Programming note: I’m back from paternity leave so regular cadence will resume. One slight change in content - I’ll still be covering AI and it’s impact on the economy, but I’m going to do so from a more explicitly European perspective.
Hank Green Got AI Cancelled
In July we had the All Ireland Gaelic Football final here in Ireland. As often happens after a final, clips of the game make it outside Ireland to people completely unfamiliar with the sport, who are surprised to see 80,000 fans in a stadium roaring for a game the outsider never knew existed.
My favourite reaction this year was a tweet that simply said:
There really is a whole other world out there.
I have shared this sentiment twice recently when it comes to AI. My own world is one of tokenmaxxing, trying to figure out if there’s a bubble, and analysing the benefits and limitations of a profoundly transformative new technology.
But there are entire information ecosystems in which AI is a completely different thing.
My first real exposure to this world was at a talk at the Dalkey Book Festival a few weeks ago with the provocative title “Is AI Theft?”. Expecting a detailed discussion about copyright, I was pretty stunned when two of the four panelists (on a discussion about AI, I’ll remind you), proudly announced that they never use AI.
One panelist, Merve Emre, Professor of Creative Writing and Criticism at Wesleyan University, even shared that many of her students would no longer be friends with someone if they discovered that the friend used ChatGPT.
Wild stuff!
This same world-view was on the rampage again last week with the near-cancelling of Hank Green.
If you don’t know him, Green is a science educator who makes very popular YouTube videos, mostly educational, always thoughtful, well researched and informative.
Last week he generated a storm in his part of the internet when he admitted to using ChatGPT for a video. No - he didn’t use it to write the script or to generate the video - he just used it for research. He didn’t even copy from the ChatGPT research notes into the script - it was all original writing.
I’m often wary of over-estimating the size of an online storm, of joining what Bo Burnham called “the backlash to the backlash to the thing that’s just begun”, or as NYU Prof. Ruby Thelot cautions us to ask of any online trend “is this a real thing, or is this three TikToks and a trench coat?”
But the reaction to his admission seems large and genuine, and was enough that Hank Green himself saw fit to issue a lengthy apology and self-reflection, and pause some of his content production.
There really is a whole other world out there!
If you look past all the doomerism and Butlerian Jihad there is, I think , one interesting tidbit in his apology which is worth dwelling on. Green says:
mostly I need to come to terms with the fact that the level of dopamine I've been getting from interacting with LLMs...with doing more and more and more and more...is not healthy for me
I think there’s definitely something to this sentiment. I think it’s related to the phenomenon of tokenmaxxing and the confusingly low correlation between increased AI use and increased productivity.
I think LLMs have removed the friction that was inherent in starting something new and, until now, we didn’t realise how useful that friction could be.
Personally, I have found that I unknowingly relied on this friction as a forcing function to push me to prioritise, to ask if this is the most important thing I need to do right now. Or indeed if it really needs to be done at all?
LLMs and Agents reduce the psychological cost of execution so much that you can now get into execution mode on any task the second it pops into your brain. There is no requirement for planning or consideration. There is no psychological pain to getting started, but there is all the dopamine rush inherent in getting into something new.
But, of course, the LLMs and Agents are not good enough to see the tasks all the way through. If you’re like me, they can get you a good 80% of the way there, but then you end up with a string of projects needing the final 20%.
And this 20% is the mentally taxing stuff. The feedback, the judgement, the planning you should have done upfront.
Or, the agent can complete the task 100%, but it’s a menial admin task that you probably should have just deprioritised in the first place.
All this is to say that I feel for Hank Green on multiple levels. Both for enduring the wrath of an online mob but also for feeling overwhelmed by the elixir of a tool that promises infinite productivity by removing frictions that we hadn’t realised were actually load-bearing (to borrow a phrase).
"I think LLMs have removed the friction that was inherent in starting something new and, until now, we didn’t realise how useful that friction could be"
Incredibly on point.