As new-fangled a concern as ‘losing your job to a robot’ sounds, one commonly cited Oxford University study predicted it several years ago – and its authors decided that, yep, bloodless automata were indeed going to take over roughly 1 in 2 careers during the next couple of decades.
Well, ok, that’s not quite what the study said. The actual finding of the 2013 research team, led by Drs Michael Osborne and Carl Benedikt, was that up to 47 percent of jobs in the US were “at risk” of being automated in the next 20 years. (There’s still some debate about the precise order it’ll happen in, although experts are largely in agreement about which jobs are most likely to be affected).
Noting an alarmed response, Osborne was keen to point out that his study “identified several key bottlenecks currently preventing occupations from being automated” – but he also confirmed that “as big data helps to overcome these obstacles, a greater number of jobs will be put at risk”.
This won’t have come as much of a surprise to another expert in the field of ‘machine learning’, namely startup leader Anthony Goldbloom. As the founder and CEO of data analysis outsourcing platform Kaggle, Goldbloom was invited to give a talk on the subject of machine learning at a TED conference last year.
He used his five-minute slot to talk about how quickly robot workers had gained ground on us over the past twenty years, how easily they’d surpass us in future – and which specific areas they’d continue to struggle in.
Goldbloom pulls no punches in singling out “the technology responsible for most of this disruption”: he calls machine learning “the most powerful branch of AI…it allows machines to learn from data, and mimic some of the things humans can do”. Interestingly, his own company, Kaggle, works right on the cutting edge of this very field – and he’d be the first to admit that machine learning is all set to cause a huge, potentially very painful shakeup in global employment sectors over the coming years.
As long ago as the early ‘90s, ‘smart’ machines were already being put to use in mailrooms and insurance offices around the world, where they’d often be delegated responsibility for covering some of the more menial or admin-heavy tasks – things like assessing liability risk based on completed application forms, or sorting deliveries by ‘reading’ handwritten addresses.
Since then, though, the progress made in developing similar kinds of AI learning has been phenomenal.
Today, data-processing algorithms have already been created that mean a computer, given enough raw sample data to work from, can grade a child’s essay to within an entirely reasonable margin of error compared to a human teacher. Ditto automated medical diagnostics programs, which can now detect numerous diseases and conditions by cross-referencing patient scans with a vast database of previous cases. Here, too, the margins of error we’re now seeing are already very comparable with those of the average flesh-and-blood doctor.
The real key to all this, as you’ll no doubt have spotted above, is data. Feed the machines with enough raw data to ‘learn’ from, and they new results they can generate off the back of it will outperform humans for speed, efficiency and consistency every time.
In fact, says Goldbloom, in terms of the machines’ ability to perform frequent high-volume tasks like these, our species has “no chance” of even competing in the future: not only do robots require no payment for their services, but in mere minutes they can pull down and cross-reference a database that will dwarf the career-long experience amassed by any one human (or even groups of multiple humans working together).
So that’s it – we’re about to become obsolete, right? Well, in some ways, we pretty much are. And yet in others, far from it.
The thing is, there are still some very specific areas in which machines – for all their meteoric progress in ‘teachability’ over the past couple of decades – remain well and truly stuck at a pre-school level. Moreover, to date they’ve shown pretty much zero sign of progressing much beyond that any time soon.
Once again, they key to this is data, and how utterly reliant on it the robots are for successful completion of those frequent high-volume tasks. As soon as you throw what Goldbloom calls a “novel situation” at them – in other words, when you ask them to perform without previous data to go on – they’re actually kind of useless.
Humans, on the other hand, use deductive reasoning in ways that computers and algorithms simply cannot replicate yet (or indeed get anywhere close to). We instinctively map our experiences in one field – our data, as it were – on to other unrelated and entirely uncharted fields. This is the basis of reasoning and judgement, and, give or take, we do both consistently well. Machines, even in 2017, remain basically horrible at it.
With that in mind, here are some areas of future employment that currently look fairly well safeguarded from the imminent threat of bot workers:
1. Non-repetitive jobs. Does your job involve tackling a subtly different set of circumstances each day, relying on astute judgement on a case-by-case basis, with little or no use for sweeping generalisations or referral to fixed precedent? You’re probably ok for a good while yet, then: we’re still a long way from devising a computer program that can reliably think on its (non-existent) feet.
2. Roles that require regular training or skills updates. Actually, roles that require no training whatsoever could also be included here: unskilled manual jobs tend not to be worth automating, as the cost of robots that can perform even basic physical tasks is astronomical, while wages in this bracket tend to be low. However, they’re often among the least satisfying jobs, too – so it’s probably better to aim for the ones that require frequent upskilling, refresher courses, or taking on new directives. Sure, those are costly, complicated and time-consuming to implement for human workers…but even more so for machines.
3. Creative jobs. Are you responsible for making humans feel something in your job – say, through crafting persuasive marketing copy, or by staffing a one-to-one service helpline? No immediate panic, then: you’re already orders of magnitude better at that than any robot we could even dream of creating right now. Keep it up, fleshbag.
4. Roles that have scope for growth, or taking on new challenges. The machines’ complete reliance on data means they’re not at all well suited to expanding their remit, or in any way expanding the reach or nature of what they’ve been programmed to do. If your employer is generally open to new ideas, or to you taking on additional responsibilities, then that should stand you in pretty good stead against any mains-powered colleagues you might go up against in future.
5. Jobs that rely on personality. Perhaps the most obvious of them all, but one that’s often overlooked: if you deal directly with other humans in any aspect of your job, and success or failure hinges on your ability to interact appropriately and personably with them, then you’re going to be awfully difficult to replace with a bunch of circuits and wire – now, and for a long time to come.