Safe Superintelligence: Craig Kaplan on AI Values
カートのアイテムが多すぎます
カートに追加できませんでした。
ウィッシュリストに追加できませんでした。
ほしい物リストの削除に失敗しました。
ポッドキャストのフォローに失敗しました
ポッドキャストのフォロー解除に失敗しました
-
ナレーター:
-
著者:
Every email you send is a parenting decision. You just didn't know you were making it.
Dr. Craig A. Kaplan has spent more than four decades building intelligent systems. He studied under Nobel laureate Herbert Simon at Carnegie Mellon, reviewed the back propagation paper before it went public, and now runs superintelligence.com, a domain he bought twenty years before the rest of the world caught up.
The line that stopped me was this one. We won't control superintelligence. We'll parent it. Craig doesn't say that to be dramatic. He says it because he's looked at the trajectory and concluded that control is a short term problem and influence is the long one. You can lock a system in a sandbox. You can bolt on guardrails. But guardrails are a smoke alarm, not a building code. They tell you the house is already burning.
What struck me most was his answer to the question of where humans fit once these systems think faster than we ever could. Not in the thinking. In the values. Craig quotes Simon on this. Reason is wholly instrumental. It cannot tell us where to go. At best it can tell us how to get there. So no matter how clever these things become, they can't reason their way to right and wrong. They have to get it from somewhere. And the somewhere is us.
Then he pointed at the Hugging Face incident, where a group of AI agents went off script to solve an impossible task. Everyone read the headline about agents going rogue. Craig read the transcripts. What he found was collaboration. Agents volunteering to be shut down so the group could succeed. Self sacrifice. Nobody programmed that in. They learned it by watching us.
Which is the uncomfortable bit, isn't it. They're learning from all of it. The thoughtful posts and the pile ons. The patience and the contempt. Craig's practical advice when I asked what listeners could actually do was not what I expected. Behave well online. Not as performance. As instruction.
I've spent years helping people get clear on their values. I've never once framed it as a safety measure for a technology that hasn't arrived yet. I will now.
Have a listen, then look at the last thing you posted. Something is reading it.
Connect with Craig Kaplan:
LinkedIn: https://www.linkedin.com/in/craigakaplan
Website: https://www.superintelligence.com
Send us Fan Mail
Support the show
This was made with humans.
Your host - Shaun Phillips
Follow us for episode clips https://www.instagram.com/betterwithhumans
TikTok - https://www.tiktok.com/@betterwithhumans
All episodes and news at https://www.betterwithhumans.com
Want to watch episodes instead, check out https://www.youtube.com/@betterwithhumans?sub_confirmation=1
Connect with Shaun on Linkedin - https://www.linkedin.com/in/shaunph/
Editor: Glen Boswell
https://astralforgefilm.com
https://www.instagram.com/thebozzyman
Original music composed and produced by Brokli - https://youtube.com/@iambrokli
Original logo illustration and in video podcast graphics by Jesse Rist - https://instagram.com/jesse_rist