Midwit AI

by Bronwyn Williams

“Evil comes from a failure to think. It defies thought for as soon as thought tries to engage itself with evil and examine the premises and principles from which it originates, it is frustrated because it finds nothing there. That is the banality of evil.” ~ Hannah Arendt

The problem with artificial intelligence today is not the problem we’ve been told it is. It’s not the exceptionality of AI but the averageness of it which we should fear.

In that regard, the advances of AI are not unlike the advances of democracy – the best of the bad ideas we’ve had so far to organise the progress of an animal so untamable and selfish as humankind. The AI tools we’re told to be terrified of taking our jobs and taking over the world today are like democracy in that they have been designed to optimise for consensus. Of course this consensus is in the bastardised cryptographic sense of the word whereby a 51% majority is taken as agreed by all, rather than in the true democratic sense of total actual agreement, an important distinction. Our not-so-very-friendly robot overlords have been trained to aim for the middle of the bell curve, the highest-lowest common denominator in order to tell us what is “right” rather than what is true.

Now the “right” or “best” answer, according to code, is either the most common answer or the most peer reviewed/“popular” choice in a world where truth is seldom popular. AI therefore follows similar logic to that of the popular vote in democracies which constantly seek popular validation rather than truth per se. This means that outputs are centred on the middles of the various bell curves involved; optimising for high probability, low variance outcomes. It also means that over time AI outcomes will perpetuate their own “popular consensus” algorithms creating self reinforcing infinite loops. These loops will be vulnerable to similar downsides to democracy, specifically an ever lowering of the lowest common denominator. Just as democracies tend towards populism, then towards socialism and on towards entropy popular consensus seeking algorithms will tend towards the average, and that same average over time will tend towards even more mediocrity and eventually entropy too.

The “great stagnation” phenomenon of our times, explained by Tyler Cowen in his book of the same title lamented how real progress had all but stalled even as so-called “progress” with financialisation and media, which distracts us from real problems, has accelerated. In other words the state whereby we have forgotten that velocity without direction is merely speed without movement. This has given us a preview of our ongoing acceleration towards average, the results of which we are able to observe around us every day. The future we are heading towards, is a future where the middle of all our markets become crowded to a point of singularity. This midwit singularity is of significant concern. The pull of the middle means the marketplaces for virtually all knowledge work (in as much as most of our very expensive and overpriced “knowledge work” economy requires very little knowledge and is not really work) are so competitive that it becomes perfectly priced. The result will be perfectly competitive, perfectly priced, perfectly in equilibrium markets, which as any good economist knows have no margin and no profits.

The middle which has hired knowledge workers forced through the sausage machine of white collar higher education, therefore, becomes an impossible place to earn an income as a human being. This the crux of the issue with AI. Algorithms will always beat human competitors at being average, at finding and perpetuating the popular consensus “perfect” outcome. Humans seeking to extract value, whether that be income or attention, from the markets will therefore be forced towards the edges of every bell curve. Forced to be more and more excellent or at the other extreme ever more wild, base and beast-like to differentiate themselves from the machine. Mediocrity is far easier automated, at scale, and far likelier to match a popular consensus seeking outcome than excellence or idiosyncrasy. The best performers in the last stages of the human dominated ideas market or knowledge economy, those “safe-bets” and reliable journeyman most beloved by corporate monoliths, will be the least valuable employees of the next stage of our automated economy.

Aiming for the edges becomes the only way to make an impression, and more importantly for those of us who live in a capitalist world, an income. Outcomes will of course adjust accordingly and we can expect extreme Pareto effects, with outliers taking all while the overcrowded, squeezed middle starves in its own homogenous mediocrity. All careers become rockstar careers. Those individuals who dare to live on the edges not yet colonised by the machine algorithms running just ahead of the advancing, all swallowing, application programming interface curve will do just fine. Likewise, weirdly, at the extreme bottom edges of the bell curve, where incompetence and idiocy lurks there will also be opportunities to make mistakes machines never would, and for populists and madmen to offer to humans the inconstant and emotive answers machines never will. High risk, high reward, winners and extreme losers take all.

Understand this and shiver at the banality of our artificially intelligently automated future as you look at your daily dose of perfectly boring computer generated pornography performed by voluptuously proportioned, Nordically homogenous avatar girls while using ChatGPT to write a 1000 word email for your boss who will then herself use ChatGPT to summarise your missive into a 100 character memo before having Alexa read it to her.