AI gedachten | thoughts

A familiar fear: what happens to my work if suddenly more people arrive who also want to work, or if new technology is introduced that takes all kinds of tasks off our hands? The impact of labour migrants has been a long-running discussion, but job losses due to AI are now becoming a topic as well.

Labour migration and image

Rebuilding the Western European economies after the Second World War required capital and labour. The capital came partly from Marshall Aid from America, which helped turn Western Europe into a prosperous customer for future American big tech. We provided the labour ourselves, but at least in the Netherlands, there was not enough of it, so workers were brought in from countries where there were apparently fewer opportunities for reconstruction: first Spain and Greece, then Morocco and Turkey. The idea of recruiting these “guest workers” was initially received with somewhat less enthusiasm by the parties on the left: after all, this meant competition for our own workers. It was the parties further to the right that were positive about it: liberals in particular felt that the economy had to keep running.

After a while, however, it became clear that the “guests” were going to stay. The positive stories often focused on visible contributions, such as sport, food and entrepreneurship. Suspicion, on the other hand, focused precisely on visible differences in religion, clothing and customs.

The terms “aliens” or “foreign laborer” from earlier 20^th century had acquired a somewhat negative connotation, and the “guest worker” followed the same pattern. Eventually, the alternative terms “temporary foreign worker” or “contract labor” were coined. That worked for a while, until that term got negative connotations, too. This effect is known as the “euphemism treadmill”.

You can see that changing attitude with AI as well. Before ChatGPT, “artificial intelligence” was something far removed from most people’s reality. It was something for high-tech companies, scientists and nerds. Specialist applications, such as medical diagnostics, beating world champions at chess or Go, or recognising the number plates of speeding drivers, were known, but at the same time also very niche. Anyone who wanted to do something with AI had to be able to program in Python at the very least. And who can do that?

ChatGPT uses generative AI: it creates new data, instead of processing lots of data and then attaching a single analysis to it (diagnosis, winning move, number plate). Generative AI works on the basis of a prompt entered by the user. In ordinary language too! No Python needed anymore, so anyone can get started. After a while, the ChatGPT-like systems became so widely known that the term “AI” was enough to refer to them. After the initial fascination with and appreciation for these new possibilities, the negative aspects of this AI have become increasingly visible in the public debate. The power of Big Tech, data theft, data centres, energy consumption, slop! (Although we do have a tendency to lump all AI together here, which is unfortunately comparable to remarks about foreigners.)

This is much to the dismay of the “classic” AI developers, who only want to build a reliable app to determine whether that odd spot on your hand is something to worry about or not. Should we therefore come up with another term for “AI” too?

We have been through this before: the term AI has been “tainted” for a while more than once.

From the 1990s onwards, AI was increasingly used in all kinds of applications, but usually “under the hood” and not under that name. Only in stand-alone applications such as speech recognition or robotics was the term “AI” used. But because big successes failed to materialise, the term became a bit less sexy and many researchers started using “machine learning” or still other terms instead. A French researcher confided to me at the time that in proposals for the French government they could not use the term “artificial intelligence”, and that even “machine learning” raised too many doubts. So he talked about “automatic optimisation of parameters”.

On closer inspection, it is striking that the term AI is now used so much. Historically, AI was always precisely the thing that was not yet possible or usable.

There are many wonderful quotes about AI. In this overview, a few are brought together that all have in common that as soon as a problem has been solved (with “AI”), it has simply become “technology”. The term “artificial intelligence” still carries something magical, and we always reserved it for everything that was “not yet possible”. This is also known as the AI effect or Tesler’s Theorem.

Apart from the words we use: they are taking our jobs, aren’t they?

Just as with the “guest workers”, that does indeed seem to be the case at first. But if you zoom out a little, you see that a great deal of work for which the local people had no time, appetite or knowledge was in fact done after all, and that this helped the Dutch economy perform rather well, with our port of Rotterdam, Westland greenhouses, and of course ASML. (This may be a bit better or worse in other countries, but in most cases, guest workers do contribute to economy.)

I myself live near Leiden, with its beautiful historic city centre that radiates wealth. In the 16th and 17th centuries, Leiden was the centre of global production and trade in cloth, and also home to what was then the most admired university in Europe. All this thanks to the Flemings and Huguenots, who had fled the religious intolerance of King Philip II. Amsterdam’s leading position was also caused around the same time by capital and knowledge from abroad. The world’s first multinational, the VOC, ran for half of its workforce on guest workers, especially from Germany. More recently: in Silicon Valley, 66% of employees are non-American; at ASML this is 28%, but the majority of new employees are now foreign. The ports, distribution centres and greenhouses full of peppers are populated by labour migrants.

Historically, much economic growth has been accompanied by a large influx of foreigners. Whether the stronger economy leads to that influx, or whether that influx in turn stimulates the economy, remains an open question.

And what about technology?

With the introduction of the Jacquard loom in the early 19th century, far fewer weavers were needed. The story goes that they frustrated this new technology by throwing their clogs into it: “sabot” in French, which is where we get the word “sabotage”. Ultimately, this invention was one of the components of the Industrial Revolution, which created a great deal of work, but which also robbed most artisanal weavers of their trade.

The rise of photography meant that painting, which in theory could have been completely replaced by photography, instead set off in new directions (something worth a blog post of its own). After Deep Blue’s victory over world chess champion Garry Kasparov, chess entered a new phase (also because of the rise of online chess). Amazon has not meant that physical shops no longer exist (admittedly, there are far fewer physical shops now).

The introduction of new technology sometimes even leads to extra demand for old technology.

Astonishingly, during the Second World War the mighty German army had never had so many horses in service, at least on the Eastern Front. Even today, the British army has more horses than tanks, although that may also be because tanks are no longer so useful in today’s warfare. (I now wonder whether there are armies with more carrier pigeons than drones.)

To stay in the UK: horses were also very much in demand at British Railways; paradoxically, the introduction of the steam train led to a sharply increased demand for biological horsepower. Delivering coal, for example, turned out to be much easier to arrange with horses. In the end, most of the jobs of all those horses really were taken by technology, but only after a surprisingly long time.

People and machines are of course not the same, but when thinking about labour there are parallels. Both labour migrants and new technology often lead to more possibilities, more ideas and more economic activity. So also to more work, in which the “displaced” workers and technologies then find their place again.

The Industrial Revolution cost many craftspeople their jobs, but “fortunately” they could then get to work on the factory assembly line. Or in the coal mine.) The car displaced a great deal of horse-related work, but created mechanics, driving schools, petrol stations, traffic engineers, insurers and logistics. The washing machine reduced domestic manual labour, but also led to changes in labour participation and other uses of time. Computers and the internet displaced all kinds of professions. The typist hardly exists anymore. Who would have thought that content manager could ever be a job?

And AI?

Thinking about AI has changed since the introduction of ChatGPT: from fascination, via total embrace, to a mixture of excitement and dismay. It seems that the reversal of opinion about “guest workers”, namely that left and right swapped positions on the issue, is already happening a little around AI as well. In any case, a former tech bro like Bill Gates is sounding the alarm.

The others simply continue developing their business, but they do want to show that they are not deaf to concerns.

Dario Amodei (Anthropic) warns of very painful effects; Elon Musk (XAI) also warns, Sam Altman (OpenAI) warns, and Sundar Pichai (Google) does too. They all do it in a slightly different way, and all of them try to let some of their earlier ideals shine through.

There have already been several reports of companies rehiring programmers they had previously laid off, partly because programs written by AI lead to higher maintenance costs. The term “devibing” has already appeared: creating software with AI is called “vibe coding”, because you do not specify exactly what you want but convey more of a feeling, a vibe, from which AI then makes a program. “Devibing”, then, is turning that software into a more precise, reliable and stable version.

A job is rarely one single thing. A lawyer reads, searches, consults, writes, weighs interests and takes responsibility. AI can speed up some parts of that, but that does not automatically mean the profession disappears. Productivity will change, though. (A nice topic for a future blog.)

From the historical comparisons above, you can conclude that it is quite possible that, certainly at first, there will be more demand for human intelligence. But after a while things will change, as I described earlier: we will come to see that cars are more than faster horses. Those changes lead to friction: the Industrial Revolution led to the exploitation of workers in factories. Not everyone benefits equally. We can certainly also conclude that work shifts, splits up and takes on new forms. Forms that we cannot yet properly imagine today – and I think not even AI can.

(Note: this text is largely translated by AI from the original Dutch blog)