AI in five minutes.
An artificial intelligence is a collection of math equations.
Mathematicians call these equations algorithms.
We have data on a million people. For 5,000 we also know their age.
We want to build an algorithm to figure out EVERYBODIES age.
We start with a set of data, the training set, the 5,000 people who's age we DO know.... like Bob.
Bob is 60 years old. we know 26 other things about Bob.
How do we find his age from using other data? Say... eating habits, shoe type, musical preference.
F. Bob eats at Denny's every Sunday at 2:00 and uses coupons
S. Bob buys orthopedic shoes
M. Bob listens to Lawrence Welk Sunday evenings
Make food, shoes , and music into numbers. F = 20, S = 10, M = 1955
We start with a best guess equation that looks like this: F + S + M = 60
1985 does not equal 60. Obviously we to do something to the value of M.
Let's divide it by 1000... that means M is getting a co-efficient of .01
20 + 10 + 19.5 = 49.5
F (1)+ S(1) + M(.01) = 49
49 is closer to 60. That's a good first step. Tweek the other variables, food and shoes.
We try them all until we get as close to 60 as possible... FOR BOB
F(1) + S(2) + M(.01) = 60.5
By this logic, why not just use food F with a co-efficient of 3... F(3) = 60
forget about shoes and music and be done with it?
Because it doesn't work for skinny assed 60 year old Charlie who has F = 5. so... F(3) = 15 but Charlie is not 15.
And then there's is Debbie and the 4,997 other people. This is why we need the computer.
Bring back the shoes and hot dogs; run the numbers for all of them: 55 Damn, still off by 5.
That's that best we can do with F, S and M ; BUT we know more than food, shoes, and music
We have a full alphabet... try everything we know.
After trying every combination and every co-efficient with all 5,000 the best 'fit' for our prediction is ...
M(.01) + W(.1) + R(.5) = 59
Music, Wife's age and Rinkles turned out to be the best predictors... food and shoes? Not so helpful.
We now have a predictive algorithm for AGE we can use on the one million people... but 59 is not 60.
What more do we need to know about the 5,000 (one million actually)? We need a better model.
Go find out some more stuff about Bob! We don't know what will be helpful, so... get EVERYTHING!
The machine is hungry. Feed the machine. After we get age down, move on to
Air temperature, all the books, bean futures, GDP, voltage drop, meme trends... (ohhh... meme trends!)
Algorithms, forever tweeking algorithms to find whatever helps make it all... perfect.
That's AI.