Preparing for my AI Doom Debate with Robin Hanson
HTML-код
- Опубликовано: 23 июл 2024
- This episode is a comprehensive preparation session for my upcoming debate on AI doom with the legendary Robin Hanson.
Robin’s P(doom) is less than 1% while mine is 50%. How do we reconcile this?
I’ve researched past debates, blogs, tweets, and scholarly discussions related to AI doom, and plan to focus our debate on the cruxes of disagreement between Robin’s position and my own Eliezer Yudkowsky-like position.
Key topics include the probability of humanity’s extinction due to uncontrollable AGI, alignment strategies, AI capabilities and timelines, the impact of AI advancements, and various predictions made by Hanson.
good luck my friend. i'm expecing this to be the best robin hanson doom debate. you're a good communicator and well prepared
Thanks!
Its amazing how much better the show looks with the relatively simple changes you've made to camera, composition and wearing a shirt.
Its like immediately it seems like a serious show and with a robin hanson debate up with these ducks in a row youll find it much easier to get more guests you want moving forward.
Also love how much due dilligence you put in ahead of the debate. That hard work speaks v well of you
Thanks, appreciate the feedback :) There’s lighting and stuff too, it was basically 5 days of total full time work tweaking the setup - share.icloud.com/photos/044X5TzLxD9uqrIPXDNLc2Huw
I’m willing to put in the work because the world needs high-quality Doom Debates to happen!
Woah! New title screen, set, and lighting is great! (Mic position at 7:37 is the best visual composition to my eyes). More importantly, though, thank you for putting the time into prepping for this; you're a service to humanity. I'll be excited to hear Hanson's take on Bostrom's "The Vulnerable World Hypothesis". What a fantastic (and frightening) paper! How do you stop ASI from pulling a superior-to-human life form out of the urn?
Lastly, as p(doom) disagreements often hinge on differing definitions of doom; I suggest to make sure you and Hanson are on the same page before venturing too far. Some think that Earth being turned to computronium would not count as doom so long as their brain gets sampled for a couple milliseconds along the way (Kurzweil?). Others think life-as-is is doom because humans aren't immortal. What counts as doom?
Best luck!
Thanks! Ya I think that mic position is the winner.
The debate happened today, dropping on Monday, and I think I did take into account what you mentioned.
@@DoomDebates Looking forward to it!!!
I'll train up some popcorn for this.
Looking forward to this one. I’ve always thought that ‘doom’ relies on a couple of things: capability and agency. The models are becoming more and more capable, but I have not yet seen an example of a model acting according to its own impulses. Yes they can be agentic in response to tasks we give them, but I have not seen any impulse on their part to act without prompting.
Of course that doesn’t guarantee AI safety. The first time a model decides to act on its own may be sufficient to cause ‘doom’.
I saw that you announced the interview will drop on Monday. Can't wait for that. In the meantime I will listen to this video to prepare for it.
I have a small complaint, if I may: the camera shakes a bit when you touch the table and for me personally is so annoying that I can't watch at all. This is the reason I haven't watch Battlestar Galactica, even though I am I huge SF fan, because they chose to slightly move the camera in almost every shot and I cannot watch that.
But I do love the higher quality of the camera and the sound. Future episodes will be lit!
@@MirceaBurdusa Thanks for the feedback, I’ll work on that
That's gonna be nuts.
Not related to the debate but I think you're a bit unfair towards evolution in the way you call it a bad designer. The difference between living beings and evolution designing things is that evolution only has a small amount of output data it can work with. Assuming I'm right it seems like it's only able to tell if a design choice is successful or not by how often on average a species is able to survive in an environment. Somehow it has direct access to information of the objective reality and it uses that to create our interface (sense perceptions). So when humans design something we are being given tools that skip over this direct information needed to gain feedback about success. It's like we're given a desktop computer to make programs and we don't have to know the inner workings of what's really going on we have high level tools that let us simply figure out what works and what doesn't work quickly. On the other hand evolution only seems to get a pass or fail and that's all it gets along with being capped at only making changes over millions of years. My point is that what it's doing is completely impossible for us because we don't perceive objective reality and we owe it to this force that designed our senses. It also has to deal with fitness pay offs and ecosystems of how the whole picture fits together. When we design something like a plane we don't have to worry about whether our new changes will affect other creatures (even though we probably should). In that way, it has to be careful not to make an overpowered species that will dominate all other life. We may be a failing of that since we exceeded everything else that has been tried. Also, when it pushes out a patch it doesn't have the luxury of just directly eliminating a species it has to deal with the consequences.
@@goodleshoesFair
dude your camera setup is 10x better
Amazing what getting roasted by RUclips commenters can do 😂
@@DoomDebates 😅haha yup
In 2 years I'm pretty sure ai won't be superhuman at math. It might discover some new stuff due to having a lot of data, but it won't be a better problem solver than Terence Tao. When it gets there, it's pretty much game over.
Please tell me you already watch Roman Yampolskiy vs Robin. I found it to be the one debate in which Robin performed the weakest. I am honestly pretty surprised he posted the debate.
@@aubreyblackburn506 I watched it ya, don’t remember it being that brutal but of course I agreed with Roman
Ai has not limits … in time it will be a god!
wasnt the debate going to be today?
The debate happened, it's recorded! The podcast episode is dropping on Monday.
@@DoomDebates #weekendearlydrop #please
@@consumidorbrasileiro222 Here's how I think about this:
1. Weekdays are better for a big episode drop since more people are around on social media to help spread it around
2. I want to make a more polished product with a couple days for light editing
3. Since I expect this to be a popular episode, I want to insert an intro beforehand where I plug a couple things like the pauseai.info organization
At some point I'm strongly considering making a premium subscriber tier, something like $10/month via my Substack (or Patreon if ppl prefer), where you could get access to "superfan" content like an early release of a rougher edit. What do you think?
Regardless, the vast majority of the content will always be free since the goal of Doom Debates is to raise public awareness.
@@DoomDebates 👏👏
I hope he's got 4 hours because you want to talk about A LOT
@@davidmjacobson Lol yup. I had Robin scheduled for 2 hours and we got through about 2/3 of my outline, didn’t hit everything I mentioned here. But I’m pretty satisfied I got to focus on the questions that were most salient to me about Robin’s worldview.