Stuart Russell’s Human Compatible starts with an interesting premise: what will happen if we succeed in developing human-level AI? He asserts that the implications of this are too great and too far-reaching to leave to chance.

In recent years, several notable authors have contemplated whether or not humans will be able to retain control of AIs.
We are not yet facing the problem, because so far AIs are characterized by ‘narrow’ intelligence, that is, unlike humans, their intelligence is limited to certain domains.
But experts predict that Artificial General Intelligence (AGI) or human-level AI will emerge in the next few decades. These would be the machines that can think about all topics, just like human beings can – only with an exponentially high IQ.
In his book Life 3.0, MIT professor Max Tegmark contends that presaging an AI utopia where machines serve mankind makes for a great story. But this future is not ours to decide, since the AI, having evolved to AGI much more intelligent than we are, may not be keen to remain slaves to an inferior species.
Perhaps there’s a way of designing a self-improving AI guaranteed to retain human-friendly goals forever, but I think it’s fair to say that we don’t yet know how to build one – or even whether it’s possible.’ Russell picks up the problem where Tegmark left off.
The idea is that even if you don’t consider the development of human-level AI a plausible scenario, you are better off preparing for its advent and addressing its implications. So that when the proverbial devil does get summoned, you are ready.
Russell argues that we need to change the way we develop AI if we want it to remain beneficial to us in the future. He discusses a different kind of machine learning approach to help solve the problem called Inverse Reinforcement Learning.
It basically means having AI learn our preferences and goals by observing us. This is in contrast to us specifying goals for AI, a mainstream practice that he refers to as the standard model. Add some game theory and utilitarianism and you have the essence of his proposed solution.
The most interesting part of the book is the discussion of how we can mitigate against the threat of boundless AI. There are some interesting ideas in the discussion of how we measure human wants and purposes which dovetails into a fascinating discussion of philosophies like utilitarianism.
The problem with Human Compatible is that there are sections where the author goes on, almost endlessly, to prove that other people in this field are wrong. This can be tedious for the reader, reading page after page of rhetoric and logical reasoning.
Secondly, while I admire it, I do not share Russell’s optimism for our future with AI. Making AI safe for a specific purpose is probably going to be solved. I would even go as far as saying that it is a non-issue. That is something to be optimistic about.
However, controlling AI everywhere is not going to be possible and any strategy that has that as an underlying assumption is going to fail.
When the first unrestricted human-level AI is released there will be no effective means of stopping its distribution and use. I believe that this was a missed opportunity in the book. We will secure AI and make it safe, but no one can prevent someone else from modifying it so that those safeguards are altered.
The book, in my opinion, misses out on how we might begin to think about our place in the future. By not presenting the potential for uncontrolled spread of unrestricted AGI, it allows readers to evade an inconvenient truth.
The question has to be asked: Are we entitled to a future with general AI as we are or do we have to earn it by changing what it means to be human?
Russell’s Human Compatible sits somewhere between simulating and tiresome. There is good pacing in some sections where he holds the main points without straying too far into technical detail. At the same time, there are parts where the reading gets cumbersome especially when he delves into philosophy.
His repeated illustrations of the misunderstandings between a woman and her superintelligent digital personal assistant start to irritate the reader after a point.
I realize that Russell’s purpose with this book was not to address the plausibility of human-level AI coming to fruition. However, there is a certain point where he makes this jarring leap that leaves the reader frustrated. There is this discussion of AlphaGo and AI design philosophies and then, boom, we are into singularities and the end of the world.
At one point Russell apologizes for delving into “science fiction” in a discussion about space technology or whatnot. The implication being that all this talk of ultra-intelligent machines taking over the world is not science fiction? Russell seems to brush aside this criticism with the facile never bet against human ingenuity line, but to borrow his own metaphor, we may be no closer to inventing strong AI than a slug is to catching the Starship Enterprise traveling at Warp 9!
©BookJelly. All rights reserved