a more subtle cost disease

hanging around, in the age of AI

Friday thorts … the Tony Blair Institute have published a thing, in which the following claim is made:

We estimate that a full and effective adoption of AI by UK firms could save almost a quarter of private sector workplace time - equivalent to the output of 6 million workers

Well, isn’t that. My problem with this is not so much with the estimate of how much time it could save, which I’m not in a position to gainsay. It’s more that even if this is correct, the underlying input-output model here doesn’t work for the kind of jobs that could use AI.

I’ll put it this way. The Tony Blair Institute work in an office (I know this, because I’ve been there). The office has a coffee machine (I’ve drunk some). The staff often hang around drinking coffee, gossiping and telling jokes. (As I say, I’ve been there). Consequently, I cannot understand why they don’t see as clearly as I do that the impact on output of full and effective adoption of any technology that saved 25% of their time would most likely be much closer to “absolutely nothing”.

I can prove it. Because for anyone who lives an hour or more away from their office, there is already a technology which could save a quarter of their workplace time - it’s called “remote working”.

(I would actually be quite interested if any of the current objections to flexible work ever get repurposed for the AI revolution. “If all these tasks are done by AI, how will our junior staff serve their apprenticeship?” “This is a relationship business and you can’t build trusted relationships with AI content”. Will we ever get CEOs sending all-staff emails saying “It has been a valuable experiment and I appreciate that some colleagues feel that AI has improved their productivity, but we work best when we work as a team, and that means a team of human beings. So, allowing six months notice for those colleagues who need to adapt their arrangements, we will be a no-AI company once more starting from Labor Day”. I have an intuition that they won’t)

Remote working did not add the output of six million extra people. Nor did the internet and nor even did Microsoft Excel. In general, productivity improvements in office jobs get massively diverted into workplace leisure. Which is like real leisure but worse.

Why? Because a huge proportion of the work that office jobs involve is that of “being available”. You’re in a big and complex system, the thing that creates value is determined by external events and has complicated internal dependencies. A lot of the time, you are basically being paid for the activity of “waiting around until it’s your turn to do the thing you do”. Sitting around, doing fairly trivial tasks, discussing fantasy sports leagues. Building relationships and specialist knowledge, yes.

But to a very great extent, just hanging around. And being paid a retainer to do so, because the company knows it can’t buy the kind of skills needed for action in a complex system on any spot market, so it needs to contract for them ahead of time. Getting better technology and tools could definitely speed things up during the periods when employees are actually creating value. But hanging around is just hanging around.

(The classic textbook example of Baumol’s cost disease is that there is no technological change which can make an orchestra take less time to play a symphony - service industries don’t have the same productivity improvements as manufacturing industries. In many ways I’m talking about the same thing here, but emphasising a different level. However advanced and well-engineered the triangle player’s instrument is, he is going to spend most of the evening not playing).

My own career history is a fairly extreme example of this. Former colleagues of mine subscribed to this list might recall that I was absolute dynamite during financial crises - no false modesty here, there were not many who could match me. They will also recall that outside of crises, I was kind of abject, like to the extent of regularly being unable to complete a simple Excel model. Everyone who employed me (and I was very lucky to find enough of them who understood this) had to accept that I was going to spend stretches of years at a time being a useless drag on the franchise, then make them a ton of money in years four and five. An AI copilot would really have helped me during the Eurocrisis, when I was rushing round trying to explain the same core facts in about a dozen different investment contexts. But there’s no technology on earth that could have made the years from 2003 to 2008 pass any faster.