Overall, \'Artificial Intelligence: A Guide for Thinking Humans\' aims to provide a accessible and insightful exploration of AI for general readers, helping them to understand the technology's capabilities, limitations, and implications for society at large.The author of \'Artificial Intelligence: A Guide for Thinking Humans\' is Melanie Mitchell. She is a professor of computer science at Portland State University and a widely recognized expert in the field of artificial intelligence. Mitchell has made significant contributions to various areas of AI research, including machine learning, genetic algorithms, and cognitive science. In addition to her academic work, she has authored several books and articles aimed at both technical and general audiences, seeking to demystify AI and its implications for society.
Like its ancestor, Samuels checkers player, AlphaGo learns by playing against itself over many games (about five million). During its training, the convolutional neural networks weights are updated after each move based on the difference between the networks output values and the improved values after Monte Carlo tree search is run. Then, when its time for AlphaGo to play, say, a human like Lee Sedol, the trained ConvNet is used at each turn to generate values to help Monte Carlo tree search get started.
id: d2b8c3eb8e24132a05a80229f2dcd9e2 - page: 162
With its AlphaGo project, DeepMind demonstrated that one of AIs longtime grand challenges could be conquered by an inventive combination of reinforcement learning, convolutional neural networks, and Monte Carlo tree search (and adding powerful modern computing hardware to the mix). As a result, AlphaGo has attained a well-deserved place in the AI pantheon. But whats next? Will this potent combination of methods generalize beyond the world of game playing? This is the question I discuss in the next chapter. 10
id: f8558d9541ec8bcc0df40169f6c31654 - page: 162
Beyond Games Over the past decade, reinforcement learning has transformed from a relatively obscure branch of AI to one of the fields most exciting (and heavily funded) approaches. The resurgence of reinforcement learning, especially in the public eye, is largely due to the DeepMind projects I described in the previous chapter. DeepMinds results on Atari games and on Go are indeed remarkable and important, and they deserve their accolades. However, developing superhuman game-playing programs is, for most AI researchers, not an end in and of itself. Lets step back and ask about the implications of these successes for broader progress in AI. Demis Hassabis has something to say about this:
id: 3f43231bd34a6612d8de7ba5ecc6699f - page: 163
Games are just our development platform. Its the fastest way to develop these AI algorithms and test them, but ultimately we want to use them so they apply to real-world problems and have a huge impact on things like healthcare and science. The whole point is that its general AIits learning how to do things [based on] its own experience and its own data.1 Lets dig into this a bit. How general is this AI, really? How applicable to the real world, beyond games? To what extent are these systems actually learning on their own? And what is it, exactly, that they learn?
id: fa7906c550f804bd9e47cefc48c7c3d6 - page: 163