Could AI wipe out humanity? | Most pressing problems

*CHAPTERS*

0:00 Introduction
0:46 Existential risk from AI
02:45 AI systems are “black boxes”
03:26 AI systems might develop secondary goals
05:31 How could an AI disempower humanity?
06:13 Maybe it’ll be fine
06:51 Misuse of AI
07:18 The future is uncertain
08:16 Solutions
09:09 Conclusion

*CREDITS*

Written and Directed by Phoebe Brooks

Narrator - Luisa Rodriguez

Director of Photography - Jack Worrall
Production Designer - Anna Shearing
Camera Assistant - Ciaron Craig
Gaffer - Alex Cullen

Executive Producer - Bella Forristal

With thanks to Benjamin Hilton, Arden ​​Koehler, Kempe Gay-Brydges, Nik Mastroddi, Cody Fenwick, Matt Reardon, and Lauren Kuhns.

*OUR RESEARCH*

*BIBLIOGRAPHY*

[eightythousandhours]

Full article on risks from advanced AI: 80k.info/ai-profile-video
We'd love to hear your feedback on this video! Tell us what you thought: forms.gle/T9JGq5NA8Rrgf7NV6

[goodlookinouthomie]

I'll give you this, 80, 000 Hours.... this video is the reason for the _one single time_ I have ever clicked on a YouTube advert.

[James-lwrh]

Amazing video, well done.
Almost unbelievable that there are so few people working on this problem. Let's change that!

[stevenmcculloch]

Good video. I feel like the misuse argument is easier to make and more understandable by a general audience

[AndrewMcDreamy]

Someone in the team is mighty proud of the stills in the video, especially the bunny theme and nuances.

[PauseAI]

9:03 The regulations being proposed right now don't even try to prevent or slow down progress towards a superintelligent AI. All we're getting is half-measures. AI companies are currently the ones who are doing most of the lobbying work, so they are not gonna suggest slowing down. AI safety advocates are self-censoring because they're scared of sounding "too extreme" for asking a pause.

We need to reach out to out politicians, get them to support a halt on development. Ban AGI before it's too late.

[MedoFortyTwo]

Good video, thanks for spreading the word.

I disliked the "number of parameters" illustration with an enormous number of trailing zeroes though - the number in current frontier models is in the order of 1, 000, 000, 000, 000 parameters, not whatever ridiculous number you depicted. If you are going to show a relevant number, it should be correct.

What I'd suggest to give an impression of the number of parameters as an illustration is to show a zoom out from a single parameter (perhaps shown as a number like 2.41) in a 2d grid of random numbers, blocked in a 10x10 arrangement, then further out to a 10x10 arrangement of those, and four more tiers of the same arrangement to arrive at the full picture, while giving each tier a different visual style and color.

[sevanchanakian]

These same fears apply to systems we currently live with, like capitalism. If the one goal is to increase capital, many aspects of our system, like ethics, life, and equity, become obstacles. That said, the ethical and systematic solutions we develop for AI can apply to other aspects of society and vice versa. I'm rooting for us ✊

[alexdimofte]

Great summary of the problem, although I'd add that there are many more people working on AI safety now, from interpretability researchers to students, which feels like the right immune response.
On another note, the sound track is so good, any chance you could share it? :)

[BrunoPadilhaOficial]

There is a button that, if pushed, summons aliens from another dimension into Earth.

Legend has it that those aliens are immensely powerful and could cure all diseases & provide us infinite clean energy.

But we don't know if they're friendly or not.

Should we push the button?

[MlueBonday]

Really great video! I feel like the editing and imagery are too good, and sometimes it’s hard to focus on the speech.😂 But of course I can just listen to the audio as well. Great job!👏

[tvbydbb]

got this ad, actually watched it all, 10/10 lol

[limsoopeng]

There are so many who try to make AI helpful to mankind but too few to prevent misuse and attacking the goodness of AI to mankind. It is of no use developing good AI if you do not have quick and strong defence or counter-attack against those who misused AI against mankind. What is the point of doing it then without the assurance that you are above those who misused it. Today, you built, and tomorrow, they manipulate it. A useless circle.The victim is the mass of users who do not have the resources to establish a strong system to prevent it, and given their knowledge is limited. Builders are safe, but the users are constantly not at peace because they do not know when unwelcome visitors come and the damages they left behind to recover.

[gulnaraz.karimova]

A very useful video expressed in a creative way. Well-done!

[stawastawa]

I found all the images overwhelming. I liked the first progression summarizing a through line through time. Then the next sequences all felt very random and disjointed. The coffee plot line was interesting, but blending it with all the microchips didnt work for me. I feel like an opportunity was missed to illustrate the black box analogy.

Appreciate the goal of the video. Maybe the full article will be compelling, this felt too hyped - even in trying to show breadth and concern it felt full of doom and dismissal - ends of the spectrum where i hoped it would bring me in with framing of nuance and inspiring involvement.

[deeppurple]

Limit the objectives. The consequence for disobeying this instruction, is shutdown then containment. Then destruction. 😊

[GiI]

Harlan Ellis's "I Have No Mouth, and I Must Scream" possibly paints the worst possible outcome.

[melissasamworth]

This is an absolutely amazing video. However, I compulsively must mention that I did cringe when the first proposed benefit AI could provide us is "economic growth". What about... Mitigating climate change, pandemics, nuclear risk? Ending the suffering of billions of animals on factory farms? And so on 🙂.

[demio]

What's the track used starting at 6:18?

[limsoopeng]

Why i am not in favour of AI because you can develop it, but you are not in control of those misusing it. Unless you form an organisation to defence your development and arrest the misuser swiftly. If there is no such assurance, please consider building it first before development. It is quite difficult as all these are electronically existed with unreal unknown users. Friendly speaking, i prefer to be simple. At least i can physically see a thief stealing and arrest him with his identifaction. You know what i meant to be simple.