As some may have noticed, I’ve been musing of late about the likely future social and economic effects of the increasingly rapid and interconnected development of ICT, artificial intelligence and robotics. This article is a bit silly in some respects but makes some useful points succinctly:
Speaking at the World Business Forum in Sydney on Wednesday, Ms Erickson said new technologies, changing demographics and the move to “knowledge work” would soon lead to an overhaul of workplace arrangements on par with the shift from craft-based industries to mass production lines during the industrial revolution.
Just as in the ‘80s and ‘90s the lower cost of communication from the introduction of computers allowed companies to radically restructure and outsource non-core activities, the same is about to happen thanks to the “lower cost of co-ordination” today, she believes. …
Just as machines have taken over industrial work, it won’t be long before they start taking over “knowledge work”, she argues.
Thinking machines are likely to take over a lot of the work we previously thought could only be done by humans. IBM’s Watson, for example, has gone from winning at Jeopardy to delivering world-class cancer treatment.
The two areas least likely to be taken over are decision-making and relationship-building.
“That is, to make choices when the options are all pretty good, the options are rational, and when the decisions you have to make are not based on fact but on values, ethics or strategic issues,” she said.
“And to connect, to form relationships. There I think people will continue to have important roles. Those are the people you’re going to be supporting — the rest of it is going to go to the machines.”
I’m not at all sure that high level subjective decision-making will remain beyond the capacity of AI for very long, although it will be wise to reserve those sorts of decisions for humans for other reasons, as Dr Dave Bowman discovers in 2001: A Space Odyssey.
As I have suggested in the context of “ride share” ICT systems like Uber, the world’s transportation industry will almost certainly be revolutionised within the next decade or so, with most human jobs eliminated. Other areas may take a bit longer, but it is likely that a very significant proportion of “knowledge work” jobs will have been taken over by machines in the next 30 years or so.
In past phases of the industrial revolution and post-industrial era, technological advances have in the long run always spawned new and better employment opportunities, but it is difficult to see how that will happen in the near future as machines increasingly take over roles involving significant elements of cognitive capacity. That may well mean that a significant proportion of people, at least in advanced economies like Australia, will be unable to obtain remunerative marketplace employment. That is the context for my tax reform suggestions in a recent post where I proposed a generous Negative Income Tax (universal basic income) funded by significant taxes on high income earners, capital gains, land and inheritance/death duties. The immediate self-interest of the owners of capital suggests that this sort of proposal is unlikely to get very far as an achievable political reality until it becomes self-evident that some such choice is unavoidable, but it would be wise to begin a serious discussion well before then.
Capitalism itself will be under threat when a large proportion of consumers are unable to afford the products and services that business creates because they can’t get a paying job. It will become necessary to redefine the nature of productive work and underpin consumption (and indeed human society) with a universal basic income that is independent of any individual’s ability to generate an income in the capitalist marketplace. If the forces of capital fail to adapt to that reality they will certainly face a potential revolutionary situation as desperate citizens fight for survival. Responses in the wake of the Great Depression, when advanced Western economies belatedly saw the wisdom of enacting a reasonably comprehensive social security safety net, suggest that our societies will have the capacity to adapt to meet such a situation in the future, though probably not until it’s staring us in the face to the extent where not even the most obtuse and greedy corporate CEO can deny the necessity.
In the medium, having machines just as smart as us might help humanity as these machines will speed up techno0logical progress in areas we approve of (health, exploration, artificial limbs and brain supplements).
In the long-run, it is a nightmarish scenario. In the long-run, those groups and humans who cannot be part of the production chains lose their political power and even their legitimacy as human beings. Nightmarish.
Hi Paul. This article published a couple of days ago deals with exactly the point you’re making:
Of course, I’m envisaging a society in the fairly near future where receipt of a NIT/UBI has become a matter of unavoidable necessity rather than “choice” for a high proportion of people. In that situation, attitudes towards income-earning work, status and social legitimacy will inevitably change, although exactly how and whether it will be for the better is the sixty four billion dollar question.
I don’t find it mysterious to predict how attitudes to those who cannot keep up will change. It wont be a question of what the broad society thinks, but what the owners of the machines think. If most of humanity becomes worth no more than what it costs to make an equivalent machine (which wouldn’t be much), most of us will have no future. A machine could be programmed in minutes, we need to be fed and educated for decades before we become useful: no contest as to what is going to be cheaper.
Our political systems might resist for a while, but the AI owners will take the opportunity to pursue their relentless dominance drive. Nightmarish. I hope that Conrad is right and that AI will simply not develop to that degree, though I don’t think he is. I think the software bottleneck that AI faces at the moment is passable.
I think the whole AI thing is over-rated — every time there is some new break through, suddenly people think machines will be talking to them intelligently in a decade. The main thing that has allowed the latest development are better and faster chips (basically very powerful GPU boards) — the types of algorithms they are using for the driverless cars, for example, have been around since the early 90s, it’s just that people didn’t have the computer power for them and so used those algorithms for other simpler tasks like digit recognition instead. Whilst the results on this particular task are clearly impressive, it is still a fairly constrained task (basically, recognize lots of visual features in lots of things).
The other obvious problem is that Moore’s law is coming to an end. You may have noticed that your computer hasn’t got heaps faster for many years. The main way the new algorithms work is by exploiting parallel processing (hence the GPU cards), but there are only so many types of task this can work for and designing good algorithms has always been a big bottle-neck for computing problems (especially parallel ones). It is also the case that it is going to get increasingly difficult to minituarize chips further, because the chips now are at 14nm, but you need other material than chips are made of now to get below 10nm, which no-one has done well at all (although it’s been done). This also doesn’t lead to much faster chips anymore, but more nodes on a chip to do things in parallel if you can think of how.
Conrad, you may or may not know that there’s a sizeable faction in the world of AI that agrees we’re some way off machine intelligence.
One such thinker is MIT’s David Mindell, author of Our Robots, Ourselves: Robotics and the Myths of Autonomy. A summary from a recent INTHEBLACK article:
Some anecdata in support of Conrad and David. Early this year a most unfortunate event brought my partner and I back in contact with a friend who is at the forefront of natural language processing (NLP) research. His two minute summary was pretty much everything we’ve done until now is a dead end, and only good for canned demos. We have no real idea how to capture meaning in a NLP system, e.g. venture too far beyond setting a timer and systems like Siri are lost.
If I were to hazard a guess re: the impact on society more in line with the OP, then like other posters have suggested there will be a real hollowing out of the middle as anything requiring human mobility is probably safe for the time being, so construction and trades, and probably basic unskilled tasks like supermarket shelf stacking (barring a complete overhaul of how we shop for food). R&D, and applying reasoning and insight across domains will be pretty resilient IMO, but if your job is to follow a deterministic process in a well defined environment then start getting creative. So lots of what Paul and you listed where people are just applying tax law etc. A small niche may remain for the more creative accountants :)
I fall into the same camp as Paul F w.r.t the likely outcome for society, history suggests it will be ugly.
Lastly human brains are pretty damn power efficient so there’s that, e.g. ~20W vs I assume quite a few KW for AlphaGo, and AlphaGo plays Go. Nothing else.
To the extent that automation of “knowledge work” is happening or will happen soon, as Tammy Jackson suggests, it seems to me likely that much of it will happen largely because many of the things we thought of as demanding the intellect of lawyers, accountants, middle managers, doctors, nurses, journalists and taxi-drivers are actually not as intellectually demanding as we thought they were. Think autonomous cars, automated legal discovery, and accountancy software that talks directly to the Tax Office.
All of these take up the time of “knowledge workers”, but none of them is exactly a triumph of human brainpower.
“Knowledge work” where the knowledge is familiarity with a large set of rules (which takes time to acquire for humans so requires specialisation), so you can apply them quickly and with a fair degree of accuracy to a mostly static environment. Close enough to what computers can be programmed to be good at, and for the edge cases send them to a reduced number of human specialists with details of where the program stalled (after which the solution is fed back into the system and that edge case is less likely to stall the program). Or if you don’t care about low probability edge cases then done!
In the finest tradition of programming I may have simplified the whole process a bit :)
I suspect that AI will keep ploughing along, albeit at a slow pace. I suspect that it will never fully replace the “sensory” attribute of human thinking, but will come close.
But we have seen the automation of commercial business, and that will continue its march forward. Has anyone been into a bank branch in the last 3 years ? I have not been into a branch of my bank in the last 10 years, and I see no need to do so in the future. Banks have closed branches rapidly (particularly in regional areas), and converting them into agencies (Auspost etc)
I did read of a proposal to have “multi- brand bank shops”, that is, as you enter a banking chamber, you proceed to CBA on the left, Westpac next, ANZ next, and so on to 10 or more brands segregated in the one chamber. Each would have their own networks, and one or two uniformed staff. The idea was to save on real estate costs, security (although quite small amounts of cash were to be kept). It was aimed at business banking for traders wanting to deposit their daily cash takings, although with the use of cards for purchases, the cash component was to diminish.
Similarly, it was suggested that fuel stations would become “multi-brand” pump farms, all brightly identified and segregated, and individually priced. So if your preference was for BP, then you would drive to one of the 8 BP pumps. Whether they would draw from the same underground tanks was not known. Again it was all about the real estate.
It is this automation and rationalization of commerce that will have a very telling effect on the general workforce.
Speaking of multi branding. I lived in a rural area for sometime, and the local farm supplies shop had a particularly well made chain saw in stock. It was a robust, well designed and engineered machine, and there was a good back up of spares (in the rare event they were needed). The owner of the shop was awarded a trip to the German manufacturing plant (for reaching a high sales figure).
He got a surprise to see that the plant produced two models only (a larger and smaller version – but made from the same fundamental parts). The difference in the assembly runs was the colour of the cowling and stickers on the machines.
“Zis month ve are making zer saws mit zer ORANGE cowling and stickers and call it zer ABC machine, and next month ve vill make mit zer green cowling and stickers and call it zer XYZ machine. Ja! “. The idea was that parts were common throughout all colours and names, resulting in small stock inventory. And to concentrate on quality. They were sold to distributors who could buy a container load at a time, and the distributors were well supported.