Demis Hassabis: creativity and AI - The Rothschild Foundation Lecture
ฝัง
- เผยแพร่เมื่อ 11 ก.ย. 2024
- Recorded at the Royal Academy of Arts on 17 September 2018:
Demis Hassabis, Co-Founder and CEO of DeepMind, draws upon his eclectic experiences as an Artificial Intelligence (AI) researcher, neuroscientist and videogame designer to discuss the implications of cutting-edge AI research for creativity and scientific discovery.
DeepMind is the world leader in AI research and develops programs to help tackle some of the world’s most pressing challenges. Since its founding in London in 2010, DeepMind has published over 170 peer-reviewed papers, five of them in the scientific journal Nature - an unprecedented achievement for a computer science lab. In 2014, DeepMind was acquired by Google in their largest ever European acquisition, and is now part of Alphabet.
Demis Hassabis is a former child chess prodigy who, following graduation from Cambridge University, founded the pioneering video games company Elixir Studios, producing award-winning games for global publishers such Vivendi Universal. After a decade of experience leading successful technology startups, Hassabis returned to academia to complete a PhD in cognitive neuroscience at University College London, followed by postdocs at MIT and Harvard, before founding DeepMind. The journal Science listed his neuroscience research connecting memory with imagination as one of 2007’s top ten scientific breakthroughs. Demis is also a five-time World Games Champion, recipient of the Royal Society’s Mullard Award, and a Fellow of the Royal Society of Arts and the Royal Academy of Engineering, winning the Academy’s Silver Medal. In 2016, Demis received WIRED magazine’s Leadership in Innovation Award and in 2017 he was named in the Time 100 list of the world’s most influential people. In 2018, Demis was awarded a CBE for services to science and technology, elected as a Fellow of the Royal Society and also awarded an honorary doctorate by Imperial College London.
The talk is followed by a conversation and Q&A with the RA’s Artistic Director, Tim Marlow.
This Rothschild Foundation Lecture is the first in a series of annual lectures generously supported by the Rothschild Foundation.
FOLLOW US:
News and blog: www.royalacade...
Facebook: / royalacademy
Twitter: / royalacademy
Instagram: / royalacademyarts
This is perhaps one of the most important topics of the century, and yet this video only has a few thousand views.
To be fair there are lots of well viewed videos on this topic (including a version of this talk by Hassabis on TH-cam with a hundred thousand views watch?v=0X-NdPtFKq0).
I haven't noticed but you are right: there are not so many views. People news are far more popular I guess, which is sad in my opinion haha.
He's not the best speaker or conversationalist. Garry Kasparov's AI talks have a lot more views, because he's much more famous and better at explaining the implications of the work done by people like Hassabis.
@@squamish4244
Same with Musk. He's not even an AI "expert", but he's a good speaker and very engaging.
So true and for me this video simplified most of Artificial intelligence in terms of machine learning. Compared to many other videos which are long and boring.
As always simplicity is the ultimate sophistication!
Definitively you should bring Demis Hassabis for a series of lectures, not only this topic is very interesting but he explained everything really well given the complexity of this field.
Thank you very much, I enjoyed every minute 🙂
Glad to see Demis articulate the idea of accelerating scientific discovery. I've been pushing this for a while. I use the example of the gap between Hero's aeolipile [Hero of Alexandria ~ 50 AD], and the steam engine invention [Newcomen~1712]. at least 1600 year gap. World could have been millennia ahead by now. Time to take on the task accelerating discovery and kick it up a notch!
54:30 the way the guy takes Hassabis's glas ;-D
Excited to see RL applied in optimization problems in the coming future. Go DeepMind!
Demis Hassabis qualifies for a clutch of Nobel Prizes on the basis of this lecture alone. See and hear a genius who is Hippocrates, Da Vinci, Einstein and humanity itself, rolled into one.
47:09 definition of intelligence
This is really cutting edge in terms of human knowledge. His implementation of a system that can imagine a scene and then move around that scene is close to what is needed to model imagination. I feel that it needs to move one step forward. Not just creating a static scene(probably constrained by the use of a neural network), but rather to create a simulation of a scene. And then actioning things in that scene, that would be real (AI) imagination - real artificial creativity.
Such a free and creative person
Thanks for sharing! Great video!
Thank you ! Demis is the Albert Einstein of our century.
Eh
He hasn't actually changed our understanding of the universe or upturned an existing paradigm. All of what he does falls within known physical laws.
He's intelligently refined one form of AI. I would compare him more with many of the physicists who followed Einstein and the other great physicists of his era and built upon their work, if we're going to stick with the metaphor.
Stop
he runs the company, most of the intellectual work is not done by him, although he certainly likes to take the credit. sure he is brilliant, but likely many at deepmind are moreso
Great lecture, kind of worrying that the analysis tools to understand AI are at best 5 years behind "black boxes" AI (1:08:00). Specially considering the exponential nature of AI development for example AlphaGo was decribed in this same lecture as 10 year ahead of predicted time...
Excellent video.
Thank you for sharing, where can i download the PPT?
Focus on creating a simulated reality to train artificial intelligence in. Simulated bodies and terrains. Train it to cope with all environments and all body shapes. Like each leg joint could have a weight or pressure. Maybe you need to work with rockstar games to make a nice small training environment for a.I. You should be able to change terrain settings, senarios, body shapes. That would be faster than Boston Dynamics.
This is indeed something people are working on. If you train an algorithm on lots of variations of a simulation, hopefully it could see reality as just another variation.
Talk starts at 1:17.
Does AI have to learn the concept of dimension and space? As humans do we intrinsically have this understanding because we ourselves are dimensional? (if that makes sense)
Yes, if we want it to solve problems involving positions/movement through spacetime it has to know something about spacetime, however it is interesting that the AI that already solve these kinds of problems never have to be explicitly taught anything about spacetime.
The knowledge is already encoded in the environment and neural networks 'simply' encode the relationship between what is observed (in the environment) and the problem they are solving, which means it naturally encodes knowledge/intuition/concepts/abstractions (or whatever you want to call it) about spacetime when the problems/goals are about relationships in spacetime and spacetime is part of the environment.
As for why we have the intuitions that we do about spacetime, it's kind of similar but a much more complex story of evolution. We have to solve problems involving relationships of objects in spacetime and have for billions of years since life started becoming capable of noticing that some positions in spacetime are different from others in important (in terms of the problem/goal of survival) ways. So that has been encoded into the DNA that creates our brains, just in a more robust and generally useful way than it is built into any current AI.
The audio is not so smooth.
This is pseudoscience.