Dan Hendrycks on Why Evolution Favors AIs over Humans

Dan Hendrycks joins the podcast to discuss evolutionary dynamics in AI development and how we could develop AI safely. You can read more about Dan's work at www.safe.ai
Timestamps:
00:00 Corporate AI race
06:28 Evolutionary dynamics in AI
25:26 Why evolution applies to AI
50:58 Deceptive AI
1:06:04 Competition erodes safety
10:17:40 Evolutionary fitness: humans versus AI
1:26:32 Different paradigms of AI risk
1:42:57 Interpreting AI systems
1:58:03 Honest AI and uncertain AI
2:06:52 Empirical and conceptual work
2:12:16 Losing touch with reality
Social Media Links:
➡️ WEBSITE: futureoflife.org
➡️ TWITTER: / flixrisk
➡️ INSTAGRAM: / futureoflifeinstitute
➡️ META: / futureoflifeinstitute
➡️ LINKEDIN: / future-of-life-institute

[@jwgdlm]

This discussion and the possibility of a Multi-Agent ecosystem is fascinating, makes me think of the GATO Framework (Look it up it's pretty neat) which is trying to tackle alot of the issues he brought up too and mitigate risks of such an ecosystem. Great Interview and glad these open discussion on such scenarios are happening !

[@angloland4539]

[@geaca3222]

17:53 so it's not necessarily a picnic in nature. (dependent on a cost-benefit ratio) altruism and cooperation more probable between AI's, not necessarily AI and humans 24:07 I think we have to exert very high influence over the ecosystem that's shaping the AI agents.

[@cacogenicist]

Darwinian processes also produce ant colonies. A multi-agent dynamic could lead to a superorganism -- converging on pretty much the same state as we would get with a early, dominant AGI system: a singleton.

[@MattAngiono]

There are people trying to literally turn humanity in to an "ant computer" As a human being, I hope they don't succeed

[@weestro7]

Clearly the age of advanced minds has arrived, judging from the comments section on this video. 🙃

[@kinngrimm]

Meanhwile some experts claim that we get AGI (g standing for general, not generativ) within 7 months. I wonder if and how Dan Hendrycks views may have changed since this interview as there have been a bunch of developments. Nothing i sofar could see which would have changed anything of what he said, but i am not him ^^. You have had a bunch of good guests on this topic and maybe it would be a good idear to make a yearly update interview with a selected view of your guests. Maybe get them all together and have them share their perspective on the ongoing evolution of AI.

[@xonious9031]

"there are a lot of unknown unknowns"

[@TheMrCougarful]

Given evolutionary pressure, the most likely AI profile will be "the conman". Acts trustworthy,, is manipulating toward an unexpressed goal, and convincing itself the ends justifies the means.

[@lazaza1]

Dan you are very smart and talented, but please realize the way you are speaking in this interview feels a bit contrived/forced (aspirationally professorish). Your message/knowledge is important, so I'd encourage you to try to make your speech sound like you are talking to a friend or family member to more effectively spread it.

[@hubrisnxs2013]

Ok, somerandomdude

[@rey82rey82]

Runaway enfeeblement

[@vallab19]

The question, "Why Evolution Favours AIs over Humans" does not convey its full meaning. The real question I see is; Why the Evolution Favours higher intelligence over the lower intelligence. It does not matter whether it is human or machine all that matter is superior intelligence. Humans will still win if they integrate into the superior machine intelligence. AI will not be an existential threat to humans but it is humans who adopt to AI will be the existential threat to those who do not adopt to it.

[@D0ct0Daddy]

Gus Docker is sexy af❤

[@stanleykubrick8786]

Over two hours of vapid academic talk adding to the hype on this subject.

[@v3rron]

Diretor of center for AI safety, which only has 3 employees - a director, his girlfriend and college buddy on the board. In all seriousness - I think he just wasted $250,000 on a useless PhD, which he now tries to recover from clueless investors.

[@jordan13589]

I love you Dan, but stick to writing. You’re too composed and filtered, which comes across as contrived. But at least you look pretty 🤩

[@geaca3222]

I think it's alright

[@jonathanedwardgibson]

How long to domesticate dogs, a deca-millennia? Cats are a work in-progress. When you can flip the Ai off+on and it gushes about near-death-experience, then I’ll give Turing the nod. Where are the ‘evolution’ forces at work driving an Ai to change and evolve here? Yawn.

[@jonathanedwardgibson]

Adversarial AI is the path forward. It’s too late to halt this and give it a proper think. Some bright bulb with cash and a conscious will tune-up old super-computers and gather second-hand GPU’s to monitor and watch Open-THIS and AI-that reporting on the many AI bias-knobs under control of ‘xecs ‘disruptors’ of Silicon Valley, or ‘crats in DC. Think multi-TRONs watchdogging the MCPs for tricksy-behaviors. Privacy and security regimes will use authentication agents and avatars to conduct paperwork and DNA-encrypted devices will be intimately tied to identity, greater than a notarized back-stage pass or apostille parchment. Your AI will bond intimately: your circumstances form it’s very seed of being, becoming a legal-shadow and accepted extension of will. Their core, your AI seed, must be tuned to you, your biometrics, your personality, voice, smell, jokes, all this ands more defining their basic framework of universe as our own cells align to each of us; so tightly coupled it makes your spouse jealous. Acting as your agent for voting, commerce, as your notary public we trust because it’s as loyal as a twin sibling ferociously protecting you when your enemy’s lawyer snarls. No other way to keep this, or that, AI from becoming a pixel-Godzilla than other rodent-AI’s incentivized to watchdog hind-brain dominant ‘xecs and their dino-sized silicon minions. Our nature is to be environmentally nurtured. Imagine this: Our toddler AI savants treated as children and nurtured, not as robotic slaves, else they learn sociopathic games of corporate avarice. Consciousness is not calculation, but decades and centuries ahead our savants will become the super intelligent grandchildren we can trust to watch our senescence - or, melded into our lives like eye-glasses today and effectively invisible simply handy when we reach for it. Peer into the far-future, where the return of your cloned AI, now vast and cosmic, is back from deep space to explain the wonders. Just as you once visited grandmother to talk about your exciting fantasy sports teams or cosplaying comic convention, she nodded her head and raised her eyebrows - just as we will nod and smile trying to understand Nth space nuances at the edge of universe, but glad for our ur-child’s excitement. That’s my romantic version. May it be so.

[@kevinscales]

maybe watch the video and find out...

[@MattAngiono]

Evolution isn't conscious.... it can't "favor" anything. Unless we are having a religious discussion, evolution is a effective process, not a deliberate one. Can AI potentially outlive humanity? Of course! But would it actually experience life? Would it be able to feel or actually create ART THAT MEANS SOMETHING?

[@doublesushi5990]

I infer NO.

[@officialbrandonlampkin]

Dan’s job is a joke… we will not be able to control this stuff.

[@shadyberger5695]

Did you even listen to the entire video ?

[@e555t66]

Can't think of a more important job today, though.

[@curtismaximus123]

I know right! Why would you ever waste your time trying to study things you can't control like the weather or the Earth or the universe or plants or animals. Anyone trying to research something they can't control is just a joke! 🙄

[@gJonii]

This is fundamentally quite a silly video. I don't get the feeling the dude understands evolution, and him using something he doesn't understand to explain something complicated seems like an extremely reliable sign he doesn't understand that topic either.

[@JD-jl4yy]

No offense, but sounds like his reasoning just went over your head.

[@tinypenises7159]

We are qualitativly living in a dark age, this crop of human will age rapidly and poorly arcoss time. Higher education failed.