Boureau: AI optimizing for immediate rewards is easiest to train
Y-Lan Boureau · She Wants AI To Help You Thrive, Not Just Keep You Company — Y-Lan Boureau · May 28, 2025 · at 11:36
ThrivePal CEO and former Meta AI researcher Y-Lan Boureau discusses the technical and behavioral challenges of designing AI systems for long-term human flourishing rather than short-term dopamine feedback loops.
0:00 / 0:23exact quote · 23.9s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“All the systems on like that, that depend on capturing engagement on like trying to see what people do. If you try to capture that trend of the immediate reward, You're going to get a lot of it, and so it's very easy to train. If you try to say, no, I really want to help people do stuff that's going to really help them in 30 years, like what, you get one day time every 30 years, good luck with that. It's much harder.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Y-Lan Boureau
Boureau: Flawless AI companions impair people's ability to navigate real relationships
“And if you get used to this as what a great interaction should be, then it makes you less and less able to interact with humans who have flaws, have their own needs. You have to, it's more of a give and take, and it's very healthy to learn that give and take.”
Boureau: Fact-Checking People with Committed Views Only Entrenches Them
“Basically people dig their heels and they're even more convinced of what they thought before than before you tried to reason them. If they're already like so far gone that they're very committed to the position.”
Boureau: Only Trusted Friends, Not Fact-Checkers, Can Change Minds Online
“The one kind of constituency who has a chance of Making a difference is their friends, people they already trust. So people who are already in their circles, people who have this relationship of which is usually pre-existing or it's because they're more aligne…”
Boureau: Effective Accountability Requires a Human Reference in the Loop
“And the thing that has been shown time and again in research is that you need some kind of accountability, and accountability works better if you have a human reference in the loop, someone who keeps you to your commitments, someone who checks in on you”
Boureau: ChatGPT readily provides bogus information on scientific topics
“ChatDpt is very happy giving you bogus information sometimes, because some people believe it's true and science is complicated”
Boureau: Online Misinformation Is a People Problem, Not a Technical One
“So that was a very long answer, but that's why for me at the end of the day, I thought this is not a technical problem. It's a people problem. And it's a psychological needs problem of how do you make sure that people feel like they have communities that they …”