In Stanley Kubrick’s 1968 classic 2001: A Space Odyssey, astronaut David Bowman finds himself outside the spaceship and asks HAL, the sentient computer in control, to open the airlock and let him back inside.
HAL politely declines: ‘I’m sorry Dave. I’m afraid I can’t do that.’
Last week, I had my own conversation with a twenty-first century incarnation of HAL: ChatGPT.
Fortunately, ChatGPT didn’t lock me out of the house. Instead, we spent the better part of a week rebuilding this website.
That may sound an unremarkable task, but it left me reflecting on one of the most profound technological changes of my lifetime.
As an academic historian, my use of artificial intelligence is heavily regulated. Universities, publishers and scholarly journals are rightly concerned about authorship, originality and intellectual integrity.
Building a website is a different matter.
Ten years ago I commissioned my first website. Although based on WordPress, it still required the assistance of a professional web developer, for which I paid about $3,000 AUD. This year, preparing for the publication of my first book in a decade, I decided it was time for a complete rebuild.
I posted the brief on a freelancer website but was unimpressed by the responses. So instead, I asked ChatGPT whether it could help.
The process and the result produced some startling insights. ChatGPT did not ‘build’ the website. It frequently misunderstood what I wanted, occasionally forgot earlier decisions, and more than once led me down a technical blind alley before we (it) found a better solution. Design decisions remained mine, but otherwise I was reduced to a human cognitive sensor for the machine, screenshotting WordPress pages and amending them according to AI instructions.
What previously required specialist assistance had become something a technically unqualified independent scholar could largely accomplish alone, provided he was prepared to invest the time to be a set of eyes and keyboard driver.
That is an extraordinary development. It lowers barriers to entry. It redistributes expertise. And inevitably, it disrupts existing business models, as Joseph Schumpeter understood way back in 1942:
The fundamental new impulse that sets and keeps the capitalist engine in motion… incessantly revolutionizes the economic structure from within, incessantly destroying the old one, incessantly creating a new one. This process of creative destruction is the essential fact about capitalism.
Re-reading Schumpeter after my week with ChatGPT, I was struck by how contemporary his argument is. But creative destruction has two faces – the creative aspect, empowering my work, but the destructive element on the other side, with the future of work built on the graves of old jobs. And the process of course is rarely orderly. Profound technological changes – paradigm shifts, as Thomas Kuhn called them – are accompanied by optimism, rapid investment, inflated expectations and very often, speculative excess.
Schumpeter would probably not have been surprised to discover that artificial intelligence has already developed many of the characteristics of an investment boom (and potential bust). The enthusiasm for what has become known as ‘tokenmaxxing’ — treating AI usage itself as a measure of productivity – is beginning to encounter commercial reality. Companies are discovering that AI is expensive to operate, computing capacity remains constrained, and investors increasingly want evidence that expenditure is producing measurable returns. The current investment boom may therefore prove less durable than investors in the ‘Magnificent Seven’ AI firms might hope, and that’s quite apart from the environmental issues involved in data centre builds.
If there are signs of trouble ahead, it would hardly be unprecedented. Railway construction, electrification and the dot-com revolution all experienced periods in which investment outran immediate commercial returns. Yet the subsequent correction did not invalidate the underlying technologies. It merely marked the transition from speculation to economic reality.
The more profound question is whether artificial intelligence is simply another productivity-enhancing technology, or something fundamentally different.
Here the recent Annual Economic Report of the Bank for International Settlements points towards a much larger issue. It acknowledges AI’s extraordinary productive potential but also raises questions about its broader implications for growth, labour markets and financial stability. Implicit in that discussion is the possibility that AI differs from earlier technological revolutions in one important respect. Steam power, electrification and computers enhanced human productivity. Artificial intelligence may increasingly substitute for human cognitive labour itself. Just as I was reduced to a set of eyes for ChatGPT, it can’t be long before even that human capability is automated.
If that proves to be the case, we may face some troubling questions. If human labour becomes progressively less central to production, where does the purchasing power come from to sustain consumption? If wages are no longer the principal mechanism through which income is distributed, how do advanced economies continue to generate demand? Those are not questions about software or computing power. They are questions about political economy.
I do not know the answers. Nor I suspect, does anyone else. But these are precisely the kinds of questions we ought to find interesting. Building this website was, in many ways, the easy part. Understanding the society that artificial intelligence may help to create will be rather more challenging.