I am spending a lot of time at the moment trying to build resources for our ethics of behavioural science initiative and preparing for a number of events on climate change as part of my role on the Irish climate change council. As said in previous posts, I am still head of the Department at LSE and spend a lot of time across things like hiring and curriculum design. We recently completed a piece of work reviewing all of our assessments in light of developments in AI, which was interesting. I am mostly through the panic phases of large-scale LLM use. I will post at some stage with some thoughts on it when we have more time to separate out the different elements. My blogging brain might also slowly return and for now at least I can share some interesting readings.
I have enjoyed reading Cass Sunstein on these topics for a long-time and good to see Manipulation now out.
Thanks to new LSE PBS colleague Ben Tappin for pointing to a range of papers on the potential persuasiveness of LLMS - one key paper he has been involved with here
There are widespread fears that conversational AI could soon exert unprecedented influence over human beliefs. Here, in three large-scale experiments (N=76,977), we deployed 19 LLMs-including some post-trained explicitly for persuasion-to evaluate their persuasiveness on 707 political issues. We then checked the factual accuracy of 466,769 resulting LLM claims. Contrary to popular concerns, we show that the persuasive power of current and near-future AI is likely to stem more from post-training and prompting methods-which boosted persuasiveness by as much as 51% and 27% respectively-than from personalization or increasing model scale. We further show that these methods increased persuasion by exploiting LLMs' unique ability to rapidly access and strategically deploy information and that, strikingly, where they increased AI persuasiveness they also systematically decreased factual accuracy.
Faisal Naru's recent paper in Behavioural Public Policy updates and elucidates his widely circulated mapping of behavioural science units around the world. "Behavioral public policy bodies: New developments & lessons".
A recent paper in Behavioural Public Policy calling for Boosts over Nudges. Firstly, BPP has been an excellent journal (I am on the editorial board but I am not a key figure in driving it - credit mostly to Adam Oliver who has pushed it from inception). It has done a great job in covering all these debates. As students might testify, I am not a convert to the argument for Boosts in terms of scalable public policy (though the key papers are often stimulating and the overall ideas are fascinating to discuss).
As said on a previous post, a podcast I have been enjoying the last few weeks is Rational Reminder which does deep dives into a lot of behavioural personal finance. There are nearly 400 episodes and the quality is very high, including some great episodes on things like how Vanguard and Dimensional work, and a lot of episodes on investor behaviour.
No comments:
Post a Comment