I wrote recently to a new paper in the Centre for Labour Market Research at the Higher School of Economics in Russia, that examined historic signs of the impact of automation and technology on the labour market.
There has been a growing trend of doomsday events over the last year, many of which took their cue from the notorious work of Carl Frey and Michael Osborne, that predicted vast quantities of job losses due to technology. Their work was widely criticized since they conflated the capability for technology to perform component of somebody’s work, with the ability for technology to do their entire job.
Now, what can machines do? That was the question presented in a current newspaper from Carnegie Mellon University’s Tom Mitchell and MIT’s Erik Brynjolfsson. The pair use 21 criteria to assess whether a job can be carried out by machine learning (ML).
“Although the economic effects of ML are rather limited now, and we are not facing the impending ‘end of work’ as is occasionally proclaimed, the consequences for the economy and the work force going forward are profound,” they state. “The abilities people choose to come up with and the investments businesses make will decide who thrives and that falters once ML is ingrained in everyday life, they argue.”
What can machine learning?
Tasks which are particularly conducive to machine learning would be usually those that demand an awful lot of information. We have seen a lot of examples of this in action, at the least in areas such as healthcare, where calculations which are trained on vast quantities of information are able to outperform trained clinicians. The authors are at pains to indicate that this doesn’t indicate that these professionals will be from job nonetheless.
“I think what’s likely to happen to dermatologists is they will become improved dermatologists and will have additional time to spend with individuals,” they state. “Individuals whose jobs involve human-to-human interaction are going to be valuable since they can’t be automated.”
It stands to reason therefore that tasks which are already performed online are some thing that machine learning may do well now, whilst those that require physical or personal abilities are much less so. Similarly, if decisions will need to be made quickly, then ML can certainly do this, but it combats with lengthy chains of reasoning or common sense.
Additionally, there are clear limits in the ability of ML to actually explain the decisions that it came up and we’re not talking the sort of ‘show your workings’ which may be required from a regulatory perspective. Whilst it may detect cancer nicely hence, it is very likely that the doctor is going to do a much better job of explaining why the patient has cancer.
It seems that there’s a great deal of misunderstanding at the moment about precisely what sort of impact technology may have on society so long because misunderstanding is present, it is going to be difficult to construct proper policy responses. Whilst not a full answer in itself, papers such as this one do offer a healthier contribution to the debate, and that’s to be welcomed.
“Although there are many forces leading to inequality, such as increased earnings, the capacity for large and rapid changes due to ML, in many cases within a decade, suggests that the economic effects could be highly disruptive, generating both losers and winners,” the group conclude. “This may require appreciable attention among policy makers, business leaders, technologists and research workers.”