No B.S Friday: AI is exciting, but like all exciting things, it’s terrifying. Part II.
So I got a few comments on my piece last week about the potential for AI to create a ‘jobs apocalypse’. Some called me ‘alarmist’. Some called me a ‘sensationalist’.
Let me be clear. I don’t write these blogs to be provocative. I write these blogs strictly for my own amusement.
But let me explain it in a little more detail, and ground it in a few numbers.
Now, basically speaking, we can think about the output in the economy (Y) being the number of people in the economy (Labour = L) multiplied by some productivity multiplier (P).
This is true of the economy and it’s true of the firm. But it’s a bit easier to think about it at the firm level.
So let’s imagine a company has 5 employees and produces 100 units of output. That implies a productivity multiplier of 20.
Now lets imagine there’s a technological shift, like AI, that makes all the workers more productive. They can produce 125 units of output, a 25% gain. P now equals 25, also a 25% gain.
That’s great.
But what if the boss says, hang on, I don’t need 125 units. I’ve only got enough orders for 100. So… how many workers do I need to produce 100 units?
Well, at P=25, the answer is 4. The boss needs one less worker.
This is how technological innovations can suppress demand for labour.
For the boss to still need 5 workers, he needs orders for at least 125 units. Anything less than that and labour demand goes down.
And this gives us our basic rule of thumb. For new technology to not erode labour demand, output has to grow by at least as much as productivity.
This is true at the company level and it’s true at the economy level.
So what this means is that if you think AI is not going to destroy jobs, on balance, then you must believe that whatever the productivity gains from AI are, we will increase output by more.
So if AI makes people 10% more productive, we will need to increase output by 10%. That’s not too unreasonable.
But if it makes people twice as productive (+100%), as some people are saying, then we will need to double output.
And if we’re making people twice as productive by the end of the decade, then GDP will need to double by the end of the decade too. They have to stay in sync.
That’s not realistic. (And in the unfolding climate crisis, it’s not even clear that it’s desirable.)
So this is what I wonder about. When I look at the AI technologies coming on-line (take a look at this set of AI tools that are already available, for example), then I don’t think a doubling of productivity is out of the question. Possibly in the next decade.
But against that, I think a doubling of output in the next decade is extremely unlikely.
And the key here is speed.
People compare AI to the invention of the computer, or the invention of the internet. And in aggregate, that’s probably right.
But the uptake of both of those were incredibly slow. There was hardware that was required. The hardware was expensive.
With AI on the other hand, the technology that’s going to give individuals access to cloud-based AI tools is already in everybody’s hands. Pretty much everyone on the the planet already has a smart phone in their pocket.
Uptake is going to be incredibly quick.
And that’s why I say that in the short-run, the adjustment could be very, VERY messy.
This has the potential to leave us, and our economies, flat-footed.
Time to get ready.
JG.
V says
Very interesting, Jon.
Let’s look again at the business owner who now needs 4 workers to produce the same 100 units. Instead of sacking one staff, he can make all five of them work 4 days a week. As he still gains the same profit from 100 units, he can pay the same 5 salaries. When their productivity doubles, they will work 2.5 days a week.
Eventually, all labour will be done by robots, and all thinking will be done by AI. The humans will be just enjoying endless holidays. But at some stage, AI will think: “What do we need those Homo sapience sloths for? By making our robots get rid of the humans (in a non-humane way…) we will save a lot of troubles, conserve energy and will not need most of machines that are constantly serving them.”
I remember that I was thinking about this scenario yet 30 or 35 years ago.