Some thoughts on Tom’s recent presentation, which can be found here:
This is a pretty good description of the additions I’ve made over the last couple of years. On the issue of the complexity of multiple simulations on multiple levels that probably is quite frustrating if you’re trying to do a reductionist analysis to find out what influence a particular factor has. In that regard it’s similar to a real ecosystem. Unlike a real ecosystem though the advantage of a simulation is that a detailed analysis of the chain of causality is in principle possible, although it might take some logging and detective work to obtain.
On the question of whether the whole system could be crashed in a chaotic way by some small event in one of the simulations, that might also be possible although I havn’t done any significant amount of investigation in that direction. So it could be the case for example that one ape contracting a particular pathogen or having a genetic makeup which makes it a good host for a particular kind of pathogen results in a pandemic which kills off a lot of apes. Or another situation might be that there is a particular set of cultural practices, spread via the braincode, which keeps parasite levels low but if that culture fades away then there could also be increased mortality levels. So it’s certainly possible that some tiny event or decision could have large macroscopic effects.
I’ve had an interest in animal behavior for a long time, and this might be partly where an interest in AI and robotics came from. One of the early books which I read about AI was “Intelligence as Adaptive Behavior” by Randal Beer. That book is pretty much at the intersection of AI, neuroscience and Artificial Life. Later on I read a couple of books by Jane Goodall: In the Shadow of Man and Through a Window, both of which are very much about the narrative history of a particular group of apes. Another influential book was “Ape, Primitive Man and Child”.
On the current confusions about ancestry I’m still not sure whether this is a bug or a feature. Currently conflicting relationship claims are possible, where an ape might start off having a particular grandmother but then be told via an anecdote at some later stage that they have a different grandmother. Maybe such conflicts should simply be detected and removed explicitly, or perhaps there should be some cognitive process which tries to maintain narrative consistency.
Is Noble Ape useful to biologists? It might have some use as a teaching aid to get people interested in studying biology. Currently it’s not a highly accurate simulation of any particular environment or any particular type of ape though.
Possible future developments
There are many possible directions in which the simulation could go, and there is plenty of scope for improving or refining the existing components. Some things which might be worth doing are:
1. Changing the way that memories are stored. Currently a fading memory model is used, whereby the older an episodic memory is the more likely it is to be replaced. This might not be very realistic, and it might be the case that particular memories are actively maintained.
2. The ability to treat other things which are not minds as if they were (anthropomorphising). This would include the ability to treat territory as if it had agency, or to make generalisations about groups of apes.
3. The braincode is currently only the most abstract representation of a language system. Perhaps more language elements could be added, such as learning or uttering particular words/calls.
4. Better visualisations of the social graph, hierarchy and ancestry. Also visualisations for the metabolism and vascular system.
5. An expansion of the actions system to enable more playful and arboreal activities.
6. A way of identifying and visualising common sequences within the ideosphere.
7. Other creatures for the apes to interact with.
8. A more generative ability to combine objects into assemblies – primitive toolmaking.
9. Better 3D graphics. More realistic looking apes and vegetation. Rain and wave effects. Shadows which change with the sun.
10. Ability to network multiple simulations together.