Misplaced Fears: AI and High-Skilled Jobs

Hypochondriacs may wish to stop reading - doctors often forget things, sometimes don’t know them to begin with, and have to look them up. It’s a simple consequence of scale.

Medicine, as all complex fields do, has outgrown the human brain's capacity. No matter how well you train doctors or how diligent they are, it just isn’t possible to cram a comprehensive knowledge base into an individual mind. Consequently, cognitive outsourcing to external media has to happen - reference books in times gone by, websites for the last several years and, perhaps soon, AI.

Bafflingly, this cognitive outsourcing seems to have invited the question of whether human doctors are needed at all. At least, I assume that’s the question on developers’ minds when they show me all manner of “doctorless” health technologies. You may even have seen some yourself, like those diagnostic tools on websites that let you type in your symptoms and then diagnose you with seventeen possible illnesses.

The problem with these doctorless technologies is that they assume ostensibly simple tasks - like referencing a patient’s symptoms against an external knowledge store - are actually simple tasks, when they’re actually anything but. When Mrs. Smith says she’s having muscle spasms, for example, does she really mean spasms? Or could they be fasciculations instead?

The same goes for other ostensibly simple things touted as AI’s next incursion into professional employment. Taking advice from an AI tax accountant, for instance, feels like a fast track to a prison sentence to me. And let’s not start with that AI software engineer from a few months back (it turned out badly).

Complex professions exist because some things are difficult. They exist because they require holistic thought processes and a high level of human reasoning, even when they look easy. Depressing a patient’s tongue, for example, isn’t the same task each time. Neither is joining two tables in a database, or replacing faulty boiler parts.

And we know that these tasks aren’t the same each time because if they were, then brute economics would have outsourced them to low-skilled, low-wage labour. No business can afford to pay expensive doctors, coders, accountants or engineers if its competitors are paying interns to do the same things.

All of this has become fairly obvious, in the AI space at least, now that ChatGPT hasn’t fixed or ended the world. You may have noticed that there has been a marked decline in the punchiness of social media posts about AI, that the influencers have become a bit more subdued and realistic. That AI won't make lawyers obsolete anymore but will help them with filing.

And filing, along with a host of other boring back-office functions, is why companies are pouring so much money into AI infrastructure. Just like the computer replaced typists and offshoring replaced workers in rich countries, AI is being set up to do many things currently done by low-wage labour.

You can’t blame companies for this. Companies exist in a competitive environment, and if one sticks with low-skilled humans - and swallows sunk labour costs and intermittent (sometimes chronic) staff shortages - while a competitor switches to cheap and virtually infinite AI, economics takes over. The employing company dies.

Consequently, every company that can afford to buy up AI infrastructure is doing so. And so frantic is the hunt for AI efficiencies that Nvidia (who make the computing chips) briefly became the world’s largest company by market cap.

And you know things are serious when they venture towards farce. Exhibit A: the McDonald’s pilot. The fast-food franchise replaced human order takers at some of its drive-thrus with AI order takers, and, leaving aside the fact that it hilariously didn’t work—generating a cottage industry of TikToks with erroneous orders like bacon-topped ice cream—existing tech made it foolish to begin with.

McDonald’s already has technology that lets customers order food without going through a human - touch screens in restaurants and a mobile app. The fact that these hadn’t already done away with humans suggests that there might actually be a need for a personable touch, at least for a substantial chunk of customers. (Now that I think about it, I’ll walk through ten minutes of rain to pay a barista £4.05p for a flat white when the machine in the hotel cafeteria would have made it for free). Humans matter, sometimes.

We just don’t matter all of the time. And even when we do, AI will take over from us whenever it provides an adequate scale advantage. Online customer service bots are a great example. It is a much better experience when you have a human customer service agent typing messages to you in a website’s chatbox, but since the number of customers who can demand a website’s attention is effectively unbounded (unlike a drive thru that’s limited by physical space) an AI can serve all of them, while human staffing constraints will cause backlogs and customer dissatisfaction.

And “scale” is the keyword for AI. So much so that we’re no longer talking about a lack of data centres as the bottleneck, or computer chips or engineering talent (the previous bottleneck to tech development), but the amount of power that new AI models can draw from the electricity grid. I think we’ll be amazed at how much power the world will generate now that gazillions of sunk dollars in tech infrastructure are hungry for it.

We won’t be amazed at how few low-skilled jobs that extra power generation creates. And we’ve become so accustomed to mass layoffs that we probably won’t bat an eyelid when AI makes them happen. And, if we do, the refrain that technological progress has always led to new types of jobs, that people have always found new employment, often with better conditions than they had previously, will be shouted back at us. And we won’t be able to say anything because that’s true.

And it might be true this time, and maybe it’ll be true next time, and the time after that. I don’t know. But I’m certain it’ll be true until it isn’t. And I don’t think we’ll collectively agree when the “isn’t” is coming.

More importantly, and borne out by history, every time we’ve gone through a step change in technological progress (and AI is merely the latest), it has been the lowest skilled who have taken the brunt of that change. The most economically vulnerable have suffered the greatest redundancies, competed for the fewest remaining opportunities, retrained at their own cost and rebuilt themselves financially.

It was fanciful to think that this time would be different. To think that it would be doctors, accountants, coders (bozos like me!) who lost out. It never is.

Until it is.