Dileep George's firm Vicarious, in which Facebook CEO Mark Zuckerberg and Hollywood filmstar Ashton Kutcher put in money, uses visual perception system to interpret the contents of photographs and videos in a manner similar to the human brain. The system claims to reliably solve modern Captchas, including Google's reCaptchas, the world's most widely used test of a machine's ability to act human. George shares advice for those setting out in the field in an interview with Krishna Bahirwani.
How much neuroscience is essential for a computer science/information technology student to pursue Artificial Intelligence? Can you make any recommendations?
What is important is to get a system level understanding of neuroscience rather than the minutiae of neuron function. There are many papers that attempt to do this. The one about 'Hierarchical Bayesian Inference in the Visual Cortex' by Lee and Mumford is a good overview paper. One could follow the references from their to build a more detailed picture. There are many popular science books that attempt to build a framework for understanding neuroscience data. 'On Intelligence' by Jeff Hawkins and 'How to build a Mind' by Ray Kurzweil. I would also recommend my PhD thesis and papers and references therein to get some of the relevant neuroscience knowledge.
How important according to you is the study of Mathematics for Machine Intelligence?
Very important. Mathematics is important to read and understand the papers in the area and to formulate problems and solutions.
Is it possible for a layman to understand the mathematical nature of Artificial Intelligence by simply observing it?
Well, a layman would be able to appreciate the mathematical structure, but to understand it and to operate with it they would need to study the underlying mathematics principles. The underlying mathematics is not very complex, but it requires you to build a large body of knowledge so that you understand what the experts in the field are talking about.
If you were given a student with some programming and no mathematical background to mentor where would you advise him to start his studies in AI from?
The student will have to first learn some basic mathematics concepts -- linear algebra, probability theory, optimisation, information theory. Once they have those concepts, one early book to master is Judea Pearls 'Probabilistic Reasoning in Intelligent Systems'. Another book is Kevin Murphy's 'Machine Learning, a probabilistic perspective'. Bishop's 'Neural Networks' is a classic that everyone should read.
You have been mentored by the likes of Jeff Hawkins and Nils Nilsson. Was it easy to get such pioneers in the field to mentor you?
Not exactly. I joined Stanford with the aim of understanding the brain, but I had to explore several paths and deal with a few unexpected setbacks before I met Jeff Hawkins. Working with Jeff required me to step out of the comfort zone of doing a "normal" PhD at Stanford. Being in Silicon Valley helps -- all around you are examples of people trying to do the impossible in unconventional ways.
There is a lot of information available on the internet about artificial intelligence and similar subject is there anything that can help students differentiate from what's real and what's not?
This is hard -- learn about all of them and then come to own conclusions.