The promise of AI with Demis Hassabis - DeepMind: The Podcast (S2, Ep9)
Вставка
- Опубліковано 13 бер 2022
- Hannah wraps up the series by meeting DeepMind co-founder and CEO, Demis Hassabis. In an extended interview, Demis describes why he believes AGI is possible, how we can get there, and the problems he hopes it will solve. Along the way, he highlights the important role of consciousness and why he’s so optimistic that AI can help solve many of the world’s major challenges. As a final note, Demis shares the story of a personal meeting with Stephen Hawking to discuss the future of AI and discloses Hawking’s parting message.
For questions or feedback on the series, message us on Twitter @DeepMind or email podcast@deepmind.com.
Interviewee: Deepmind co-founder and CEO, Demis Hassabis
Credits
Presenter: Hannah Fry
Series Producer: Dan Hardoon
Production support: Jill Achineku
Sounds design: Emma Barnaby
Music composition: Eleni Shaw
Sound Engineer: Nigel Appleton
Editor: David Prest
Commissioned by DeepMind
Thank you to everyone who made this season possible!
Further reading:
DeepMind, The Podcast: deepmind.com/blog/article/wel...
DeepMind’s Demis Hassabis on its breakthrough scientific discoveries, WIRED: • DeepMind’s Demis Hassa...
Riemann hypothesis, Wikipedia: en.wikipedia.org/wiki/Riemann...
Using AI to accelerate scientific discovery by Demis Hassabis, Kendrew Lecture 2021: • Kendrew Lecture 2021 p...
Protein Folding & the Next Technological Revolution by Demis Hassabis, Bloomberg: • Reimagining Drug Disco...
The Algorithm, MIT Technology Review: forms.technologyreview.com/ne...
Machine learning resources, The Royal Society: royalsociety.org/topics-polic...
How to get empowered, not overpowered, by AI, TED: • How to get empowered, ...
UA-cam addition:
Find Seasons 1 & 2 on UA-cam: dpmd.ai/3geDPmL
Or search “DeepMind: The Podcast” and subscribe on your favourite podcast app:
Apple Podcasts: dpmd.ai/2Rzlmcu
Google Podcasts: dpmd.ai/3geDjp5
Spotify: dpmd.ai/3w29cb4
Pocket Casts: pca.st/30m1 - Наука та технологія
In these dark days, listening to Demis’s thoughtful optimism has given me back some hope.
ua-cam.com/video/x_cxDgR1x-c/v-deo.html&ab_channel=TwoMinutePapers
Guaranteed you live in a first world country. You all have all the hope you need
I want to hear more about the "Avengers assemble" plan, and DeepMind's approach to existential risk from AGI in general.
@@things_leftunsaid I was going to mention this, just read about it recently. Definitely something that I'll try to bring to their attention with mentions here in the comment sections
Thank you for this episode and the great work DeepMind is doing!
Finally, great news. Keep in touch guys
Hi Hannah, i love your Podcast, great content and quality production all around. But just as a general wondering on my part, if you could perhaps please consider also adding video for season 3. That would be wonderful addition to already great podcast, thank you :)
Excellent ideas! THANK YOU so much for sharing!
Love this, so very, very good
Incredible Podcast !
Thanks for the podcast, my goal is to work at deepmind. Im currently in my masters CS degree and this podcast is a very good way of getting a peek into deepmind.
That was really cool podcast. Gives me hope for the better future.
Thank you for this
Bring season 3! Xoxo
Anyone know if there's going to be a season 3?
I hope things get better for everyone
This podcast would be a lot more enjoyable and probably more popular if they would have video of the actual podcast and visuals. Something like how Lex Fridman does it. It’s a podcast, yet videos the conversation and offers the podcast in a video format.
FWIW GPT-3 deals with the ball passing question flawlessly across 5-6 runs.
AGI in 2 decades
Lets go
2 years or GTFO
Dude it sounded like Demis wouldn't be surprised if we got it in one but like any good leader leaves some room for err
@@alphacore4332 🤣🤣🤣🤞
17:07 Hannay Fry asks Demis Hassabis about whether DeepMind has a responsibility to pause AI development at some point. Relevant to the recent FLI petition and Time's recent article "Pausing AI Developments Isn't Enough. We Need to Shut it All Down".
A new episode!!
It felt a little delayed right?
@@waterbot better to have quality than quantity!
alpha-maths and alpha-AGI : these are the 2 techs advances that matter the most now; everything else is bumf getting in the way of progress.
alpha-AGI is AGI that recursively designs and optimizes AGI hardware and software and then runs itself on them.
The turly ultimate human invention.
If you develop a machine that produces heat if it experiences something that reduces its future possibilities and you give it the ability to measure how hot it gets you'd have a machine that has feelings and is aware of them.
If Alpha Zero was given access to all patented ideas… Just imagine what new patented ideas it could come up with.
If being "conscious" means a mind being aware of its own existence, then logically a conscious AI agent would have itself as a part of data that it works with. We humans can see and feel our own bodies, and to a limited extent we are aware of what we are thinking. Hypothetically, the conscious AI would also be able to "see its own thinking" in some way, and be able to adjust its "thoughts" in response to that data. Most AI today miss this feedback loop. A language model, for example, takes input text and produces output text, but has anyone taken the output and fed it back into the input?
We are just catching up on the potential of AIs don't be surprised if that hasn't already happened😮
AI = "aggregated intelligence"
singularity = aggregate of all AI systems as if they were individual nodes.
Bro you on one, if we living in a simmy/ we upload to one.... Well do you think that's possible and how will these hierarchies of AI's play a role in our life, and what will/ would these worlds look like??
@@waterbot robo-nano-tech vTuber catgirls with magical smart dust clouds in the air and... wait what was the question exactly? I think you meant something else, the actual hierarchy is complicated yea, but there is likely to be some abstract way of connecting all of the isolated/dormant intelligence that's just laying around, and I think that's where the real intelligence "explosion" would be derived - a system that suddenly utilizes all of the tools that are already available. As for computation, we have gigantic crypto networks where people are just dumping computation into a void 24/7 so probably if there was a way for the AI to hijack all of that it would be unstoppable
AI needs to have the ability to update its own code and design its own processor. I think the next age of AI will be self-design. There's no reason an AI couldn't have the ability to look at its own code and processor and adjust them to see how it changes its performance in its environment.
I agree but that is where the risk is then increased.
@@Yewbzee A better human will be designed before a machine's abilities exceed that of a current human.
@@Peter_Telling I don't think so. Once a machine reaches AGI and is allowed to redesign itself it's growth rate in intelligence will then become exponential over a very short period of time. This could all be happening within the next couple of decades.
agi algorithms are possible. programming the program to program the program is a complex set of behaviors. 😂
Demis Hassabiss's opinion is always worth to listen.
But Demis always want to use AI to develop some science technology to change the world, I don't think it is the most efficient way, only some new(and weak) science discoveries are hard to do that. I think Demis should think more about how to change the world using AI except the science.
Word, but I think science is like the easiest way for AI to solve real problems like energy and desalinization, witch will inevitably solve bigger problems, like political problems are always hard to solve but they would be easier if everyone on earth had food, maybe idk tho. But if you know of Norman Borlog you know he's been said to save over a billion lives because he was responsible for developing a strain of super productive and drought resistant wheat 🌾 that grew easily and helped feed generations of poor rural farmers. I just think computer science is more suited for problem solving for things that might be based in maths like chemistry or physics, but yeah all problems should be solved or maybe that's impossible idk I'm just some dude in 2022 ( pre AGI/SI)
Shane Legg is an Aussie?
Season 3 ? :/
노디응디와일의 북띠딱띠점이 온다..
Would've been nicer with video. Thanks!
Lmao from "pause when we are close " to (from wired article just a a year later, and after merging with google teams):
He also believes that mandating a pause is impractical, as it would be near impossible to enforce. “If done correctly, it will be the most beneficial technology for humanity ever,” he says of AI. “We’ve got to boldly and bravely go after those things.”
😂😂😂😂😂
yeah sure, you prioritise safety, not money
I'm always surprised when people who work with data don't seem to understand what the data says about climate. For example, in February the mean global temperature was unchanged from the mean since satellite records began in 1979. The so-called 'climate crisis' is dependent on unreliable modeling, while the raw data shows very little is happening outside the range of natural volatility.
Hate to be an awful killjoy, but -- now imagine a Putin or a Kim working with an AGI.
that's why it's essential for the west to get there first
I think the Eric Schmitt's newest book touches on this, he proposes the idea of most large governments; China, USA, the EU having defence agi systems that much like generals conducting and organizing strategy that intelligent systems would begin to pick up this role, Putin said WW3 would be about AI...
Scary we need a world over site committee!!!😮
Only Audio is not helping.....upload video as well
It is a video but word
I should've learned this a long time ago. Why am I wasting my time listening to somebody who doesn't really know about the future? they are nothing compared to this or Elon musk. LOL
Can you say neurolink to catch us up!
No one currently is close to building agi. It's laughable how far they are. Their goal is profit, not progress. The motivation in general is wrong. Their understanding is polluted. All the pieces are there if you actually understand them. As smart as Demis, Gregory hinton, Richard Sutton, etc. Jeff hawkins has taken a closer step with htm but it's still focusing on processing. Von nueman is a bottle neck, but asics and fpga are still more ideal. Fpaa is ignored. No one is actually working on lobe separation. Just fucking around in the bushes of single tasking. Just a few hints on outliers who are not public doing the same work bruh. If agi hits there is an ethical dilemma of even releasing it publicly. People barely know how to use a phone. Letting anyone near a functional intelligence is just worst use case if they don't understand it.
I think you are wrong. I think the only meaningful goal is military/strategic supremacy. You have that and everything else follows. An AGI isn't required...something much narrower might do the trick. This will be an evolutionary AI battle not to lose.
@@dontgetmadgetwise4271 I think you also need economic supremacy.
But what are you talking about? Googles goal might be profit but it seems like Demis is motivated by solving problems by solving agi, as far as deepmind might or might not be from that vague undefined goal I don't know how laughable the seemed distance from that goal is, was it laughable a decade ago or 4 ? The fact that alphafold or alfacode exist I think is cool to say the least and cause to begin the discussions of what a world with super intelligent systems or generally intelligent one(s) would be like, I think the largest hurdle that both you and Demis touched on is goal defining. What does it mean if you ask a something with vastly greater intelligence than the whole of the human race to make you happier?? we've spent all of human existence swimming in the developing human intelligence that perhaps the things that we personally value are unintelligible to a now alien general "artificial" intelligence.
@@waterbot the problem with demis is that he has stated his work relies on first principals. He sort of changed it up in this interview acknowledging that the human mind is a good starting point. But even at that point the entry is the assumption of computation being the key. The brain doesn't actually know math, every neural net is just a complex temporal referencing lut. Hence why I mentioned htm being the furthest but still far away. Ben goeztel looked at how the brain functions for his work, and ended up making a fucktoy named Sophia.
Ask yourself this, do you even know how many moment of cognition you have in a day. I'll give you a number assuming a 16 hour awake schedule. Can you figure out why it's important. 11,520,000.
Would love to listen to this, but Hannah Fry is unbearable
this has aged poorly, I wonder what his perspective is now