A.I.
Artificial
intelligence
Way, you love me
Love the way you love me
It's so clear
You make everything inside me feel
Just automatic hurt
Bring me back tonight'
Cause you're intelligent, so real
So real
Artificial Intelligence?
Are you kidding?
While Artificial Intelligence, AI, is getting a lot of press these days, it is not really anything new. Only the terms are new. It used to
be called Machine Learning. Before machines, it used to be called the Wisdom of
the Crowd. When lower animals do it, it is called Neural Networks or Instinct. “Artificial”
is because it is done quickly by computer, the artificial part that does not mind
processing many, many observations. The “Intelligence” part should more
properly be called “Inference”, because there is nothing intelligent about it.
If it were intelligent, it would know what caused those inferences. What results are correlations given a multitude of observations However, the correlations can themselves have
problems.
1. Spurious Correlations
Just because there is a strong correlation between an event and something else that is observed at the same time, it does not mean that this correlation has any meaning. There is a book about spurious correlations. (Virgen, Tyler; Spurious Correlations; Hachette Books; New York; 2015) My favorite is that there is a strong correlation between drowning deaths in pools and the number of Nicholas Cage film appearances in a given year. No mater how bad people think Nicholas Cage is, I don’t think that is the reason people drown in pools.
2. Absence of evidence is not evidence of absence.
A correlation can only be made of observations that were made. If the sample excludes certain observations, any correlations would be suspect. Famously, the Literary Digest predicted that FDR would lose the 1936 election to Alf Landon because its readers said in a poll that they were overwhelming voting for Mr. Landon. However the public included many, many voters who were NOT subscribers to Literary Digest. In November, Roosevelt won the election in an unprecedented landslide, winning every state except Maine and Vermont while also winning the popular vote by 24%. If you don’t make an observation, then the correlation to that observation will of course be zero. And therefore your sample will be biased.
3. Correlation is not causation
Just because there is a correlation it does not mean that the observation was caused by the event. It takes intelligence to make sense of how and why a correlation exists. There were thousands of years of observations of the planets. The popular wisdom was that the planets revolved about the Earth, and that observations of planets were explained by retrograde motion of some planets. In fact it took Copernicus to intelligently propose that all of the solar system planets, including the earth, really revolved about the sun.
If someone tells you that AI told them so, remember this. Then you won’t have to be like
Galileo when the inference was that all objects in space move around the Earth,
and be forced to recant, as he did, of his observations
of the moons of Jupiter. But still they move.