Thanks Wes. After reading your review, I have just moved her book up on my List Of Books To Read List.
I bought “The Devil in Dover” to be my companion read on a recent business trip. That was the plan anyways. It arrived on the Friday prior to a Sunday flight. By mid-day Saturday, I had finished it. Lebo’s personal tale made it a far more compelling read than Hume’s “Monkey Girl”, which was a darn good book in it’s own right.
Perhaps I should wait until you can blog about the experiment and analysis you refer to at the top of this post, but I found it very interesting that you were running Avida. Wandering over to the Avida site, I found that you are part of a team on a project “Evolving Intelligence”.
Let’s ignore the overly grandiose title of the project, the pomposity of the project description, the name dropping, and the studied ignorance of the field of sub-symbolic AI, and let me ask one question. How is Avida the right tool for this task?
Historically, Avida digital organisms are these autistic little assembly language programs. They don’t communicate with each other. They have no environment to speak of. They compete for CPU time and rewards for encoding logical functions.
How does any of that relate to the evolution of pattern recognition? The IO instruction does not probe the cellular environment in which the organisms are (perhaps) embedded, nor are the “reactants” embedded in that environment. I’m sure the project team can jam new functions into Avida to try to support whatever you think will be required, but why bother? It feels like an attempt to evolve Avida into Sugarscape or John Holland’s ECHO.
I’m very sympathetic to the desire to probe the origins of any kind of intelligence, human, cephalopodian, or whatever contingency allows. I’d love to hear why Avida is the right tool for the job, because I have a hard time seeing it.
I can offer some general comments here. Certainly there is a pragmatic element to using Avida, since that is the platform most used by those of us at the MSU Digital Evolution Lab, and thus there is a lot of platform expertise available from my colleagues. I’m sorry that the description of the EI project seems to have caused aggravation to you. Part of the reason that I am on the project has to do with my background in sub-symbolic AI, I’ll note. I think that Avida’s underlying design is made for expansion, so providing new functions for an infrastructure that can readily accommodate them is not so outrageous a proposition as it seems that you are asserting. There are a number of assumptions built into Sugarscape and ECHO concerning relatively high-level concepts, so I think you might concede that taking a somewhat different conceptual approach in research could also require substantial back-fitting of either of those platforms as well, which, if we are being consistent, would also be a strike against them.
If you plan to be at the SSE conference starting on the 20th, I’ll be presenting a poster there on some preliminary results. Do feel free to drop by and rag on me in person.
I understand that at the start of any project there is a build/buy decision, and it’s easy to default to what the lab has been building on.
Avida may be designed for expansion. (I remember that at some point during the development of programming editors for the PC, some wag said that the next version of such a product would just be a C compiler, “an editor shipped without a default UI, but with a world class scripting language!”) But that implicitly admits that the core, digital ant farm Avida isn’t germane. Looking at the SourceForge distribution, I didn’t see a modular architecture designed to be expanded, I saw code pulled in a lot different ad hoc directions for different research projects.
Certainly ECHO and Sugarscape suffer the same faults, and more of their own. I was hardly recommending them as alternatives, only that it seemed to be where some projects were pulling the Avida codebase. Add events, local resources, sensors, and communication to Avida and you are well on your way to either of these.
I’m sorry I won’t be at SSE (what is it?), but I would love to see a copy of your presentation. I was amused, not aggravated, by the project description, and I would like to see what direction this is going to go. Are you focused on the evolution of pattern recognition or on contingency, with pattern recognition simply an exemplar?
thank you for such an informative review>>>
Comments are closed.