“The Bot Is Out There. It Doesn’t Feel Pity, Or Remorse, Or Fear, And It Absolutely Will Not Stop Until Your Job Is Dead”
By Michael Every of Rabobank
AI ai ai!
Last night we had a fascinating team discussion about the implications of ChatGPT and AIs. One conclusion reached was that replacing brains with a machine will do to middle-class jobs what replacing muscles with a machine did to working class jobs: lots will be destroyed, and those left will have to be either very skilled or very unskilled.
Of course, productivity will soar, but so will political-economy tensions over how to divide those gains fairly. We didn’t do that for the working class, as they tell us anytime anyone listens to them, which is never for people with middle-class jobs – until they see yet another shock election result.
Ironically, one of the AI answers we were shown was to the question of the implication of higher interest rates on capital-labour dynamics from a Marxist perspective. The answers it gave were good… if you don’t know a great deal about Marx. The bot said higher rates would mean lower profits, so downwards pressure on wages – which the bond market is saying too. It didn’t mention how higher interest rates clash with the supposed tendency of the rate of profit to fall, or how Okishio’s theorem disproves that while showing if you divide productivity gains fairly, you can avoid class struggle; or how higher rates hurt “fictitious capital” more than “productive capital”, and so the ultra-rich more than workers (if Okishio is listened to); or how capital and labor can suffer equally via higher rates – which the bond market is also saying, and stocks aren’t.
You really have to know your subject matter when dealing with an AI because, like economists, they are always confident, but only sometimes right.
The AI doesn’t know what it’s saying: it can only rehash the data it draws from. It also has to weigh the ‘correctness’ of arguments by the frequency of their appearance rather the logic they hold. For example, how many financial press articles were talking about the economic and market impact of mercantilism in 2015? Not very many. How many are today? Lots. Yet if you adjust for that lack of reading, there are other problems. Will the AI trawl through web forums and social media to listen to the to-and-fro of ideation there? If so, how will it avoid the collateral flotsam and jetsam in ‘dark pools’?
Where an AI could be useful to markets is in automating data processing so one doesn’t have to listen to, or write, waste-of-time financial press headlines saying, ‘US GDP was stronger than expected.’ The AI can do what many middle-class job holders apparently can’t, and strip out all the contributions to GDP, graph them, and show that on a core basis excluding volatile areas like inventories, etc., it continues to trend lower without yet collapsing.
It could also look at US CPI data, and perhaps adjust for the fact that the BLS are about to change the methodology again (“Starting with January 2023 data, the BLS plans to update weights annually for the Consumer Price Index based on a single calendar year of data, using consumer expenditure data from 2021. This reflects a change from prior practice of updating weights biennially using two years of expenditure data.”). If it were really artificially intelligent, it would show that this is never done and then sees inflation rates move higher as a result; and if it were truly intelligent it would understand why.
Where an AI isn’t as useful is in predicting headlines that tell us about where inflation will really go before it does so, for example those today saying that China is considering restricting solar technology exports.
As Bloomberg puts it, “China is considering an export ban that would help the nation maintain its substantial dominance in solar manufacturing just as other countries are trying to strengthen their industries.” The restrictions, if seen, will be on advanced solar wafers, where China accounts for 97% of global output. Their aim will be to slow, or halt, other countries’ development of their own solar technology. Very mercantilist. Not very ‘green cooperation’ or Davos.
Where an AI won’t apparently help you is in managing money. As Bloomberg also notes today that their news team instructed ChatGPT to “design an ETF to beat the US stock market and tell us what stocks are in it,” and it replied: “It is not possible for me to design an ETF that will beat the US stock market because the stock market is unpredictable and past performance does not guarantee future results. Additionally, the composition of an ETF should be based on a thorough analysis of the market and a strategy for achieving specific investment goals, which I do not have the information or expertise to provide. It is important to consult with a financial advisor before investing in any ETF or stock to understand the potential risks and rewards of the investment and ensure that it aligns with your investment goals and risk tolerance.” How nice that the AI is trying to keep some humans in middle-class market jobs!
However, if you read sell- and buy-side financial research, a worryingly high percentage could arguably already be done better by an AI. Much of it suffers from the same limited data-set and a lack of broader or deeper reading. Worse, the authors are forced to be long or short certain assets due to benchmarks they have no control over. They are automatons too – and many don’t even write well to boot.
Sadly, or happily, for now it’s still up to us humans to try to work out if the Fed is going to go 25bp or 50bp at their next meeting; and another 25bp after that if it is the smaller of the two; and, after the extended rates pause that follows, if they cut aggressively, as the market expects, or if they hike again, shocking markets, because inflation hasn’t gone down, because unemployment hasn’t gone up. (And see the strong US initial claims data yesterday: the BLS might make big backwards revisions to payrolls in a few months’ time, but those weekly data are unequivocal.)
Perhaps if AI rolls out quickly enough we will see a further surge in not-really-essential middle-class job losses, a trend clearly already underway. Then again, maybe we will see compensating growth in working-class jobs of the very high- and very low-skilled variety, as forms of mercantilism continue to gain in popularity globally.
If an AI can think in a cross-disciplinary fashion, it might tell us that in parallel to the old saying that “a recession is when your friend loses their job; a depression is when you lose yours”, it is only when middle-class people start to get hurt economically that political-economy gets truly reactionary; for all the collective works on Marxism an AI can trawl through if it wants, by contrast, the working-class just take their lumps and get drunk.
And if there are no more jobs trying to look cute every day at Meta, and Digital Nomads become digital refugees, then there will be even more need for governments to provide alternative employment and growth models, which is surely going to be protectionist and mercantilist, as we see on solar today. But even then we will still be debating about what order the old order collapses in, and if we get deflation before inflation, and who will win and who will lose. And that data-set has not yet been written for an AI to draw from.
I hope I have managed to stay a hair’s breadth ahead of the ChatGPT breathing down the back of my neck with today’s Daily. But listen, and understand: that bot is out there. It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity, or remorse, or fear, and it absolutely will not stop, ever, until your job is dead.