Burn, baby, burn - ChatGPT x CO2

This blog first appeared on Monday 24 April to mark the publication of episode 72 of the Small Data Forum podcast. Have a listen here.

Fire and music go well together. Sixties rocker?Arthur Brown?– a long time-resident of the liberal enclave of my home town, Lewes – made an entire career out of his 1968 cult classic,?Fire. Indeed, I even played roadie to him and had the honour of putting him out when?he caught fire?during the first chorus of Fire in a Sussex gig back in 2007, my pimple on the backside of rock ‘n’ roll history. And the first time Arthur had gone up in flames since the 1971 Windsor Jazz festival.

The least successful rockstar of all time,?John Otway, was given a 50th?birthday present to remember when his fans “rigged” the charts in a totally legal way and bought him a second, top-ten hit in a 5,000-plus gig career, and that catchy ditty?Bunsen Burner?stormed the charts. It’s chorus features the line “Burn, baby, burn”, a lyrical echo through the ages, from?The Tramps?to (appropriately enough)?Ash.

Generative AI x climate change

And “burn baby burn” is exactly what it appears the planet will be doing – even quicker than the entire combined scientific consensus has unequivocally determined it will do, thanks to our crack-like addiction to fossil fuels – if we don’t shake our very recent, very deep love of generative AI. Sam starts episode 72 of the Small Data Forum podcast with a look at the latest developments in this new technology, whose poster boy is ChatGPT and one of whose early funders was Elon Musk. But more of the Musky one, anon.

In recent weeks, Sam points out, there have been a growing number of respected sources calling bullshit on the carbon intensivity of generative AI. It makes sense, right? A set of algorithms that, in an instant, rummage through the world’s knowledge (or at least the world’s knowledge up to about September 2021) is going to drain a lot of energy. All those gigaflops burn, baby, burn through rainforest canopy after rainforest canopy, as heavyweights from the?MIT Technological Review?to?Computer Weekly?are demonstrating as they weigh into the debate.

The choice is pretty stark. If we use generative AI in its current form, it can – stress can, if the prompts are good and don’t need hours of refining – help us with structure, content, and thinking, based on humanity’s knowledge to date (read September 2021). But that comes at a very real cost – and that cost is hugely elevated carbon emissions. It’s up to individual users, teams, and companies to determine whether that’s a cost worth paying. (Hint: it may not be, certainly from our grandchildren’s perspective).

From Gutenberg to Zuckerberg?

In our previous, low-carbon emission, Sam poured some cold water on the “yet another Gutenberg” moment hypothesis of AI, and though we Podnosticators Three all see the potential of mass-availability AI tools, it’s fair to say we represent a spectrum of opinion. Neville is the most enthusiastic, Thomas the most pro-regulation, and Sam somewhere in the middle getting splinters in his ass from sitting on the fence. Thomas has experienced at first hand the confabulating – nay, bullshitting – nature of ChatGPT and its ability to create fake citations; what its inventors and investors gently refer to as “hallucination”. Since we last met, Thomas has discovered an excellent browser plug-in for Chrome –perplexity.ai?– that sources and returns only vetted links.

But the bullshitting tags won’t go away, and more evidence is emerging about why the current raft of algorithms make things up, and it’s all to do with both convergent (rather than divergent) thinking AND?unsupervised learning. In a classic case of?quis custodiet Ipsos custodes?–?“who guards the guards” – the algorithms’ learning experience is itself unsupervised by anyone (or rather, anything) other than … the algorithms themselves.

Thomas is concerned particularly about the training data on which generative AI text tools are learning, and cites a recent?Guardian?article?which shows that, while Wikipedia and the open access scientific journal hub PLOS (the Public Library Of Science) are reassuringly in there, so are some rather more unsavoury sources. Sources including white nationalist site VDARE, far-right news source Breitbart, and the Russian-back propaganda site RT. (No backlinks required, thanks).

Is ChatGPT the new Bitcoin?

Neville draws analogies to the humungous?carbon footprint and impact of cryptocurrency, and can’t resist the irony of shiny new electric vehicle charging facilities being powered by diesel generators (fact-checked?here?by?Reuters). But there’s apparently unquenchable computer user appetite for generative AI – “The people want it!” urges Neville, observing that every time he’s on LinkedIn (up to three times a day) there’s yet another bro (or soeur) giving top tips for how to channel ChatGPT to power your business forward. Or it might just be FOMO, and FOMO that accelerates the planet burning, baby, burning.

Neville is impressed with the array of bodies – from the UK and EU Governments to the US and UN – who are proposing regulation of AI, and as the carbon impact is going up the agenda, they will have to address this issue, too. For Thomas, it all – as so often – comes back to the keen-eyed analysis of the University College, London economist,?Mariana Mazzucato. As she has pointed out for many years – at least since her 2011 book,?The Entrepreneurial State: Debunking Public vs Private Sector Myths?– there would be no Silicon Valley without state investment, from NASA to DARPA. Yet Silicon Valley is neoliberal capitalism on steroids, prioritising profit over all else and abdicating responsibility for external costs. External costs like carbon emissions causing climate change.

“When will we wake up?” he wonders. “When will companies go beyond the nonsense of CSR and actually behave responsibly?”, fixing the problems it causes and doing rather more than buying or planting yet another immature forest. “A bollocks idea!” he believes.

There’s a musky smell around here

Our second major focus in this episode is Elon Musk, much in the news following his recent?“warts-and-all” interview (given very much on his terms) with the BBC, and the launch (and almost immediate ‘rapid unscheduled disassembly’) of Elon’s massive SpaceX rocket. For Neville – and he uses the phrase as a term of praise – Musk is a modern-day?robber baron, in the mould of Vanderbilt or Rockefeller. Unethical, eliminating the competition, and accumulating massive wealth – and at the same time pioneering and taking the world and commerce in novel and diverse directions. Neville cites his own research assistant for this section as Bard (not ChatGPT) and – perhaps swimming against the mainstream – found Google’s new AI tool much better than Musk-funded OpenAI’s.

The BBC interview left Neville a bit “Meh!” (he left after about half an hour), not quite as “Meh!” as Thomas (who lasted ten minutes), but rather more “Meh!” than Sam (who watched the whole thing). As Neville observes, Musk gave the BBC correspondent a hammering – a hammering repeated on Twitter from Musk fans, rather in the style of Trump fans. For Neville, the interview was a well-architected PR coup. But his admiration for Musk is more about his achievements in power, energy, and battery storage than his management style, not to mention smart investments in both OpenAI and?Neuralink. “He’s a robber baron who makes things happen,” concludes Neville.

RIP the blue tick

Thomas cites our (other) favourite podcast,?The News Agents,?staffed by three, whip-smart, now unchained ex-BBC journalists – Maitlis, Sopel, and Goodall. They critiqued the BBC x Musk interview as a turning of the tables. The BBC reporter was given no time to prepare and the meeting took place entirely on Musk’s terms. As Neville reports, Musk streamed the interview live and unexpurgated on Twitter Spaces, with an audience peaking at 717,000.

The day before we recorded this episode, Musk’s reign at Twitter had seen the removal of blue ticks?of accreditation from all not prepared to pay the equivalent of US$8 per month, including Neville. That happened on April 20, or?4-20, the traditional dope smoker’s day of celebration of all things marijuana. Sam wonders whether some of the ever-shrinking leadership of Twitter had been overindulging – like much of the rest of San Francisco these days – in recent decision-making. He also wishes Elon were a little bit more like?Tony Stark, who – in the Marvel Cinematic Universe at least – ends up having a heart and a conscience.

Thomas didn’t enjoy the Musk-Twitter-interview persona’s passive (and active) aggression. No filters, hyper-intelligent, and likely neurodiverse, he rips up the rule book, rewrites the rules, and those who want to carry on playing with him are required to play by these new rules. For Thomas, this is the very manifestation of?modal power, a concept coined and developed by?Steve Fuller, the Auguste Comte Professor of Social Epistemology in the Department of Sociology at the University of Warwick. Modal power explains why the right can insult the left (but not vice versa).

No time for British politics

After too much chat about GPT and Musk, we run out of time to talk about brutish, British politics. Probably just as well. When we started recording,?Dominic Raab?was still the UK’s Deputy Prime Minister. When – and I got straight into it – these show notes were half-written,?he’d resigned?in rather a bullying way over … accusations of bullying. Probably just as well we didn’t podnosticate about Raab’s chances, not least because none of Sam’s three likely outcomes bore fruit.

But we’ll be back to British politics in our next episode – out mid-May – in which we’ll survey the rubble of the defenestration of the last person standing from those who’d co-authored?Britannia Unchained.?Raab has now joined his fellow authors Priti Patel, Liz Truss, and Kwasi Kwarteng on the political scrap-heap. Perhaps?Chris “skidmark” Skidmore’s time will have come by the next time we three meet again. What will certainly have happened are the local elections in England, and a chance to see whether Sunak really is closing the gap in the polls, or whether his Cabinet has suffered the fate of many an IKEA cabinet – and the SpaceX rocket: rapid unscheduled disassembly.