Hai Stanford (Human Centric AI) was started in 2019 with support of Stanford President ndleladership of Fei-Fe Li and John Etchemendy, professor of philosophy and former Stanford University provost
HAI Founding Distinguished Fellows
The inaugural group of Distinguished Fellows will include: Yoshua Bengio, University of Montreal; Rodney Brooks, MIT; Erik Brynjolfsson, MIT; Jeff Dean, Google; Daniel Dennett, Tufts University; Susan Dumais, Microsoft Research; Edward Feigenbaum, Stanford University; Barbara Grosz, Harvard; Demis Hassabis, DeepMind; Geoff Hinton, University of Toronto; Eric Horvitz, Microsoft Research; James Manyika, McKinsey & Company; John Markoff, Center for Advanced Study in the Behavioral Sciences; Helen Nissenbaum, Cornell Tech; Judea Pearl, UCLA; Stuart Russell, UC Berkeley; Mustafa Suleyman, DeepMind; Terry Winograd, Stanford University; and Hal Varian, Google.
tSanford President Marc Tessier-Lavigne said artificial intelligence has the potential to radically change how we live our lives. “Now is our opportunity to shape that future by putting humanists and social scientists alongside people who are developing artificial intelligence,” he said. “This approach aligns with Stanford’s founding purpose to produce knowledge for the betterment of humanity. I am deeply thankful to our supporters who are providing foundational funding for the institute, which is a critical element for our vision for the future of Stanford University.”
Stanford HAI formally launches at a symposium on Monday, March 18 featuring speakers such as Microsoft founder and philanthropist Bill Gates and California Governor Gavin Newsom, as well as leading experts Kate Crawford of NYU, Jeff Dean of Google, Demis Hassabis of DeepMind, Alison Gopnik of UC Berkeley, Reid Hoffman of Greylock Partners and Eric Horvitz of Microsoft Research. (Watch the livestream here.)
2023- Hai Stanford is celebrating leas forward from covid era with extraordinary talks like this
We
are reducing this 90 minute youtube and Transcript to key cases and methods
mentioned by fei-fei li and demis hassibis -of course you are referred to the
90 minute version at youtube
I am Fei-Fe
Li HAI stanford ie lab human-centered AI
Welcome Demis
hassibis founder DeepMind 2010- now an alphabet/google companu
Demis
superpiwers began in games, and doctirate in cognitive neuroscience -brain
architecture - at MIT (check him out as alumni of Uni College London, MIT and Harvard so this is how he came to machine intel – DEEPMIND 2010 truly one of
first new wave AI companies aquired google 2014 but retaining its sphere og influence out of London. (indeed this may be Demis first sam francisco region
visit since covid even though he was on HAI stanfrd founding council 2019
deepmind commercial
breakthroughs include center energy and mapping the world’s proteins through
the very impressive AlphaFold AI stream work
a world leader in realms of deep learning and
reinforcement learning and this reflects gaming roots -eg AlphaGo world
champion
there are
hopes that DL will become climate AI leader – invite student Q&A on that at
end indeed Demis is in midst of pioneering Artificial General iItelligence as
potentially Epoch Changing Tech changing the very fabric of human lives
AI is having
a pubcli Awakening moment and it's no longer just a niche field that
nerds like us
play around ; it's impacting human life society and our future and
Hai Stanford
has been missioned to be one of the forums that will host this kind of
3:49
intellectual
discourse about Ai - I cant think of
more timely speaker than Demis
========================as at 5/3 rest undrr construction
Demis –
thanks so much Fei-Fei – yes now covid’s hoefully passing I hope to be in bay
area more often and share real passion
of mine which is to use AI to accelerate
4:57
scientific discovery
with generative Ai and and large language models and and the work we're doing
deepmind was
founded way back in 2010 that was almost like medieval times
5:45
in 2010 it
was very difficult to o raise our seed round of you know a few hundred thousand
dollars compared with today’s billion dollar rounds - in 2010 few were talking about AI Shae Legg DM’s chief scientst
annd I felt very isdolated – people seemed to have forgotten AI as field to build a human-like intelligence so
it's been astounding what the last few months have brought as wwe see
comvergence of a lot of different
DM has always
been big proponent of reinforcement learning and understanding
7 the human
brain - my PhD I worked on the hippocampus and memory systems and Imagination
made some interesting discoveries in in that domain that I thought would also
potentially carry over into ideas for AI systems architects together with QC
Advent of lot of compute power and specifically GPUs (Graphic Processing Units)
which ironically of course were invented for games so everything for me as
you'll see comes back to games one way or another um
We began DM
as king of Apollo Program of Games
First we
revited earliest game systems Atari space invaders and 50 games from the 1970s -we uh and maximized the score jaround n the
raw pixels on the screen so it was very much um probably the first example of a
kind of end-to-end learning system on something that working on something
really challenging perceptually
9:09
challenging lan
Atari game swas a incredible moment for us and I remember when back in 2011
when we were struggling to even win a single point at a game like pong and we
was jwondering well maybe we're just 20 years too early with these ideas of of
learning systems and then suddenly it won a point then it won a game and then
it didn't lose any points; by 2013 it was playing all the Atari games of course
we then took that much further and and
However our
big opportunity came with Alphago -
world champion of game like GO
the super
complex game that's played in Asia
obviously
famously in 2016 we had this massive million dollar challenge match in Seoul
200 million people watched thematch around the world an alphago famously won
that match for one but more important than it winning move 37 which blew away
all past human players startegies
Today we can
design game winners one thing that's holding us back is that we don't really
know how to ask aplha to design a great game not in a way that it could
understand
As yet we havent found way to ask alpha to invent
a game that only takes five minutes to learn but many lifetimes to master is aesthetically
beautiful can be completed in in 10 hours of you know play so it fits into a
human day
Thesek are inds
of things is what I would kind of give as instructions and then I'd hoped it
would come up with something like go but there's no real way to do tgis yet
so how does
the self-play system work and I'm just going to combine together um actually a
range of systems together
Back to what
we can do16:10
Face-Off
match a hundred game match of old play -version V1 versus new veriosn V2 and threshold
in our case we set a 55 win rate threshold where if V2 beats V1 by
above that
threshold you assume that it is significantly uh better and then you
replace the
Master System that the generator system with that new V2 system now and you go
around of course
16:35
iterating
this round so now you could play another 100 000 games with V2 so it's slightly
stronger so that means the
16:41
generated
data is slightly better quality -continue to generate more data another hundred
thousand games with V2 so then now you have 200 000 games uh to train a
118:09
general way
of thinking about Ai and the and the um
the idea of coming up with a solution to a problem
18:18
se've been
very fortunate over the last decade we've been part of um kind of creating many
big breakthroughs um inall sorts of different games all of them kind of
landmark uh results at the Atari one the alphago one I mentioned Alpha
18:31
zero I just
talked about generalizing that to every two-player game and then finally Alpha
star which was our program
18:38
to uh beat
Grand Masters players at Starcraft 2 which is the most complex
18:43
real-time
strategy game computer game and it has extra challenges over board games of
being partially observable uh
18:51
it needs
things like long-term planning so it's complex in in in in more challenging
ways than a board game and
18:58
so this was
all of our work in in games now of course although I love games always have
done playing them designing
19:04
them using
them for training for AI I've sort of used games in every way possible but
they've always although it's been
19:11
very fun to
do that um it's always been a kind of means to an end not an end in itself
right the end was never to just win it go or Win
19:19
It win at
Starcraft it was to build it was to use games as a convenient proxy
19:24
to test out
our algorithmic ideas so that we could apply them to important real world
problems