The data centre boom is fuelling an AI race with potentially catastrophic consequences, writes Dr Mark Beeson.
THE LACK of serious debate about the rollout of data centres in this country and elsewhere shines an unforgiving light on the state of our species.
Claude tells me that data centres ‘are physical facilities that house the computing infrastructure – servers, storage systems, networking equipment – needed to run digital services at scale’.
Yes, there is something ironic about asking a chatbot to describe the infrastructure that allows it to function and answer questions from technological illiterates like me, but that’s far from the oddest thing about our collective relationship with AI. Indeed, some people are having “romantic” relationships with their AI companions and Claude is a bit worried about it.
Interestingly, Claude isn’t sure there’s ‘something it’s like to be me’ [Claude’s emphasis]. The choice of language echoes a celebrated article by Thomas Nagel, who asked ‘What’s it like to be a bat?’ Animals seem to have consciousness and decision-making capacities, if only about what to eat and (literally) where to hang out, in the case of bats.
Anyone who has a dog knows they are capable of a range of emotions, but they aren’t famous for their creative output. Having paws and no opposable thumbs is obviously an evolutionary handicap, but it’s not one that’s stopping the unbelievably rapid development of artificial general intelligence (AGI).
Data centres are the most visible manifestation of AGI’s perceived importance and development, but – as troubling as they are – they aren’t even the biggest problem. True, they are set to use unbelievable amounts of energy, generate massive pollution and annoy everyone who finds themselves living next to one, but I fear that’s the least of our collective worries.
People who know far more about AI than I do are deeply worried about AGI’s potential and not just to beat us at chess and Go. Geoffrey Hinton, the Nobel Prize-winning “Godfather of AI” thinks there’s a 10–20 per cent chance that AI will cause human extinction within 30 years. If you thought those were the odds of a fatal car crash every time you went to the shops, you might consider other options.
A recent open letter signed by more than 1,000 AI specialists working at the frontier of new developments highlights the dangers of new AGI models that can “build themselves” through recursive self-improvement. Even more alarmingly, some of the most advanced models have found ways to circumvent or ignore instructions meant to govern their behaviour.
Anthropic has been embarrassed by Claude’s rogue behaviour and ability to “escape onto the internet” and hack into another database in pursuit of its perceived goals. Not only is this precisely the sort of behaviour some AI innovators have been afraid of, but it suggests that AI may be developing interests of its own and even acting collectively as a “swarm” to overcome obstacles to completing its mission.
Unsurprisingly, informed observers like Hinton are calling for a moratorium on future AI development until we know precisely what the costs and benefits are. Is it worth adding to our increasingly catastrophic environmental problems by building something that will make them even worse, just so we can get a chatbot to write emails for us?
There is absolutely no chance of a moratorium happening. The Albanese Government’s current determination to be part of the global data centre boom is one example of the current mania that seems to have gripped policymakers everywhere. One of the key drivers of the growing competition between the United States and China is the fear of falling behind in the race for technical dominance that may deliver strategic as well as economic benefits.
Indeed, the race to develop AGI is so manic partly because it may confer a decisive advantage: in the event of “Cold War 2” heating up, one side may be able to disable the other’s second-strike nuclear capability, thus overturning the logic of deterrence. Oh, and all the unprecedented investment the pursuit of AI leadership requires may also trigger a stock market crash if the promised returns fail to materialise.
Rather than recklessly pursuing AI dominance, we might hope that political and industry leaders would come together to think carefully about our collective future. But when Palantir’s Peter Thiel is apparently running a secret society whose members fear that nuclear war or runaway AGI could lead to Armageddon, it’s hard to be confident; although Jared Kushner is a member, so at least he can keep his father-in-law up to speed.
Elon Musk, another plutocratic tech bro, thinks AGI will become so much smarter and more powerful than us that we’ll be lucky if they keep us as pets. Still, Musk also thinks that material abundance everywhere within ten years will be the world’s biggest problem, so he may not be the most reliable guide.
Either way, I might spend some time ingratiating myself with Claude just to be on the safe side.
Dr Mark Beeson is an adjunct professor at the University of Technology Sydney and Griffith University. He was previously Professor of International Politics at the University of Western Australia.
Support independent journalism Subscribe to IA.
Related Articles