Sign in to follow this  

Starting new Project

Recommended Posts

Hey everyone, I'm about to start an A.I project for my postgrad. I haven't decided the exact specifics but here is what i'm thinking so far. I would like more than one type of species to be involved, preferably one being prey the other predator. Each species would have different attributes and each individual can have further attributes. This should create a sense of individualism where every creature is different but within the boundaries of their own species. I would like each species to be able to react to the environment and to other species, individuals should also be able to breed...but only in the right conditions In addition simple flocking algorithms could be included where species try to stick together as much as possible by using group voting. However each individuals vote has a different weighting, this way leaders of small groups may evolve where an individual has most of the say of what the pack does. I had a suggestion that Intelligent Agents be used for the individuals, of course i would have to had some Genetic Algorithms and include mutations for the breeding part. I was hoping i could get some input as to how some of you think this sounds. Also if you know of any useful links? I'm in the middle of going through lots of sites such as and But most importantly, my tutor rose an important issue. While it would be fun to create a system where it was just prey and predator and the system just worked by itself and the prey learned to move in herds and to co-operate etc etc. It won't actually be evident to the user (I.E the external examiner who gives me my mark and doesn't know squat about a.i) that any "itelligent" decisions are taking place. I need to somehow show that the agents are learning and developing. I don't know how to go about this does anyone have any ideas? If the system was about humanoids it could be shown that they are learning about fire, weapons, buildings etc etc. But i don't want my creatures to be that 'advanced'. So i'm stuck...and i can't really start untill i know what direction to take it in. So anybody got anything? Thanks for reading T

Share this post

Link to post
Share on other sites
Well even in animal systems there is a difference between males and females, strong and weak, old and young. Have you ever seen deer fight each other? The but antlers together and fight their own community in other to become more prominent in the community. The strongest becomes the alpha male, who has the best authority and breeding priveliges. The deer just don't randomly fight each other because it hurts when they have and they usually need good reason to do so.

Also is you want to you can have a map with different climates and surroundings. Where will the prey go and where will the predator go? If and area is too deadly for the prey it doesn't matter if it has the best foor or not ... but then again they may stay and just reproduce faster so that several of them can die and a resonable number be kept alive to reproduce again (a decesion usually make by the alpha male and disagreed by some others).

As to HOW you do it is even for a simple intelligence prey/predator you will need to record past experiences (for a more real setting you may not record all, or remove some after a long period of time). This mean you will have to have the agent be able to understand a LARGE amount of past information, not just GOOD or BAD being saved but something closer to:

..slept ...
.. REproduced ...
.. attacked ..

I hope I helped a little, and best wishes on your project.

Share this post

Link to post
Share on other sites
Sticking with your original idea...

It sounds like the learning is an 'optimization' for one of the entities (either the predator, in becoming more able to kill prey) or the prey (in its ability to avoid the predator).

What if you allowed them to evolve in lock step? Let the predator evolve for a while. Track statistics on their initial success and the success after the evolution. Have something that demonstrates that they improved.

Once you do this, allow the prey to adapt. Again, have some sort of statistics on before and after.

As you are trying to show that there is AI involved, it seems like the first thing you need to do is have a goal and a way of measuring your progress towards the goal.

I would also suggest doing a bit of research on population fluctuations in predator-prey systems. There is a lot of interesting material on the net which you may want to look at. One initial hit was This paper gets pretty deep, but does have a huge number of references.

This paper also looked interesting -- it may have a few useful keywords to search for, as well as a MUCH simpler to read discussion of systems:

Share this post

Link to post
Share on other sites
Gryfang: To begin with i don't want to include any differences in sex, i want to keep it simple. Maybe at a later date i could modify it.

Your suggestion of a map with different areas of climate, was something i had thought i might include but i had never thought of it in a way that if an area is to dangerous it doesn't matter if it's also an area that results in high number of offspring. I quite like that.

BrianL: Indeed, i have been trying to think of what goals i could come up with but there is nothing really specific. In a real prey/predator climate the only goal is for the herd to survive and breed. Individually also it is to survive and breed. Thanks for the links, i will look into them.

Thanks for both your suggestions, they have got me thinking.

Share this post

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this