A persistent modern fear is that artificial intelligence and robot technology will advance so much that smart robots will soon be able to perform many of the tasks that we humans currently earn our crust with. Since they will come off the production line in a matter of minutes, they will be lots cheaper than us.
With the loss of our productive value, the fear is that our political value will soon follow: the value of a human life will sink to the cost of replacing us with a robot that does not need 20 years of feeding and education, but can be programmed in seconds. Why bother sustaining huge investments in unproductive humans? The usual historical answer has been that rulers do not bother with humans for whom they have no use.
Much has been written about this, but I here want to take the perspective of mainstream economics because I think it is particularly sobering and useful for this sort of question: mainstream economics already takes a very long-run view on markets and production, so now that we finally have a question that is truly long-run in nature, we can draw on standard economic thinking to illuminate the way.
Within mainstream production-function economic thinking, advancement of AI technology is like a drop in the price of a particular form of capital, sustained and followed by an increase in the use of that capital.
In the short-run, it is possible that this additional ‘IA’ capital will substitute for all kinds of current investments and for labourers. This is what new technology in the past has done, from the agricultural boom of the 19th century to the desktop age we live in today. So there will surely be losses to labour in the short-run.
But in the long-run, the glut of cheap new capital eventually raises the value of all other factors of production, including labour, unless and only unless labour can be fully substituted for the newer form of capital.
It is the standard logic of production-function thinking that any factor of production that is available at almost zero cost raises the value of everything else that is not a complete substitute (of course we can think of weird production functions that don’t have this property, but I am yet to see one). The argument is that whatever cannot be perfectly substituted starts to become the relatively scarce production factor whose value increases. This does not mean that labour would necessarily start to command all of the output, but it does mean that the marginal value increases and that hence labour would get a bigger piece of pie, even if it is a relatively smaller piece than before.
This is arguably what the agricultural revolution and the desktop age lead to: humans adapted so that they themselves specialised in different skills than before. The new skills were worth more than the old skills because of the cheap new capital. Hence droves of poor peasants were replaced with somewhat richer factory workers working alongside lots of machines. And those factory workers were subsequently replaced by even richer office workers who work with lots of computers.
There were losses in the interim, but in the long-run we were better off: the labourers found other human skills to focus on as the previous ones they focussed on became of little value. Those new tasks did often not even exist before the revolutions that made our previous jobs obsolete.
So in the long-run ‘we’ are going to be fine as long as there is something, no matter how unimportant or rarely performed by us currently, that robots and computers absolutely cannot do, no matter how much research time the research-robots of the future throw at it. That something has to lead to the production of some output that has consumption value to humans. Something that is pivotal in the production of either some basic goods, or in social goods that cannot be perfectly mimicked.
Even if it is sex, gladitorial entertainment, wombs, visual acuity, or humour that saves us, ‘we labourers’ would then be fine. In absolute terms, life would continue to get better: the value of our unique ability will creep to infinity and our families could live on a smidgen of that skill. So even if our ‘uncrackable skill’ is something as simple as a good joke, then whole human dynasties would in the long run be able to survive on just one decent joke in ten generations. Even the Germans might manage that.
If there is no human ability that cannot be mimicked perfectly well by robots, then, to put it in technical economic terms, labour is stuffed. Our value as labourers then eventually drops to zero and we can expect only the owners to prosper, and even their numbers might dwindle as they compete with other owners and their super-intelligent artificial advisers.
Is there hence something that is totally out of reach of AI and robot technology? Frankly speaking, I personally suspect the answer is ‘no’, but I hope it is ‘yes’. It is more a question for engineers and computer programmers though. Perhaps the reader can think of something?
Will it matter if we humans team up with AI and robots, so that we get augmented humans that combine the ‘best’ of both technologies, which is for instance the future envisaged by Raymond Kurtzweil, who famously predicted that in 2029 artificial intelligence will emerge that is indistinguishable from human intelligence? Think of additional memory and processing capacity outside of human brains, but connected to it.
This kind of ‘interfacing’ will not matter in the slightest in the long-run, for the same relentless economic logic applies: if human interfacing with other technology really becomes widespread and valuable, then the value of figuring out how to replace the human bit in that arrangement will keep rising. So any ‘hybrid’ technology that still requires a human bit is also doomed to be replaced by a fully artificial entity that does not require 20 years of feeding and education, unless the human bit truly cannot be replaced. Raymond’s prophesy thus needs AI to have its limits!
It is hence only in the short-run that it matters whether or not there are particular combinations of human and non-human technology that are more productive at something than either on its own. In the long-run, the key thing that matters is that humans truly have a productive skill that robots and AI cannot attain.
At present, we humans can still do lots of stuff that robots and computers cannot. Unlike computers, we can make babies, recognise a keyboard from a pizza 100 times out of a 100, and run with two legs without looking ridiculous. Well, maybe we cannot quite do the last one!
Interestingly, what I believe are amongst the easiest skills to mimic with AI are research skills, so my kind might disappear more quickly than others. That is because people like me get sit on their behinds all day, don’t have to recognise anything visual, and communicate in a language (maths and formalism) that is already pretty close to computer language. Hence a robot replacing me does not need to walk, recognise his surroundings, be able to hold a conversation, or even understand other humans. Indeed, evil tongues have rumoured that understanding humans is a liability anyway in some branches of social science!
Harder skills to mimic have to do with lots of quick spatial and visual recognition, combined with reading other humans. Those skills requires mobile robots able to simultaneously understand many different entities in new surroundings. Think of plumbers who have to understand the garbled mutterings of distraught home-dwellers whose bathrooms just mysteriously flooded, after which they have to creep through weird cupboard to reach some medieval boiler that needs special re-fitting.
It is somewhat of a comfort to fantasise that the last human workers to get replaced are tradies. Meanwhile, I will console myself with the hope that me and my loved ones are ancient history before that long-run arrives.
Paul, this is an absolute possibility and is happening now. I watched a show on the telly some months ago (and for the life of me I cannot recall the title of the show (must be the heat !) – will keep searching) where a vehicle production line was filmed, and there was ne’er a human to be seen. It showed the engine being placed in the engine bay, and with ultra precision placed it perfectly on the engine mounts. Even the engine bolts were placed and tightened. So even bolt placers and tighteners are not needed.
There was one human walking along the line with the outdated clipboard and biro (oh so yesterday !) – just observing everything, or perhaps to see if any robots started to get cranky and nasty, and take over the joint, or begin duelling with their high precision jaws.
I would imagine the management of the factory spend most of their time looking at screens since there would be almost no need to manage people (does the tea lady need managing ?), considering they are in very short supply. The HR department probably consisted of one.
Yes, we will still need tradies for their vital work, but even plumbers have been challenged with the “flick mixer” taps, which seem to last forever. I suspect the trade in tap washers is diminishing.
BTW your articles are always informative, instructive and educative to read. Keep it coming ! Are you in London yet ?
Andy, the car assembly line is a very carefully constructed, very expensive, very artificial environment. And there have been major parts of assembly that have had virtually no humans for at least thirty years. Not really a good example of the kinds of AI that are coming on stream now.
“I would imagine … there would be almost no need to manage people (does the tea lady need managing ?)”
Exactly wrong. Remember, we get more of the types of labour that cannot be substituted.
The tea lady can certainly be replaced by a smart robot that delicately dunks a gratuitous gingernut bicky into your tea for you – and dunks it better than you could ever do – but the person who pens the minute forbidding such generosity as a violation of company policy will never be.
I wonder how long it will be until the robots are labour rather than capital. In one of his stories, Charles Stross speculates how the limited liability company might form the legal underpinning for person-hood for AI.
I said that 15 years ago in a comment somewhere in some incarnation of this blog.
Sadly, not an original insight. Read Kurt Vonnegut’s first novel “Player Piano” (published in the 1940s).
Talking about a UBI or social dividend as being the feasible and likely response to mass technological un- or under- employment misses the point that it’s certainly feasible but not likely.
A generous unconditional UBI is technically feasible now – but it is politically extremely unlikely. And the more AI-driven inequality we have then the less politically likely it becomes, even as it becomes an ever better solution to that inequality. Because the more unequal a society, the more fear and downward envy amongst the shrinking middle class, and so the more they want to believe everyone is getting what they deserve.
Humans are weird – most people are much more like Job’s companions than Job (the only book in the bible where the psychology rings true, BTW, despite its rewritten Hollywood ending). Their reaction to others’ misery is not “there but for the grace of God go I” but “If I believe it can’t happen to me then it won’t happen to me. So I believe they must have deserved it and I don’t. The judgements of the Lord are altogether righteous”.
There is probably someone in the robot design cell working on reducing the wheel nut “place and tighten” procedure, from 67 seconds to 49 seconds thus achieving a 26% reduction in tyre placement time. That would look good on the resume.
New potenital area for study:- “meta-demand side” economics. What economic factors lead to a demand for demand side economics when anything, any item can be produced and delievered in under 30 seconds.
Meanwhile some moron worries about dietary restrictions.
Hi Paul
Some trades are effectively already being largely replaced with prefabricated factory made components. For example:
Many of today’s builders simply ‘bang together’ ready made components and that trend can only grow.
And many car mechanics these days do little more than ‘ just plug the car into computer run diagnostic software’ and then ‘pull out an assembly and replace it with a new pre-made assembly ‘.
In the case of cars, while there is a genuine and growing ‘enthusiasts’ market for people who have the skills needed to work on classic cars, think its unlikely to ever be that large.
I started off in machine learning, but didn’t pursue it because I didn’t want to spend my whole life thinking about maths — however, based on this and still having some interest in what is going on, I think your idea that “Harder skills to mimic have to do with lots of quick spatial and visual recognition” is entirely the wrong way around.
These are the easiest things to learn — this is what self-driving vehicles use and the algorithms have been around for decades (it just took GPU cards from gaming technology to allow them to work fast enough). Try looking at, for example, Yann LeCun’s web page when this stuff was applied to recognizing digits — which was about all you could compute then. They’re basically what the general public know as “deep learning networks”, of which there are two main types.
My feeling is that this looks like a major advance (it is in some respects), and so the public get excited, but it’s very context dependent and when something else doesn’t come along quickly, people forget. You might remember VR was supposed to take over the world in the nineties, but now it’s scope known (games, a bit of simulation technology).
Which is a classic illustration of what people like Robert Gordon have been saying for a long time – that technological innovation has a history of transforming our lives far beyond even its originators imagine, but taking a far longer time to do so than anyone expects.
The reason is developing new technologies is the relatively quick and easy part. Embedding them into our society and economy is a far more gradual – yet traumatic – process.
If technology can replace human labour to satisfy all our needs and most of our wants, it shouldn’t be too difficult to give everybody an income so that they can demand what technology can provide. Some in the technology owning class might feel entitled to all of the wealth but they will be a minority and they can be constrained through the ballot box. Maybe most of the technology could and should be socialised at this historical juncture. Maybe this could bring about genuine Communism, whereby a man will truly be able “to hunt in the morning, fish in the afternoon, rear cattle in the evening [and] criticise after dinner” as Marx wrote in Die deutsche Ideologie.
Paul, the thing you haven’t really talked about is who owns the machines – who draws the dividends from their production. Surely that’s the critical thing in the foreseeable future.
IMO there are three major possibilities, Nicholas:
1. the machines are owned by a small owning class who reap all the benefits save a few crumbs, thus immiserating most of humankind
2. the machines are owned by a small owning class but they are taxed heavily so demand for their services can be transferred to the non-owning class
3. the machines end up being socialised
Possibility 2 is preferable unless the machines can innovate and improve to such an extent they render obsolete the need for capitalism, which currently is responsible for much of the innovation and improving.
I agree that is quite crucial and I did talk about it, but very summarily. I didn’t want to sound too negative.