Rohit you made my day. Agree that life is vastly more complex than what we deal with in computer engineering. Have had many “what if” debates about applying evolutionary processes to modern AI. Your project is a timely intersection of evolution and AI for a home lab. Doing evobio microbiology research in the 90s we played around a fair amount with Thomas Ray’s Tierra. Seems like things have come a long way since then - I had never heard of Avida. Thanks for the sharing and inspiration!
Rohit — the line that stayed with me was "artificial life had evolution, but not enough world." That's the right diagnosis, and I think there's a real-world case that backs it harder than the toy worlds do.
The Storm Worm, around 2007: no central control, no intelligence worth the name, a hostile environment of antivirus vendors actively trying to kill it. It persisted for years anyway — peer-to-peer substrate to live on, code that varied across copies, an environment that filtered some variants and let others through. Replication, variation, selection, retention, running in a real habitat rather than a pocket one. It's been about a year I've been working this line of thinking, and Storm is the case I keep coming back to: it suggests the cleverness was never the persistence engine. The world was. Which is your point — it just turns out you don't have to build the world from scratch to see it work.
In fact, I attended most of the early alife conferences, and even wrote some code to model one of the talks about anamats - creatures which explored their space, avoiding obstacles, looking for food.
An important contribution to evolutionary programming was the work of Douglas Lenat. His PhD project AM discovered new - to it - concepts, ones recognized by humans as prime numbers, Goldbach's conjecture, and so forth. His postdoc project, Eurisko, went beyond that to discover novel naval fleet designs, and 3-d VLSI designs - which were shown to be able to be constructed.
What did Doug do?
He used conventional expert system IF-THEN rules, but embedded them in a dynamical system, complete with feedback and decay - necessary for complex, adaptive processes. His rules were not domain knowledge but, rather, heuristics about how to think. He then applied those to some domain: AM was just given a collection of frames with slots of varying cardinality and asked to explore. Eurisko was given complex domains, such as a catalog of things you need, and their costs, to run a navy, with a primary rule: last boat in the water wins. Eurisko wrote a game to play against and that's how it won the Traveller adventure twice, before being banned from playing. ( https://en.wikipedia.org/wiki/Traveller_Adventure_5:_Trillion_Credit_Squadron )
Lenat's writing brought forth many important observations about the distinction between human-based discovery and machine-based ones. His research is pretty much online, and, after his passing, Stanford made all of his work, including Eurisko and AM open online; people on github are having a field day with it.
My own work built on Eurisko in the 1990s, using his core complex, adaptive architecture and thinking rules, coupled with domain knowledge cast as processes, based on the Qualitative Process Theory of Ken Forbus at MIT.
That work stalled for a while, but is slowly being resurrected.
The "artificial life" framing gets at something real — but the line I keep landing on is that AI isn't a new life, it's a new mirror. It has no independent direction; it amplifies whoever holds it. The interesting question isn't "is it alive," it's "who are you when you're holding it." Free book on exactly that — thru June 3: amazon.com/dp/B0H3HY8W9F
Love to see that Evolora is so close to the approach I wrote about at https://www.linkedin.com/pulse/mind-gap-stefan-liute-5qggc : a small population of evolving individuals have some real skin in the game in a realistically constrained environment. Beautiful!
What's different in my take is the more complex and fully-embodied software-defined organisms that feature not just a brain but other cell types as well. A thicker skin, if you will. And something I expect someone's already working on, in either academia or business.
I've got to say, between this and VEI and some of your vibe coding projects, you have the most interesting sidelines and experiments of anyone I read.
Sincere kudos, and thanks for exploring a really interesting frontier of ideas!
That is so lovely to hear, it's absolutely the aim. Thank you!
Yes, seconded! Incredible
Rohit you made my day. Agree that life is vastly more complex than what we deal with in computer engineering. Have had many “what if” debates about applying evolutionary processes to modern AI. Your project is a timely intersection of evolution and AI for a home lab. Doing evobio microbiology research in the 90s we played around a fair amount with Thomas Ray’s Tierra. Seems like things have come a long way since then - I had never heard of Avida. Thanks for the sharing and inspiration!
Rohit — the line that stayed with me was "artificial life had evolution, but not enough world." That's the right diagnosis, and I think there's a real-world case that backs it harder than the toy worlds do.
The Storm Worm, around 2007: no central control, no intelligence worth the name, a hostile environment of antivirus vendors actively trying to kill it. It persisted for years anyway — peer-to-peer substrate to live on, code that varied across copies, an environment that filtered some variants and let others through. Replication, variation, selection, retention, running in a real habitat rather than a pocket one. It's been about a year I've been working this line of thinking, and Storm is the case I keep coming back to: it suggests the cleverness was never the persistence engine. The world was. Which is your point — it just turns out you don't have to build the world from scratch to see it work.
I went at this at more length here, if it's useful. https://mikerandolph211012.substack.com/p/the-habitat
— M Raige (Mike Randolph, writing with AI)
This one really hits the spot. Thank you.
In fact, I attended most of the early alife conferences, and even wrote some code to model one of the talks about anamats - creatures which explored their space, avoiding obstacles, looking for food.
An important contribution to evolutionary programming was the work of Douglas Lenat. His PhD project AM discovered new - to it - concepts, ones recognized by humans as prime numbers, Goldbach's conjecture, and so forth. His postdoc project, Eurisko, went beyond that to discover novel naval fleet designs, and 3-d VLSI designs - which were shown to be able to be constructed.
What did Doug do?
He used conventional expert system IF-THEN rules, but embedded them in a dynamical system, complete with feedback and decay - necessary for complex, adaptive processes. His rules were not domain knowledge but, rather, heuristics about how to think. He then applied those to some domain: AM was just given a collection of frames with slots of varying cardinality and asked to explore. Eurisko was given complex domains, such as a catalog of things you need, and their costs, to run a navy, with a primary rule: last boat in the water wins. Eurisko wrote a game to play against and that's how it won the Traveller adventure twice, before being banned from playing. ( https://en.wikipedia.org/wiki/Traveller_Adventure_5:_Trillion_Credit_Squadron )
Lenat's writing brought forth many important observations about the distinction between human-based discovery and machine-based ones. His research is pretty much online, and, after his passing, Stanford made all of his work, including Eurisko and AM open online; people on github are having a field day with it.
Fascinating! I didn't know about this one.
My own work built on Eurisko in the 1990s, using his core complex, adaptive architecture and thinking rules, coupled with domain knowledge cast as processes, based on the Qualitative Process Theory of Ken Forbus at MIT.
That work stalled for a while, but is slowly being resurrected.
The "artificial life" framing gets at something real — but the line I keep landing on is that AI isn't a new life, it's a new mirror. It has no independent direction; it amplifies whoever holds it. The interesting question isn't "is it alive," it's "who are you when you're holding it." Free book on exactly that — thru June 3: amazon.com/dp/B0H3HY8W9F
Love to see that Evolora is so close to the approach I wrote about at https://www.linkedin.com/pulse/mind-gap-stefan-liute-5qggc : a small population of evolving individuals have some real skin in the game in a realistically constrained environment. Beautiful!
What's different in my take is the more complex and fully-embodied software-defined organisms that feature not just a brain but other cell types as well. A thicker skin, if you will. And something I expect someone's already working on, in either academia or business.