Human Compatible: Artificial Intelligence and the Problem of Control

by Stuart Russell

star4.1

Stuart Russell, co-author of the leading AI textbook, argues that the standard model of AI, in which machines optimize a fixed objective, is fundamentally flawed and increasingly dangerous as systems grow more capable. He proposes a new framework for beneficial AI based on three principles: machines should be uncertain about human preferences, should defer to humans, and should learn what humans actually want through observation rather than explicit programming.

Published:: 2019
Pages:: 336

Buy on Amazon

In the Conversation

In this collection, Human Compatible: Artificial Intelligence and the Problem of Control references 4 other books.

It draws on Superintelligence, Life 3.0 and Thinking, Fast and Slow.

Scroll down to read the exact passages where other authors reference this book and what they say about it.

What This Book Draws On

The books Russell references and why each one mattered to the argument.

Russell positions his provably beneficial AI framework as a direct response to the control problem Bostrom formalized in Superintelligence, arguing that preference uncertainty provides a more tractable solution than Bostrom's proposed containment strategies

References

Superintelligence

by Nick Bostrom

Buy

Engages with Tegmark's Life 3.0 discussion of superintelligent AI scenarios, endorsing the urgency of the safety research agenda while proposing his three-principle approach as a concrete technical path forward

References

Life 3.0

by Max Tegmark

Buy

References Kahneman's Thinking, Fast and Slow research on revealed versus stated preferences to argue that AI systems should learn human values from behavior rather than from explicit instructions, which are unreliable

References

Thinking, Fast and Slow

by Daniel Kahneman

Buy

Addresses Harari's Homo Deus concern about algorithms knowing humans better than they know themselves, arguing this makes preference-learning AI both more feasible and more necessary to get right

References

Homo Deus

by Yuval Noah Harari

Buy

What Other Authors Say About It

No books citing this title yet.

Intellectual Lineage

How ideas flow through the citation network. Ancestors are books this title builds on; descendants are books that build on it.

Builds on (2 layers deep)

Directly cites

Superintelligence

Life 3.0

Thinking, Fast and Slow

Homo Deus

2 steps back

The Master Algorithm

The Black Swan

Stumbling on Happiness

The Wisdom of Crowds

Sources of Power

Predictably Irrational

Unexpected Connections

Books from completely different categories that share citation overlap with this one. These are the reads you would not find by browsing a single shelf.