Some notes on Nils Aall Barricelli
https://nautil.us/the-computer-maverick-who-modeled-the-evolution-of-life-234936
“I have not heard of him to tell you the truth,” says Mark Bedau, editor of the journal Artificial Life. “I probably know more about the history than most in the field and I’m not aware of him.”
https://www.chessprogramming.org/Nils_Barricelli#Barricellian_Symbioorganisms
Thanks @S_Conradi for https://mathstodon.xyz/@S_Conradi/112983346836554624
@akkartik @S_Conradi First time I encountered #Barricelli was in George Dyson's "Darwin among the machines". He truly was one of the early anti-heroes/enigmas of evolutionary computing and ALife... For quite a few years I found it very hard to find detailed literature/papers about his approach(es), but still have been riffing on some of his (mostly only loosely/vaguely & sometimes contradictory documented) ideas, experiments and research about "numerical symbiosis" since ~2010, years later (and more recently again) also working on some actual projects and also still aiming to publish some of these experiments as libraries for others to play around with... The poster image for my website is also from one of these projects (visualizing different numbers as 3D particles of varying lengths)
Some more related links:
https://www.tim-taylor.com/selfrepbook/
https://www.tim-taylor.com/assets/docs/barricelli-suggestions-for-starting-theoretic-papers-6-6-1987.pdf
https://www.chilton-computing.org.uk/acl/literature/books/gamesplaying/p004.htm#index78
Attached: 2 images @[email protected] Ahh, I found my bug! Now things look much more interesting. All numbers are equally likely to die out. As you know, the simulation moves arr[i] to j where: j = (i + arr[i]) % len(arr) 0 values never move, and later phases introduce and eliminate 0s in carefully balanced ways. The mistake I made was in compensating for Lua's 1-indexing: j = (i + arr[i]) % len(arr) + 1 Now 0s start to move, which introduces bias. It should be: j = (i-1 + arr[i]) % len(arr) + 1
@akkartik Already looking great! Barricelli's conceptual setup is so wildly flexible, you can have a lot of fun (and literally can spend days/weeks/months) coming up with new replication rules for the different cell collision cases, i.e. deciding how to resolve them and adding more factors to encourage emergent behaviors, for example using combinations of:
- recursively search for free neighbors, incl. using varied step sizes (per gene/number) and allow scattering/spilling to future generations if needed
- use per-gene LUTs with different "symbiosis" equations for combining numbers and choosing behaviors in case of collisions
- including multiple neighbor cells into symbiosis/mutation decision making (a kind of halfway house to cellular automata)
- define fields (e.g. image/mask based) or probabilities for using smaller ring sizes (for modulo, per cell!), acting as a kind of environmental and/or time-based gene reset/limiter/filter (also include an ring offset to remove bias and so that resulting genes aren't always zero-based)
- add probabilistic noise (either by mutating genes/numbers themselves or their target location, pre/post collision also can make a big difference here)
- define/randomize gene priorities, in case of collision migrate the gene with lower priority (related to scattering in #1
These are just some ideas I've been experimenting with myself (attached image uses several of the above approaches)...
Look forward to seeing more of yours!!!
Kartik Agaram
Karsten Schmidt
Simone Conradi