Techie Tuesday (on Wednesday)

I figured I would share some interesting tech tidbits. Actually I am going to try and share these each week on Tuesday, so check back for good techie tuesday news and thoughts.

My first one is about CADIE - Google's newest artificial intelligence creation: http://www.google.com/intl/en_us/landing/cadie/

Well actually this was really their latest April fools joke. Which makes it now 2 years running that it has taken me most of the day to figure out that they were trying to trick me with their april fools jokes! b@st@rds!

Does anyone remember the send it in the past Gmail trick? Yeah that one got me, oh it got me good. I was so excited last year that i actually blasted it out to everyone i know as a good example of how google's engineers just really understand US, the people who use their products.

OK, but I digress, I am actually hear to talk about the would-be CADIE and whether such a thing could be possible. Well the truth is maybe, but maybe not in the way you think.

According to Google the make-believe CADIE (who actually says a whats up to an old professor of mine Dr. Norvig - no seriously his name is Peter Norvig and he is the head of research at Google) is based on "a significant breakthrough" in how to approach "reinforcement learning" problems. People may not realize just how close to reality some of these topics really are - so I thought I would write a short spot on some current topics in AI and more importantly how it relates to our lives.

What the heck does that mean? Essentially it means that they have a way of taking data and learning what about it, then reinforcing what they have just learned by studying a similar set of data again.

Its kinda like slapping your kids wrist everytime they reach for the candy jar. Except that kids usually learn - most kids usually learn (*cough* Jarod... ) - to stop reaching for the candy jar. Whereas most computers reach for the candy bar A LOT - they just really want to make sure you are going to hit them before they decide it is a good idea to truly stop trying to get that candy.

The goal is to create algorithms which can figure out this whole candy jar business in just a couple of wrist slaps - and better yet extrapolate that since your mother told you not to eat candy and she told you not to drink soda that you might get a similar wrist slap for drinking too much soda as eating too much candy.

Well I am not going to get into a discussion about the meaning of intelligence. I will leave it up to you guys to decide if this would truly constitute intelligence. But one thing that is interesting, is to think about how good computers can be at faking intelligence. One of the classical tests (the Turing Test) of intelligence for a computer was whether or not it could converse with a human being and fool the human into thinking it was a human as well.

Then, of course, right after we decided that it was a test worth beating, we went and solved the darn problem and in less than 250 lines of code (if you are not a developer just read - very very little work)!

This is all important because we want to think about what a CADIE-like program might be able to do for us.

Well here are a few things, if a computer were able to learn in the way that CADIE is, it might be able to figure out what food products stores like yours were purchasing and recommend which products you should be looking at. If it were REALLY intelligent it might suggest to a food manufacturer just which food buyers they should market a given product too, it would be like the perfect marketer playing match maker between buyer and seller (in fact these are a couple of the crazier things we are working on at Findood).

In fact we all get functionality like this today from Amazon (people like you bought this product, don't you want to buy this product too?). However, what would be truly revolutionary about CADIE would be that reinforcement and extrapolation part. The part where the program doesn't just look at people like you, but also looks at people that are a little bit like you and other people who are also a little bit like you but in different ways, and then learns and extrapolates what applies to you, because after all we are all at least a little bit special. Seroiusly there really isn't anyone else out there quite like you or me, and it would be amazing if the next time your support call were routed to India it was actually routed to a computer that treated you as unique as you actually are... just an april fools, or is it?

thanks for reading.

PS Do you have questions about how to use technology in the work place? If so, send me an email elliot AT findood.com and let me know what questions you have, I will be happy to discuss them here.