Demis Hassabis: Towards General Artificial Intelligence – talk at Center for Brains, Minds and Machines (CBMM). [Background: r. Demis Hassabis is the Co-Founder and CEO of DeepMind, the world’s leading General Artificial Intelligence (AI) company, which was acquired by Google in 2014 in their largest ever European acquisition.
The talk draws on Demis’ eclectic experiences as an AI researcher, neuroscientist and video games designer.
One synthesis of Stuart Russel’s reflections on Deep Learning:
“ Some deep learning networks get up to one thousand layers or more … The deep learning hypothesis suggests that many layers make it easier for the learning algorithm to find a predictor, to set all the connection strengths in the network so that it does a good job. … to a large extent, it’s still a kind of magic, because it really didn’t have to happen that way. why this happens is still anyone’s guess.”
in Marty Ford’s new book: principal architects of AI.
Russell Stuart is lead co-author of THE must have AI textbook
this textbook is the main AI textbook used in over 1000 Universities and Colleges. If he says it’s a kind of magic … it is a kind of magic! So if you want to be one of the contemporary magicians or wizards … get on with the deep learning wands.
BONUS … Queen’s kind of magic
To place Artificial Intelligence in appropriate context is a complex and intricate challenge. Marty Ford, a master explainer presents interviews with some of the principal architects of AI. The Architects in this case are Yoshua Bengio, Stuart Russell, Geoffrey Hinton, Nick Bostrom, Yann LeCun, Fei-Fei Li, Demis Hassabis, Andrew Ng, Rana El Kaliouby, Ray Kurzweil, Daniela Rus, James Manyika, Gary Marcus, Barbara Grosz, Judea Pearl, Jeffrey Dean, Daphne Koller, David, Ferrucci, Rodney Brooks, Cynthia Breazeal, Joshua Tenenbaum, Oren Etzioni, and Bryan Johnson.
What a great list … Dan Ferrucci is of course known from his amazing work with IBM WATSON, and the first ever amazing win of a machine over the best of the best at Jeopardy! Dennis Hassabis , of Google/Alphabet’s Deep Mind, brought us AlphaGo, AlphaGoZero, and now AlphaGo that exceeds the best of the best in Chess, GO and Shogi (all with the same MCTS algorithm). NYU/FAIR/FaceBook’s Yan LeCun did some serious stuff with Mastering and claiming ‘the prize’ over ImageNet Challenge. Rodney Brooks with iRobot, … each one of the Architects is truly a master architect. We’ll explore their contributions and significance later … their thoughts are really worth checking out. I am looking at all kinds of things right now … and there’s just so much. Maybe I need a nice Intelligent Machine Assistant to help me pull this al together. 🙂
Ford, M. (2018), Architects of Intelligence,
“The occupational activities of children are learning, thinking, playing and the like. Yet we tell them nothing about those things.” per AI Pioneer Seymour Papert – In Pam McCorduck’s Machines who Think, (an outstanding book; Pam is a great author, turns out she’s the wife of Joe Traub who was Computer Science Dept Chair at Carnegie Mellon University & Columbia University … and had amazing insight into the real story 🙂 – not found elsewhere ) https://amzn.to/2FwGmIu
EXCELLENT EXCELLENT BOOK … It’s really packed with amazing insights and details hidden from the public view …
I didn’t realize Papert’s connection with Piaget and his deep understanding and interest in how children learn. Of course Papert and Minsky’s Perceptrons were widely known [ and got a refresh boost . The Perceptron. ideas… which, in prehistoric times, with Marvin Minsky, helped pave the way to the AI we know today. — that’s where the real action was and maybe still is … check the reboot. over at https://amzn.to/2TNjok7
One of the main categories of discussion in this book is that of worthwhile tasks for AI. I will devote some time to stating some of the recognized questions, problems, and tasks. I will also mention some notable AI accomplishments and highlight a few of the recognized scholarly achievements. Another topic for discussion is the classification of Intelligences. What is Natural Intelligence? What is Artificial General Intelligence? What is Superintelligence? What about human measures such as IQ? G? What does the AlphaZero algorithm beating the best human players in Chess, Go and Shogi mean? Can the Paperclip Apocalypse really happen?
All these and more … coming soon …
OK, so I started perusing Terry Sejnowski’s recent book, The Deep Learning Revolution. It’s dedicated to Bo and Sol, Theresa, and Joseph and is In memory of Solomon Golomb. Nice!
- It’s a great book. In the short time I spend with it, I learned quite a lot. I decided to see what’s most important to Terry looking at the topics he spends most of his time on. Here’s what pops out first …neural networks and deep learning . [To be expected], then the items getting most discussion are:
- the brain
- machine learning
- learning algorithm
- artificial intelligence
- the world
- visual cortex
- the network
- boltzmann machine
- the cortex
- Geoffrey Hinton [looks like Geoff is really getting attention and kudos from everyone!!]
- network models
- the future
- self driving car
- learning networks
- cost function
- deep learning networks
- hopfield net
- primary visual cortex
- the visual cortex
- independent component analysis
- real world
- the internet
- the perceptron
- facial expressions
- reinforcement learning
- Francis Crick
- hidden units
- the retina
- information processing systems
- neural information processing
- neural information processing systems
- td gammon
- the boltzmann machine
- computer vision
- driving cars
- simple cells
- the hopfield net
- cerebral cortex
- David Hubel
Somewhere further down the list I came across Soumith Chintala over at FaceBook AI / Courant Institute. His was a new name for me. Looks like he’s a PyTorch maven, super-coder. Nice! his Wasserstein Generative Adversarial Network (GAN) paper seems pretty nice. Apparently FAIR has advanced the ball a lot with Generative Adversarial Networks. I need to be paying much more attention. Also noted a new name to follow, Cade Metz who writes about technology for The New York Times/
All this from my first glance at The Deep Learning Revolution.
read it … I will get deeper into the deep learning as well.
Happy Holidays …