Build God, Then We’ll Talk

In God AI, longstanding Artificial Intelligence researcher and University of New South Wales academic Toby Walsh outlines what AI is and what it could become, and explains why people are rightly concerned that AI could lead to the extinction of all human life.

God AI is Walsh’s fifth book about AI, and it shows: his explanations are cogent and clear, allowing him to quickly cover everything from how the human brain is different to a computer to the history of mass extinctions on Earth. For the most part, the reader is taken along on the journey: what AI is capable of, how it could become a “General” intelligence as smart as the average human, or a “Super” intelligence smarter than any human, and how that could end life on Earth.

Walsh identifies advantages to smarter-than-human AI (curing disease; liberating people from the need to work) as well as downsides on smaller and larger scales (surveillance states, mass automated killings in war and the aforementioned extinction of our species).

Walsh gives several reasons why “General” and “Super” intelligence might be out of reach. But make no mistake, Walsh expects smarter-than-human AI is coming, probably within this lifetime.

God AI boasts many vivid analogies. Why do we imagine that humans could create smarter-than-human intelligence? Bees are intelligent, in their way, but we don’t imagine they could create smarter-than-bee intelligence.

Also pithy are Walsh’s descriptions of the human brain: three pounds of wrinkled fat that requires about as much energy as a dim lightbulb. If meat can think, feel, reason and create, why can’t “sand” (silicon computer chips)?

God AI shares many of the grand concerns of progressives without indulging the more petty ones. The book takes seriously workers’ rights, human dignity, peace and equality. Climate change is recognised as the cause of most past mass extinctions, and a possible cause of the next one.

Walsh accuses AI boosters of over-rating intelligence and downplaying the role of power, a critique that will be familiar to readers of The Point. Humans have already created super-intelligent entities more powerful than any individual – they are called corporations. No human can make an iPhone from scratch, but Apple can.

By taking these grand concerns seriously, Walsh avoids the trap so much progressive discourse about AI falls into: of treating the creators of AI like any other corporate villain. He gives issues like data centre electricity and water use, the wasteful AI investment bubble and copyright infringement their due, without letting them distract from the larger, unique and existential threats that AI poses.

The weakest chapter of the book is unfortunately the longest: on “X-risk”, that is, existential threats. Specifically, human extinction, rather than mass death and suffering. Walsh focuses on threats that could conceivably wipe out all humankind, like theoretical cobalt bombs, “mirror life” or a genetic bottleneck hundreds of years in the making. This neglects war, genocide, nuclear armageddon, climate change, famine and disease that could kill millions or billions of people but not the species itself.

Walsh concludes, “The lessons we can learn from past extinctions and other existential threats are, I fear, disappointingly modest”. I agree. Time spent on esoteric and remote threats was a missed opportunity to anchor AI to more familiar and pressing dangers.

Walsh does correct this omission in later chapters, proving that AI is already causing harm. AI doesn’t have to be as smart as a human to drop a bomb on a school, take someone’s job or report a democratic activist’s movements to the secret police of an authoritarian state.

The book’s wide-ranging approach leads the reader down some dead ends. Walsh links fear of AI extinction to eight supposedly overlapping ideologies: Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism and Longtermism (TESCREAL for short). Most of these are dispensed with in a single sentence. We are to understand from the summary of TESCREALs as “white male geeks” that the thinking of these people can be dismissed, but it’s not clear why their fears of AI extinction are wrong if Walsh’s are credible.

As befits the title God AI, Walsh rightfully spends time on the quasi-religious thinking of many Artificial Intelligence proponents. But he does not suggest they are entirely wrong either, recognising that a superintelligent AI would approach god-like powers of omnipresence, omniscience, incomprehensibility, immortality and omnipotence.

God AI is a clear, broad and easy-reading primer to the most extreme danger of Artificial Intelligence: that if machines achieve super-intelligence, they could readily kill all human life and destroy our species and everything we have worked for. Walsh recognises the arguments against this ever happening while refusing to be complacent about the catastrophic consequences if it does eventuate.

The book makes worthwhile reading for anyone worried about AI risk – and for anyone who is not worried enough.