Will AI fry the planet, implode or find a sensible way forward?

© Overearth, istockphoto.com
One of my claims to fame, in my living room at least, is to have coded two neural networks in my time. One static network as a student engineer on placement in a Government research department in 1990, and a recurrent neural network three years later for my final degree project. These were laughably simple with one intermediate level of nodes between input and output compared to 7 or 8 in today’s versions – but it still took an hour or so to see if they were actually learning anything from the training data. The main difference is the sheer processing grunt available today – and processing power requires electrical power…
Today of course, we can’t get away from AI. The US mid-terms results, the PHP hack, Palantir, Jensen Huang’s love life, protests against data centres in Edinburgh and Donald Trump’s abiding love affair with AI slop, the headlines are just one AI thing after another. But the big question is: are we at the peak of the Gartner hype cycle with the trough of disillusionment gaping ahead of us?
There are several reasons to believe so:
- While who Mr Huang has been sleeping with (and why they’ve been sleeping with him) appeals to our prurient appetite for scandal, the fact that he and his company Nvidia has been investing in AI to keep demand for Nvidia chips booming could be a recipe for disaster. With Nvidia facing post-Jensen turmoil, what will happen next? How long can the developers sustain ongoing losses?
- Data centre developers have to find grid connections and power for each data centre. To put this in context, a project to retrofit social housing with heat pumps near me has a three year wait to secure a modest grid connection. Yes, there is talk of behind the meter gas power, but gas turbines also have a waiting list and you still need a gas grid connection.
- Public acceptability, or lack of, was one factor I didn’t have on my bingo card. In the US, the Democrats campaigned hard and successfully for restrictions on data centres – I know from experience that politicians love a bandwagon to jump on, so this could spread fast. Here in the UK, the Government has designated data centres to be Critical National Infrastructure (CNI), but would that hold in a political onslaught if, say, the Green Party decided to go in hard?
As a typical centrist dad, I’m taking a measured approach to all this – I suspect the industry does face a reckoning, lots of investors will lose money, and then we’ll revert to something more sensible. It has always bothered me that data centres are sized by energy consumption rather than computing power – a rather macho perspective. If suddenly there is pressure to make AI pay, will this flip to a focus on output and a drive for energy efficiency? We shall see!