Using AI to Accelerate Scientific Discovery | Campus Lecture with Demis Hassabis

Demis Hassabis, Founder and CEO of DeepMind, discusses the development of AI, from self-learning systems like AlphaGo, the first program to beat the world champion at the complex game of Go, to AlphaFold, a solution to the 50-year grand challenge of protein structure prediction.
During the event, he receives a Doctorate Honoris Causa from EPFL.
#EPFL #ai #artificialintelligence #google

[@Martin-jp8pw]

Demis starts speaking at 7:28

[@noahhall1502]

Demis is a legend

[@ravd8082]

demisis and deepmind started the ai revolution

[@rufaiguruboy7677]

No one can disagree

[@mistycloud4455]

ai will be like the Gutenberg press, we will Accelerate Scientific Discovery exponentially

[@LilaSilk]

Great presentation, merci for sharing.

[@micbab-vg2mu]

Great talk! - Thank you.

[@MG-fr3tn]

When you fold a protein the device folding it exerts some torson,then you have hydrogen bonds or proximity of water molecules to amino acids. But most of it should be amino a joining amino b bends at such and such a angle. Given the diagrams they have now you must've solved it.

[@gabrielstahl5629]

Fantastic talk!

[@Myrslokstok]

Should be noble prize!

[@squamish4244]

I think he'll win one as soon as AlphaFold discovers a drug that works. So far it has discovered one drug in record time, but we don't know if the drug works yet. It will no doubt discover many more soon, as that drug was discovered early this year and AlphaFold can get a lot more sophisticated - along with everything else in the drug discovery pipeline - but as soon as they find something that works, then I can see Demis and his team winning the Nobel Prize.

[@spagetti6670]

HERO!

[@hanskraut2018]

Woow finally switzerland doing SOMETHING in AI and machine lerning and not just biology. I want to do AI and machine lerning stuff and not be bored to death. I wan to see swizerland flurish. Many many good things here. Also i hope elitism wont be a problem but so far seems not bad. Lovely video thanks a lot. The introduction seemed quite thoughtful im plesently mildly suprised. Demis is awesome gona be a interssting talk.

[@nb6808]

eth is literally top 5 in the world in computer science XD

[@squamish4244]

Switzerland was already world-class in terms of AI and machine learning research, especially considering its size. Combining what it's doing now with its historical expertise in biology is a big effing deal, because that means that AI and machine learning are finally doing something that actually matters for our quality of life, in Switzerland and in many countries.

[@famillepol342]

Amazing

[@spagetti6670]

DEMIS HASSABIS+++

[@geoattoronto]

Intuition is my word for the output of the human spirit that perceives at a spiritual level and accesses wisdom from above and beyond the human senses. Our spirit can access wisdom from good (godly) sources or bad (evil) sources and which one makes a fundamental difference in a person’s character.

[@Myrslokstok]

37'th 6-line move, they kind of explained tht it has been an human error and missunderstandmentbetween human an AI because the move was crazy and the computer couldn't be this bad.

[@Nova-Rift]

I love Jennifer Doudna!

[@blaeks]

boom!

[@geoattoronto]

The process of imagination creates a link to the highest, other-dimensional capabilities of the human. Tesla had an amazing imagination that allowed him to create to the point that he was a threat to the capitalists. He wanted ‘free energy’ for everyone.

[@kirkgilbreath5094]

AlphaFold has NOT solved the protein folding problem: AlphaFold 2's results at CASP were described as "astounding"[6] and "transformational."[7] Some researchers noted that the accuracy is not high enough for a third of its predictions, and that it does not reveal the mechanism or rules of protein folding for the protein folding problem to be considered solved.[8][9] Nevertheless, there has been widespread respect for the technical achievement.

[@squamish4244]

It hasn't solved it, but experts were still stunned. Three years ago, people spent years modelling ONE protein. It triggered serious competition from Meta, for instance, whose ESMFold is not as accurate but 60 times faster. The team that developed it has been disbanded, but the software is open source, and obviously other companies are also working on protein folding. Everything Demis discussed about AlphaFold being used by a million scientists has happened in the last two years, since it became open source in July 2021. Right now DeepMind is working on rebuilding and retraining AlphaFold. Idk, I just get mildly irritated when people make statements that don't see the forest for the trees. AlphaFold and everything else that is happening with drug discovery is finally here after a long time coming but someone just HAD to clarify that AlphaFold hasn't completely solved the protein folding problem yet when DeepMind was founded all of 13 years ago and the software that would lead to AlphaFold only began to reach maturity all of eight years ago with AlphaGo.

[@devilsolution9781]

Quotes from where?

[@kirkgilbreath5094]

@@devilsolution9781 Read the sources given in Wikipedia entry for AlphaFold in the 4th paragraph.

[@MrEmbrance]

Starts at 7:29

[@darlenewaldron3621]

AI is often surrounded by hype. It seems to have periodic breakthroughs followed by periods of stagnation.

[@squamish4244]

AI went through two 'AI winters' after it started to develop about 50 years ago. This time, however, there are no foreseeable obstacles and the amount of knowledge and resources at our disposal is much larger than before. Unless a bunch of things all go wrong at the same time, it's hard to see us stagnating this time.

[@darlenewaldron3621]

@@squamish4244Achieving Artificial General Intelligence (AGI), or machines that can perform any intellectual task that a human being can, presents multiple challenges. Here's a summary of some of the main obstacles: (1.) Complexity of Human Intelligence: Human intelligence isn't just about solving mathematical problems or playing games. It encompasses emotion, intuition, creativity, general world knowledge, self-awareness, and more. Emulating such a wide range of capabilities in a machine is a complex task. (2.) Computational Challenges: Even with today's most advanced supercomputers, the computational power may not be sufficient to replicate the intricacies and efficiencies of the human brain, especially when it comes to tasks like understanding natural language or learning from limited data. (3.) Lack of Understanding of the Brain: While we have made significant progress in neuroscience, we still don't fully understand how the brain works. Many aspects of consciousness, cognition, and subjective experience remain mysterious. (4.) Data Challenges: Training machine learning models, especially deep learning models, requires vast amounts of data. The right kind of data might not always be available, or it may be challenging to acquire. These challenges are substantial, so the timeline to AGI remains uncertain, with predictions ranging from a few decades to perhaps a century or more, or even some believing it may never be fully realized.

[@squamish4244]

@@darlenewaldron3621 I am aware of the technical challenges. However, predictions by most experts have drastically shortened since the recent breakthroughs in deep learning. The median used to be 2040-2050 and now, experts say they just don't know and many do say it could be five years or less. Demis is as much an expert as anyone and he's saying a few years. What reason does he have to hype anything? He's already a multimillionaire. Today, expert predictions are actually the same as asking random people on the street. Comparing computing to the human brain is now a good way to gauge AI. It's why we were so wrong about the timeline regarding LLMs and AlphaGo. Max Tegmark says the mistake we made is in thinking computing has to work like the brain in order to be AGI. It doesn't. He compared it to how we were able to achieve human flight with steel and heavy engines 120 years before we were able to replicate how birds fly. We were able to leapfrog biological evolution with the use of inorganic materials. Similarly, computers operate 10,000 times faster than human brains do, so even though they are far less efficient they are still capable of many things we cannot do. Combining those capabilities is easier than we thought, at least up until now. They are working hard on the problem of accessing more data. There is still a huge amount of non-text-based data to pull from - like KZbin?!? And then there's synthetic data, where computers start making their own data. And _then_ we can make AI way, way more efficient with the loads of data we already have. Only the most pessimistic predictions put AGI out a century or say it may never be fully realized. That is utterly unrealistic. AI would literally have to suddenly go wrong in everything. Also, we lack a true definition of AGI. We don't really know how to define it and whether we will have achieved it or not until it is quite possibly far along. If you're dealing with a superior form of intelligence, how do you necessarily know? It's a distinction humans have never faced before. We'll see what DeepMind is able to achieve with Gemini, due for release later this year or early next year. Regardless, AI does not have to be true AGI to utterly transform the world. What Demis lays out here does not need to be AGI to change everything about everything. So obviously my assessment of the state of AI is not nearly as pessimistic as yours.

[@arthurrobey4945]

I see a need for a machine psychiatrist. Ref: Asimov.

[@geoattoronto]

I am working on a ai Christian therapist.

[@goldnutter412]

Games and reality are analogous, the brain is a layer for what we don't use to play this game. The main computer, a distributed non physical system. 10:45 metaphorically this is free will "The problem is choice" .. no it is the solution ! choice leads to evolution of our personal information system. By extension, our shared information system evolves. Entropy is relentless for obvious reasons.

[@jorgedavidtarrillo8554]

☕

[@ili626]

AI for science makes sense, but using it to vacuum up countless artists’ unique creative styles etc is a social and cultural disaster for humanity

[@squamish4244]

I can see people still wanting art because it's produced by humans, in the same way that there's been a movement back towards as many practical effects and locations as possible in movies and not just having them be CGI orgies with five people in a blue room. Or like there seems to be little interest in 'AI girlfriends/boyfriends' like in Her beyond the socially very maladjusted. If anything, even in our hyperconnected social media world people crave in-person contact more than ever and the online space has become a means to facilitate meeting real live people instead of just talking to them online. The anger at how badly a lot of dating apps are still designed and how the designers keep things crappy on purpose in order to keep people swiping so they can make more money is one reflection of this.

[@Gingnose]

That's what exactly artist said when camera was invented back then.

[@alanrobison4761]

53:38 - Exactly. Can't make any promises. Even this guy know it's mostly HYPE.😁

[@squamish4244]

No scientist can ever make any promises until something actually works. Nobody knew the steam engine, cars, planes, electricity, radios, television, computers etc. were going to work until they actually worked. Or more topically for this year, the nuclear bomb. Oppenheimer et. al. didn't know it was going to work until it actually _worked._ So was all of that hype?

[@devilsolution9781]

Are you over 60?

[@alanrobison4761]

@@devilsolution9781 What's the point??

[@devilsolution9781]

@@alanrobison4761 you seem like a pessimistic old person

[@devilsolution9781]

@@alanrobison4761 so, am i right?

[@bld691]

🐇

[@MG-fr3tn]

Cleaver people might have nothing to do but crosswords and games.

[@justice929]

Cleaver people have an insatiable appetite for learning. My uncle who Lectured at MIT and Caltech all he did was learn reading books, after he retired worked as a night watchman so he could read all night...

[@deeplearningpartnership]

Haha

[@hanskraut2018]

?

[@thembelssengwayo6896]

I am NOT really a fan of Elon Musk but I should admit that Elon Musk is NOT just superior but Elon Musk is mostly NOT a l human being. When Elon Musk is speaking he always say:" Mhu...Mhum....Ham".....for the major part of his speech and it is very difficult to follow Elon Musk speech because of that, why is he doing that in his speech. Elon Musk's brain is almost thinking beyond his speech and he is getting billions every second that he is able to speak, that is number one...... Number 2, Elon Musk is always asking questions than just speaking......and always know the answers to those questions and people don't like to be asked questions. So in essence Elon Musk is the most odd and hard nut to crack than almost all geniuses and progidies of the world combined......Hate him or love him, it does not matter.

[@alanrobison4761]

No AI has solved any significant problem faced by humans like cure for terrible diseases like cancer and Parkinson's, not solved global warming problem, or anything for that matter. It's just HYPE.

[@somrajdutta8366]

You missed the plastic eating enzymes work underway using Deepmind

[@alanrobison4761]

@@somrajdutta8366 The development of plastic-eating enzymes is still in its early stages. However, just like other technologies such as Thalidomide, Artificial blood substitutes, DDT, Fast Breeder Reactors, Ethanol from Corn, and Carbon Capture and Storage (CCS), it will fail to deliver anything meaningful.

[@somrajdutta8366]

@@alanrobison4761 Point taken. We still have to stay hopeful for a better future for ourselves and our children. Edison failed 10000 times to invent the incandescent filament - it changed the world . Mankind progresses one failure at a time, don't you think?

[@alanrobison4761]

@@somrajdutta8366 It's more complicated than that. There's substantial evidence suggesting that human intellect has its limits, and there's a growing body of evidence indicating that science is becoming unbelievably complex. Nowadays, the faith people place in science is beginning to resemble the faith one might have in a religion.

[@somrajdutta8366]

@@alanrobison4761 interesting thought. As an Indian growing up with Vedantic thoughts I do agree that science is a product of the mind. And the mind, being a feminine energy, can only complicate in its machinations. It is in her nature to entangle further and further into gross materia nature. But if we leave aside the so called path to 'liberation/ emancipation/ moksha' or whatnot, for the moment, what we will be left with is man trying to imitate God in this material world. Like a child trying to imitate it's Father. You are right in the essence that perhaps this savage progress is not in Man's best interest. But that is only to the point of man's path to realise himself independent of his material existence. However that apart to live in this material world is to try to live better. Man is forever trying to survive and reproduce better more intelligently - and this is where the problem lies : his misplaced ideals away from acknowledging the existence of a power beyond material sciences. Unfortunately we are now too far displaced to even go back. For now Man must never cease from exploration. For at the end of all exploring is to arrive where he'd begun and know the place for the first time. Thank you for your thoughts.