A robot didn't write this review

THE AGE OF AI And Our Human Future

Author: Henry A. Kissinger, Eric Schmidt and Daniel Huttenlocher

Publisher: Hachette

Pages: 256

Price: Rs 799

It is extremely unusual for a book to shock and give hope both at once. To provoke, yet calm the mind, to alarm yet relieve, to show why artificial intelligence (AI) will transform, remake the world and us in ways yet unimagined. Deeply insightful yet disturbing, the book asks what in the future will define and set apart a human from AI and why humans should be worried and elated at the future of AI. It is exceptionally well written. What sets apart the book from others in the same genre is the fact that three stalwarts from different walks of life — Henry Kissinger, Eric Schmidt and Daniel Huttenlocher —have collaborated on it, producing an unusually rich tapestry of thought leadership that covers statecraft, diplomacy, entrepreneurship, technology, and academic brilliance.

Let us consider some of the extremely disturbing questions that the book touches on. Suppose an AI engine driving a car predicts that killing the occupants would save many more in the event of an upcoming crash and the decision must be made in seconds. What should it do? Should it protect the occupants of the car or kill them to save many more in other vehicles nearby? Suppose an AI-controlled weapons system predicts that killing a thousand people in a war-like situation will lead to saving a million more lives. How should the system respond? It is disturbing to know that AI can make these choices but it is impossible to figure out its preference ordering even by those who program it, for what goes into the black box of the decision-making is something completely unknown to humans yet.

The book has some remarkable examples. The first one is about AlphaZero, a chess game purely trained on machine learning that beat every known chess player and computer game in the world. What made it different from any of its predecessors was that it was not taught by humans. Instead it was simply fed the rules of chess, told to maximise the wins, and it did the rest. GarryKasparov after seeing it commented that chess had been shaken to its roots. As AI breached the limits of the game, players who had spent their lives mastering it watched and learned as they saw it sacrificing unusual pieces, including the Queen, to win.

Likewise, yet another AI beast is the MIT-designed Halicin, which is an antibiotic that could kill bacteria that is resistant to all known strains of antibiotics till then. MIT said it would be “prohibitively expensive” to do it via traditional methods of drug discovery. Instead, AI, while it did not need to understand why the molecules worked, could scan the library of candidates to identify one that would perform a desired albeit still undiscovered function: To kill a strain of bacteria for which there was no known antibiotic. Halicin was like a chess game with thousands of pieces, hundreds of victory conditions, and rules that are only partially known. After studying just a few thousand cases, AI found a new antibiotic that no human had perceived.

What marks a turning point in human history is the fact that AI obtained results that were beyond the capacity of human minds to compute. In other cases, it obtained results by methods that humans could, retrospectively, study and understand. In others, humans remain uncertain to this day how the programs achieved their goals.

The book has a remarkable chapter on current technology platforms and how they influence our lives today and will do so in the future. Where it could have done better is to explain the human-machine collaboration and partnership. This theme recurs throughout the book but there is no cogent analysis of how these interactions can be either designed or executed in the years ahead except for some high-level thoughts on how user data is being used to create better machine learning. Questions such as how machine learning operates in individual cases, how it makes decisions using the data that it collects and the risk of the data being used to exploit those very people from whom it is collected (and how to mitigate it) are not discussed in any detail.

Today there are many AI programs available that by using simple tools can write a book review (or even an academic thesis) by just skimming some parts of the book. Does it mean that columnists and book reviewers might be out of a job soon? I’m quite sure that in the years to come you may be reading this review written entirely by a robot. Whether it is as good as a human review, or even better time will tell, but in the age of soaring newsprint prices, declining forests (and need to conserve paper) and declining subscriptions and ad revenues thanks to digital, I’m sure editors would love to be free to use an AI tool that can review any book in a few minutes!

The reviewer is an IAS officer, not a robot. These views are personal, 

not AI or machine-generated