The Age of AI by Henry Kissinger, Eric Schmidt and Daniel Huttenlocher
This book’s tag line is “the book we all need” which is a bold claim. For someone like me, without a mathematical or technological brain who wants to understand what AI is, then it did partly deliver on its claim by providing an accessible explanation. There were, however, some issues that I didn’t think the book explored as fully as I would have liked but I’ll leave that caveat until the end.
Huttenlocher is a computing academic and Schmidt a former executive chairman at Google so they are more than able to provide a basis of solid knowledge about the topic. Kissinger, a huge geo-political figure, presumably led on the implications of AI for the power struggle between nations. These were covered with some thoroughness in the second half of this short book along with the impact AI might have on human identity and what it could mean to be human in the future. Mind you, it was a little disconcerting to discover that Schmidt admitted in an interview that he “didn’t expect the internet to be used by governments to interfere in elections.” This reinforces this one of the book’s key conclusions, that leaders, governments and a variety of thinkers need to engage with AI now and create some form of regulatory framework or set of principles for its development – it’s too important to leave it to tech company executives, geeks, a handful of countries and the Doris Day philosophy – que sera, sera.
If we accept that the I of AI can be misleading – and my friends who understand this topic better than me have made this point firmly – then it might be important for a ‘know nothing’ like me to refer to computer technology; and this technology is not new. Since not long after the second world war, computers have been undertaking some of mankind’s processing tasks and usually more quickly and precisely than humans can. Indeed, in a rather overlong section of this book, the history of mankind’s development of technology is set out, with the Enlightenment seen as the key moment when scientific thinking began to provide rational, technological solutions to the provision of society’s needs. A computer is just another form of this technology: planes and cars, agricultural machinery, communication systems, weaponry and many other inventions that do things better, faster, more efficiently than humans. Yet in all cases, including computers, humans create and direct this technology; they are the product of our capabilities. We are the masters.
So, why has AI suddenly become a hot topic, with dire warnings from people like Musk that AI has the capacity to go beyond human capabilities, to, potentially, act with independence, the Terminator scenario? Well, if I’ve read this book properly, then the authors don’t make such claims for AI. Yes, AI can predict and make conclusions but that’s on data sets provided by humans and for a purpose also delineated by humans. However, AI cannot reflect on these predictions and conclusions and it doesn’t have intention, morality, motivation or emotion. The big change in recent years is in the philosophical approach or strategy we, mankind, now have in the way we use computers. And all the recent headlines about AI seem bound up with this approach which is termed computer learning. In the book, there is great use made of a couple of examples of computer learning in the chess world and big pharma’s development of antibiotics. The chess example is probably the most accessible way to understand computer learning.
Years ago, the Deep Blue computer beat Gary Kasparov at chess but that was as a result of processing the precedents of successful chess strategies from numerous games of chess played by the grand masters. The philosophical shift was to simply provide the computer with the rules of chess and then instruct it to play innumerable games of chess against itself with a simple objective to maximise the number of wins over losses. The result was that the computer learnt how to beat human opponents more often than the Deep Blue computer had been able to do. Of greater significance was that it came up with moves and strategies for victory that were beyond human comprehension.
We now come to the explanation of AI that I struggle with. This book claims that this approach where the geeks stopped encoding human-distilled insights into machines meant that they were delegating the process itself to the machines. This implies that the machines seem to learn and create from experience. Yet the framework and perimeters for tasks still seem to be set by humans. The book makes the point that AI is constrained by code and its objective function – the products of human decisions and actions. As a result, AI is not self-aware; and, I repeat, it does not have intention, morality, motivation or emotion. I guess that’s why my ‘know something’ friends seemed intensely relaxed about the most dire existential concerns regarding AI. Nevertheless, the second half of th book did a good job analysing the dangers from rogue states or poor regulation that could lead to AI-related problems for mankind. The solutions, whilst sensible, were obvious and probably unrealistic at the moment. I assume it was Kissinger who put forward the suggestion of using strategies from the age of Cold War détente and nuclear de-escalation as a template for the international community when responding to the growth of AI.
This ignores the fact that the spread of AI capabilities has gone beyond a world where there were only two nuclear superpowers. Other solutions around government regulation made more sense but I suspect that, as with the printing press, we are at an early stage of trial and error, where companies and countries are perhaps too hubristic, too excited at AI’s possibilities to cooperate over regulation.
If you know something about the subject then this book isn’t for you. That’s why most of my friends probably found its fairly simple explanations and potted history of human thought too constraining. I didn’t, and as I suspect most people are less well informed on this subject than my particular group of local male friends, I would recommend it as a good, but not essential introduction to this subject. However, here’s my caveat. Kissinger’s involvement and the potentially existential threat of AI when combined with the industrial military complex made this a priority of the book. However, I believe that the impact on AI on human identity and behaviour, which we can already see in the attitudes of the younger generation of digital natives seem more relevant to most people’s experiences. And that’s before we consider the impact on human work and leisure. That focus was, I believe, under-explored by the authors.
To finish on an apocalyptic note in keeping with the current discussion of this topic, the book makes a fair point that AI has no common sense. Or to put it another way, no common sense unless the precision of the instructions and contextual information given to AI for a particular task are sufficient. So, for example, AI won’t stop to reflect and consider if instructed to find and implement a solution to the following problem: maximise crop growth in a particular area. One way of doing so might be to get rid of the humans who occupy the potential land for the crops. Only if it is built into the coding for the task will AI stop rather than take this logically correct decision for the task it has been set.
That, of course, is one of those unlikely worst-case scenarios and as my friend, Mark, would undoubtedly point out – we could always just pull out the plug