Murphy’s law really applies to AI.

“Artificial Intelligence has an enormous affection for euphemism [and its practitioners don’t recognize] the superficiality and immeasurable naivete with respect to [certain] concepts”.

The assertion above is the gist of a talk on the dangers of Artificial Intelligence… in 1986.

As one example of that thesis, thast talk describes the case of a doctoral student developing software that basically makes the computer image of a harmless teddy bear and an equally harmless kitten play with each other… according to the instructions spoken by the child playing with the computer.

But from a technical point of view, that playful, harmless application would actually be, with minimal changes, the same software needed to make a fighter jet drop bomb on some target, as soon as its pilot utters certain words.

Fast forward to AI in 2022

In real life, the John in the picture above is a guy who never went anywhere close that naked woman which is not his wife. Still, that and other pictures “prove” that John is a really bad guy involved in all sorts of illegal, or otherwise deeply troubling activities that he never took part in in reality.

This is because creation of potentially life-wrecking fake images and videos has become really easy this year, and one reason why it happened is exactly what that talk of 1986 was talking about (emphasis mine):

“from the beginning, Google’s researchers knew that if you can put humans (like John) in any fake situation, trouble might follow. That’s why when they announced Dreambooth, they used pictures of Corgis and not people as examples. [But this is] an example of how AI scientists often try to minimize or avoid criticism for the potential negative impacts of their research."

Maybe Murphy’s Law should be renamed “AI Law”.