Courseware / Machine Learning / course-016
An In-Depth Look at the Journey of Demis Hassabis and the Development of Artificial General Intelligence
Tweet@sairahul1View Source →

🎙 Podcast Version

2-host dialogue — ALEX & SAM discuss this course.

An In-Depth Look at the Journey of Demis Hassabis and the Development of Artificial General Intelligence

Overview

This course provides a comprehensive exploration of the journey of Demis Hassabis, co-founder of Google DeepMind, in his pursuit of building artificial general intelligence (AGI). We will delve into his early experiences with AI, his time at Cambridge, and the inception and mission of DeepMind. Additionally, we will discuss the two primary approaches to AI, the shift from expert systems to learning systems, and the applications of these systems in scientific discovery.

Background & Context

Demis Hassabis is a renowned computer scientist, neuroscientist, and gaming expert who co-founded Google DeepMind, a leading company in the development of AGI. His fascination with chess and AI at a young age led him to study computer science at Cambridge, where he was inspired by the intellectual giants of the past and the theoretical underpinnings of computer science and computation theory. In 2010, Hassabis established DeepMind with the mission of solving intelligence and using it to solve everything else, focusing on accelerating scientific discovery and medicine.

Core Concepts

Artificial General Intelligence (AGI)

AGI refers to AI systems that can perform any intellectual tasks that humans are capable of, encompassing all cognitive capabilities.

Expert Systems

Expert systems are AI systems that are directly programmed with solutions, often inspired by logic systems. They struggle with unexpected situations and can be rigid and brittle.

Learning Systems

Learning systems are AI systems that learn for themselves and learn directly from experience or data, inspired by neuroscience ideas. They have the potential to go beyond the knowledge of the programmers and are valuable in areas like scientific discovery.

Hassabis' Journey and DeepMind

Hassabis' early fascination with chess and AI led him to study computer science at Cambridge, where he was inspired by Alan Turing, Charles Babbage, and other intellectual giants. In 2010, he founded DeepMind with the mission of solving intelligence and using it to solve everything else. The company started with a two-step process: step one, solve intelligence; step two, use it to solve everything else.

The Shift from Expert Systems to Learning Systems

DeepMind's approach to AI focuses on learning systems, which have the potential to go beyond the knowledge of the programmers and are valuable in areas like scientific discovery. In contrast, expert systems like Deep Blue, which beat Kasparov at chess, struggle with unexpected situations and can be rigid and brittle.

Applications of Learning Systems in Scientific Discovery

Learning systems can help accelerate scientific discovery and medicine by going beyond the knowledge of the programmers. For instance, AlphaGo, a system developed by DeepMind, played the game of Go at world champion level and even invented new strategies, demonstrating the potential for these systems in scientific discovery.

How It Works / Step-by-Step

DeepMind's AI systems, like AlphaGo, are trained up through a system of self-play. In the case of AlphaGo, it starts with a version that doesn't really know anything about the game, just the rules, and it plays randomly. After playing 100,000 games against itself, a new database of game positions is created, and a second version is trained based on that data. This process continues until the system becomes proficient at the game or task at hand.

Real-World Examples & Use Cases

  • AlphaGo, a system developed by DeepMind, famously won a million-dollar challenge match against 10-time world champion Lee Sedol in 2016, demonstrating the potential for AI in complex games and problem-solving.
  • DeepMind's AI systems can potentially be applied to various scientific disciplines, such as medicine and biology, to accelerate discoveries and solve complex problems.

Key Insights & Takeaways

  • AGI encompasses all cognitive capabilities that humans possess.
  • Expert systems, like Deep Blue, have limitations in dealing with unexpected situations and can be rigid.
  • Learning systems, inspired by neuroscience ideas, can go beyond the knowledge of the programmers, making them valuable in areas like scientific discovery.
  • DeepMind's AI systems, like AlphaGo, are trained through self-play, starting from a random state and gradually improving through repeated games.

Common Pitfalls / What to Watch Out For

  • When working with AI systems, it is essential to consider their limitations and potential pitfalls, such as their inability to handle unexpected situations in the case of expert systems.
  • Overreliance on AI systems may lead to a lack of understanding of the underlying principles and concepts.

Review Questions

  1. What is the difference between expert systems and learning systems, and what are their respective advantages and disadvantages?
  2. How does DeepMind's two-step process for building AGI differ from traditional AI approaches?
  3. Describe how AlphaGo was trained and how it demonstrated the potential for AI in scientific discovery.

Further Learning

  • Explore the development of other AI systems and techniques, such as reinforcement learning and deep learning.
  • Investigate the applications of AI in various industries, including healthcare, finance, and manufacturing.
  • Delve deeper into the theoretical underpinnings of computer science and computation theory, as well as the latest advancements in AI research.
← Previous
Next →