Swarm intelligence

Hi all,

I am presently completely fascinated by unanimousAI - google around for some of the news articles on it - here's their official website - To enter you just enter a username and password as guest,thats it, no email involvement at all.
Picking the kentucky derby 1-2-3-4 was a knockout haha, its a brilliant idea.
I've added a question about the English premier league football champions, as a test, in categories Sport - if anyone fancies interacting on that, user is chonji.

Its given me the idea to code my own population and run my own sims. I am presently trying to design a template agent.

This has revealed many questions to me: Like...

Lets say, randomly applying knowledge values to its attributes - do you allow an agent to have a high intelligence but a low score on a 'well informed' attribute? would that make sense? does that match the profile of anybody you could think of? Could it be somebody that is very intellectual in a field, yet is just not interested in watching or reading the news, eg current affairs. i am finding I can formulate a case for most profiles - people are people!

Quote:

---

Originally Posted by rogster001

Lets say, randomly applying knowledge values to its attributes - do you allow an agent to have a high intelligence but a low score on a 'well informed' attribute? would that make sense?

---

If I understood it well, it does make a lot of sense. And I'd wager, makes for a better turing agent. I would probably implement this as a sort of "tech pool" for each attribute.
Maintaining all random properties of the decision tree, the agent decisions nonetheless draw from the tech pool of those attributes affected by each decision. Once exhausted, it takes some amount of time (or some amount of new decisions not affecting that attribute) to refill. During that time, the agent decisions are less informed, by lowering the attribute score internally. The more empty the pool is, the lower the internal attribute score. I suppose you could then randomize the amount of effect of each decision on each attribute's pool to maintain a balanced AI, but that still shows signs of small bouts of continuously wrong decisions, as a human would.

I hope I explained myself right.

Is this like - a finite amount of each attribute is available to be distributed amongst the population?
The approach i was formulating has something along those lines - EG for a 'general intelligence' property, allocate say 15% of agents with "high", 25% above average etc. I had actually done a little sample in Excel while thinking about this originally, see below.

This was just some random scores for intelligence, 'informedness' , knowledge of sports (as a problem specific value). I was just interested to see what came out.

Attachment 15024

Agent1 turns out to be intelligent, not very well informed though and has little knowledge of sports.