If you want a breezy and accessible argument about the far future of humanity in an age of Artificial Intelligence, this is the book for you. Max Tegmark is a well-connected polymath, with a longstanding interest in AI and a knack for bringing big ideas to life. Who knows what life will be like for humans in the next billion years? Certainly not Tegmark, but why let that stand in the way of the fun? It’s enough here to ask the questions: what will AI want, and why? Can machines ever become conscious, and if so, how might that feel? Is it possible to gain immortality by uploading ourselves to the net? What happens when machines merge with humans as cyborgs, or ‘augmented intelligence’? These are longstanding questions in philosophy of mind, computer science, neuroscience and a bunch of related fields. Since we can’t know the answers to these questions, the fun comes in sketching out the possibilities — often revealing more about ourselves in the process: what does consciousness do for us, for example? How far is our cognition shaped by our bodies, and by our evolutionary imperative to reproduce?
There are plenty of popular books that cover similar territory. Nick Bostrom famously warned of the dangers of an AI ‘Superintelligence’emerging rapidly and with disastrous consequences for humanity, as it single-mindedly seeks to fulfil the task we humans have set it. Another similarly engrossing account of advanced AI is Murray Shanahan’s Technological Singularity– Shanahan was the brains behind the movie Ex_Machina, and, like Tegmark, is adept at asking the big philosophical questions in easy to digest prose.
In his book, To be a Machine, Mark O’Connell has produced a witty and sceptical look at this sort of AI expert, especially transhumanists who explore the technological possibilities of transcending the human body. Transhumanism is the yin to the yang of superintelligence — its sunny utopian promise of eternal human life contrasts with the dystopian vision of a malign superintelligent AI obliterating life as we know it. Tegmark and Bostrom personify the debate — both fascinated and repelled by the possibilities of machines with minds. Ultimately, there’s something deeply human about their existential search for meaning in a universe with machines that can produce and refine themselves without our help.
But is that really where AI is heading? Other introductions to AI offer more insight on the underlying science — the best works here are Margaret Boden’s AI or Kevin Warwick’s AI: The Basics. Warwick, incidentally, walked the walk — turning himself into a pioneering cyborg, complete with DIY implants. Those in the know call this ‘bio-hacking’, and it’s not for the faint of stomach.
Unlike those two books, one thing you won’t get from reading Tegmark is very much detail on the current state of the art in AI — the underlying processes, and the inherent difficulties. AI research has developed in fits and starts, with promising advances followed by frustrating AI ‘winters’ as research stalls a long way short of achieving anything like human-level intelligence. Right now there are breakthroughs aplenty, and it’s easy to go along with the sometimes breathless hype about what AI might achieve. Things are moving so fast that sometimes a superintelligence seems entirely plausible — often replete, in the public imagination, with the sort of glowing red eyes of Arnie’s T-1000. Much of the running is currently being made by ‘connectionist’ AI that are (very) loosely modelled on the brain’s neural networks. But the godfather of this approach, Geoffrey Hinton doesn’t think that connectionism will ultimately bring about the sort of advanced intelligence that Tegmark, Bostrom et al have in mind. Modern AI lacks the ‘common sense’ of humans, as anyone using Siri can attest. Without much sense of what things ‘mean’, connectionist AI might be headed for another winter — highly adept at decision-making in narrow, limited domains, but wholly incapable of applying its reasoning more flexibly. These Artificial Neural Networks can pick out a cat picture on the internet, but they can’t tell you if they like cats. It doesn’t compute. And while they can demolish you at chess and even poker, they can’t on a whim decide to throw it all in and write a novel. In fact, AI’s have written novels — but they’re truly awful.
The other thing that Life 3.0 won’t tell you very much about is war. Breathless and unrealistic hype about Terminator AI aside, AI has profound implications for those interested in strategy — and especially for air minded strategists. As yet, there’s not much written in this important sub-field. The debate has mostly been about the ethics of autonomous weapons. Will they be able to discriminate between soldiers and civilians? Who can we apportion blame to when things go wrong? Is it possible to ban, or at least control, such weapons? This is important — and there are some big challenges to face — how can we regulate AI given that it’s a dual-use technology? Can we afford to keep a human ‘in the loop’ to watch over decisions if the enemy chooses not to? Compared to these ethical conundrums, the broader literature on how AI will shape warfare is rather nascent. That’s changing this year with the publication of Paul Scharre’s Army of None and my own book, Strategy, Evolution and War– but for now the action is mostly in think tank reports, like DCDC’s recent work on human machine teaming. Whether it’s swarm-like mini-UAVs in the attack role, the use of algorithms for encryption and intelligence analysis, or AI deploying ‘fake news’ to deceive adversaries, this is a rapidly changing and important field. In the course of careers now underway, we will see some rapid and radical changes to organisational culture. New skills, concepts and equipment will undoubtedly be required. If Tegmark’s thought provoking romp through some big ideas is your introduction to this brave new world, that’s no bad thing at all.
