Confabulation - I warned you

A while ago I started to worry about the direction that AI was taking, and that it could lead to machines making incomprehensible decisions that people would make up explanations for.

Then I realised that we’d start outsourcing these explanations to machines too, and sadly I was right.

http://t.co/5wruSm5zdK if Big Data creates AIs humans can’t understand, is confabulation the next thing AI researchers should work on?

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) March 25, 2013

@maximolly @Techsoc my prediction is that next frontier for AI is confabulation - making up plausible explanation for big data intuitions

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) April 20, 2013

@asaadhutchinson @pierre the next task for AI is creating plausible explanations for algorithmic decisions. Confabulation as a service.

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) March 6, 2014

my big fear is that we teach AI this kind of confabulation by accident.

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) December 21, 2016

“understandability, where a model can explain its decision and actions to users” - machine confabulation is coming https://t.co/oeYZbDU6OE

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) October 4, 2017

This is my deeper worry about AI - as we demand that the AI justify its decision to us, we create an incentive model for confabulation, just as we have for natural intelligence. Personalised reassuring salesmanship and gaslighting.

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) May 3, 2018

I'd argue that "confabulation" is closer to what's going on

— Kevin Marks 🏠kevinmarks.com xoxo.zone/@KevinMarks (@kevinmarks) January 6, 2024