understanding how we model the mind and the deep issues involved in cognitive science.

๐Ÿ’กEpistemology

Epistemology is the branch of philosophy that investigates the nature, origin, and limits of knowledge and beliefs. It is concerned with the processes of belief formation and the criteria for determining what is true or false. In the video, epistemology is relevant to the discussion of how we know if our models of cognition are accurate or on the right track.

๐Ÿ’กReverse Engineering

Reverse engineering is the process of deconstructing something to understand its workings, typically with the intention of replicating or improving upon its design. In the context of the video, reverse engineering is used to describe the approach of analyzing the human mind to create computational models that mimic its functions.

๐Ÿ’กArtificial Intelligence (AI)

Artificial Intelligence refers to the development of computer systems that can perform tasks that would typically require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. In the video, AI is discussed in relation to its current state and potential, as well as its limitations compared to human cognition.

๐Ÿ’กDeep Learning

Deep Learning is a subset of machine learning that involves the use of artificial neural networks with multiple layers to model complex patterns in data. It is inspired by the structure and function of the human brain. In the video, deep learning is mentioned as a key technology driving progress in AI, particularly in the context of self-driving cars.

๐Ÿ’กIntuitive Physics

Intuitive Physics refers to the innate understanding that humans have of the physical world, including concepts like gravity, motion, and collision. It is a form of common sense that allows us to predict and explain the behavior of objects and events around us. In the video, intuitive physics is used to illustrate the kind of computational models that can capture human-like understanding of the physical world.

๐Ÿ’กIntuitive Psychology

Intuitive Psychology is the innate ability to understand and predict the behavior of others based on an intuitive understanding of mental states like beliefs, desires, and intentions. It is part of what allows humans to navigate social interactions and empathize with others. In the video, intuitive psychology is discussed as a core component of the 'common sense core' that cognitive scientists aim to model computationally.

๐Ÿ’กProbabilistic Programming

Probabilistic Programming is a paradigm in computational modeling that involves writing programs which can execute stochastic processes, or processes that involve randomness. These programs can handle uncertainty in data and make probabilistic inferences about the world. In the video, probabilistic programming is presented as a key tool for building models of cognition that integrate neural network pattern recognition with symbolic languages and probabilistic inference.

Highlights

Professor Josh Tenenbaum discusses the computation of cognition and the reverse-engineering approach to understanding the human mind.

Tenenbaum's research focuses on building models that mimic AI systems to understand natural intelligence.

The importance of comparing models to data, both qualitatively and quantitatively, to ensure accuracy in cognitive modeling.

The gap between current AI technologies and the flexible, general-purpose intelligence of humans.

Self-driving cars exemplify the challenges and limitations of current AI in real-world scenarios.

The potential of probabilistic programming to integrate neural network pattern recognition with symbolic languages and probabilistic inference.

The concept of 'the game engine in your head,' which models the intuitive physics and psychology of everyday interactions.

Intuitive physics in young children and the computational models that attempt to replicate their understanding of the physical world.

The use of simulation programs and probabilistic inference to create computational models of the human mind's theory of mind.

The significance of counterfactual analysis in understanding causality and causal responsibility in cognitive science.

The application of probabilistic intuitive physics in predicting human behavior and creative problem-solving.

The development of a probabilistic generative model that captures multi-agent interactions in a virtual environment.

The future of human learning and machine learning involves algorithms that write algorithms, moving beyond traditional neural network training.

The challenge of creating physical circuits or chips that implement the probabilistic computations of cognitive models.

The coherence of scientific models and the quest for a satisfying understanding of the mind through computational approaches.

The integration of symbolic representations, probabilistic inference, and learning as essential components of understanding cognition.