AI for Dummies, in Bite-Size Pieces, Part 1

(WaPo) A curious person’s guide to artificial intelligence claims to offer “Everything you wanted to know about the AI boom but were too afraid to ask.” It’s a decent lexicon and taxonomy. This is for readers who would like to scratch a little deeper.  Since I have worked in AI, the  explanations I will provide are well intentioned approximations. Perhaps the reluctance of academics to relax rigor is why such explanations are uncommon.

Deferring  the beginning of the AI story, we start with the juicy middle. You want to know about neurons. The neurons of AI are inspired by, but rather different from those of your brain, simplifications based on  common features of all biological neurons. An individual neuron cannot think; billions of them apparently can.

What does an individual neuron do? A useful simile that might make specialists  cringe: It resembles an odd kind of voting machine:

  • One type of constituent – typically another neuron, can only vote yes, or abstain.
  • The other type of constituent neuron can only vote no, or abstain.
  • Votes are individually weighted for importance. The weights are determined during training.
  • The votes are summed. If the weighted sum of votes exceeds zero, the motion passes, in which case the neuron sends a “yes or no” to some of the billions of other neuron of a thinking machine or chatbot.
  • If the motion does not pass, the neuron sends an “abstain” to some other neurons.
  • Typically, a “yes-for” or “yes-against” result is signified by the number “1” and a “abstain” by “0”. The sign of the weight on the input of the recipient neuron determines “for” or “against.”

In biology, a neuron is a living cell, an actual thing. In AI, it can be an actual electronic circuit, or an abstract simulation as lines of computer code. The first is experimental. Simulation is standard practice.

Next, a little history.