Theme: Agency

Back when I thought AI had a lot of promise I created a bit of software that used emotions. (a Tank). I stored memories as problems, actions, and consequences, each as symbols referencing other symbols constructed from limited operations and ‘feelings’.

(I was not in favor of neural networks which i saw as useful in creating symbols but unnecessary if we already can work in symbols. )

I think google is doing an interesting job of associations but I don’t see those associations reduced to operations and changes in property, and the corresponding emotions, which is what would be necessary to produce a sympathetic intelligence.

It wasn’t until much later I understood that it’s property that’s the unit of commensurability and recipes that transform states as just another set of actions.

And it wasnt until a friend at MSFT told me about using manifolds as data structures that I began to see how all of this would fit together.

I see at least three avenues to AI. We all prefer the one we understand. And I think it will be a sony-betamax problem of we invent what is useful but not best. And this will delay us in getting to best, because it requires a lot of infrastructure to produce these components and we will have to exhaust that venue before we try an alternative.

That’s what I think I see happening.

I’m the only one working with property. although some bit-coin nut might stumble across it.