If one had to characterise the state of mind among knowledge and creative workers, musicians included, in 2022, worries about meeting demand for quantity or variety of content would not figure high. Misinformation, provenance, and the difficulties of making a living, yes. Being short a few million blog posts, product reviews, or new music tracks, no.
So it was perhaps a bit surprising when the appearance of a flock of AI engines, and their ubermensch ChatGPT, an easy to use interface to a new text AI, generated (pun intended) wild enthusiasm, along with some doomsterism, and of course digital reams of new content as the commentariat contemplated their own industry’s potential demise. Perhaps the boosters thought that with AI doing the creative stuff they would have more time for the real work; self-promotion.
There were some practical responses too. Stack Overflow, the techie Q&A site, moved quickly last month to ban answers generated by pasting questions into ChatGPT. Their reasoning? “…while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce”. New York City schools followed with another swift ban, adding concern for safety and appropriateness.
Behind ChatGPT is a language model developed by OpenAI. It can output school essays, reviews on Amazon, answers on Stack Overflow, even rather lame sonnets (try it!). Other AI models produce images, music, and video, churning out in seconds what would take humans hours or days.
AI output is being widely monetised in copyright markets intended to nurture human understanding, research skills, and creativity, over which it has unfair cost and quantity advantages. Here’s California’s Boomy Corporation, an AI music generator, boasting about its productivity: “Boomy users have created 10,900,994 songs, around 10.8% of the world’s recorded music.” Of course the bit about Boomy users creating songs is a lie – the users filled in a short form and tapped some buttons.
Anything that can be consumed as news, opinion, entertainment, or art, can be over-produced by these generative AI systems, making them part of what I term ‘hypercompetition’. This is a seemingly ineluctable rule of the classical platform business model, which bundles distribution with curation, and follows up with interventions to increase supply. The aim is to heighten competition with and between human practitioners in the monthly zero sum subscription share-out games.
AI could make light of much drudgework and formal, logical writing, freeing time for more human creativity. It will almost certainly accelerate knowledge extraction and cataloguing in some specialist fields. But almost all ChatGPT output is spam, low quality stuff we really don’t need more of. However I’d argue we should not push string by trying to ban AI. Instead we should treat it like the pollution it is, and slap a large Pigouvian tax on it.
Some activities create costs that are borne by people who don’t share in the gains of those undertaking them. Congested roads, smoke from factory chimneys, and unusual demands on publicly funded health services are examples. Economists think of these as costs external to the economic activity associated with producing things, driving somewhere, or smoking tobacco. There are rare cases of external gains too; think of how good state-funded schools inflate property prices in their catchment areas.
The external costs from ChatGPT are the teachers’ time and effort, and pupils’ lack of learning caused by cheating at school, incorrect Stack Overflow advice followed, as well as revenue extracted from copyright markets and the extra efforts human creators need to put in to compete. Arthur Pigou himself, Cambridge Professor of Economics in the first half of the C20th, was deeply interested in welfare. He argued that tax was the simplest and most effective way to deal with gains and losses that fell to those outside of a transaction.
As well as being a redress for harmful activity, Pigouvian taxes can provide funding for more of what we do want. This is important – interventions need to have broad popular understanding and support. In this case the harms fall on education and the arts, two significant absorbers of taxpayers money which could be supplemented or offset.
I asked ChatGPT if it thought it should be taxed; it denied any knowledge of itself. It might not yet write a decent poem, but AI has clearly got its head around tax evasion.