I watched 1986 version of the 'The Fly' last week and realised that the core of the film was about machine learning and it was a good demonstration of how it works and how some limitations. I shan't explain the plot but at one point Jeff Goldblum (playing his familiar role of maverick scientist) realises that the reason his telepods are not teleporting animate objects properly is that they do not understand what organic material is so they create their approximation of it. He proves this by teleporting a piece of steak and cooking it for Geena Davis, his co-star. The steak looks OK but tastes wrong. After he realises this problem he trains the computer by teleporting steak until comes through the other side at a quality that is edible. He then has a workable teleportation machine that successfully transmits live animals and he tests it and it works. 1- 0 to machine learning v human overconfidence.
Alas, when he transmits himself there is a fly in the telepod and the machine hasn't been taught how to deal with two entities and has a guess and decides to fuse them at the DNA level with unfortunate results. The machine could have been taught how to deal with multiple entities but Jeff Goldblum did seem to be a scientist in a hurry with a taste for shortcuts. 1- 1 machine learning v human overconfidence.
Another subtle warning in the film is that the Goldblum character is a scientist/inventor but he does create any of the components. He designs them, other people build them and then send them to him. He puts them together to get the effect he wants without anyone being aware of the objective but he really doesn't know how they are built.
If he had been able to write a few lines of code he could have written something along the lines of:
If NumOf Entities >1 Then
Print "Clean Out The Telepod"
Exit
Endif
Then he would've avoided all the unpleasantness of being turned into a human fly and picked up his Nobel Prize and we would not have overcrowded trains on the way into work.
