Google creates self-replicating life from digital 'primordial soup'
newscientist.com
newscientist.com
At the time I learned about Tierra I didn't have a way to get at any of Ray's papers or code. Now I do, but I'd kinda forgotten about it. I really should revisit it. There's probably a lot of fun to be had there.
Aside: Steven Levy's "Artificial Life"[1] was where I learned about Tierra. That book, along with Levy's "Hackers"[2], and Gleick's "Chaos"[3] were formative books for young me.
[0] https://en.wikipedia.org/wiki/Core_War
[1] https://www.amazon.com/Artificial-Life-Frontier-Computers-Bi...
[2] https://www.amazon.com/Hackers-Computer-Revolution-Steven-Le...
[3] https://www.amazon.com/Chaos-Making-Science-James-Gleick/dp/...
Genius. Thanks!
Here's a direct link to the paper. I have not read it yet but the abstract reminds me of a recent post discussing using brainfuck as a way to do program synthesis.
I'm excited to read more about Tierra and others prior art to understand if observation has precedent. My introduction to all of this is Shiffman's Nature of Code, so I am not familiar with the literature.
The other new research introduced in the paper was the way the BFF VM and "tapes" were observed. By using entropy and compressibility, the authors observe complexity as a heuristic for presence of self-replication in the "primordial soup" simulation.
Computational Life: How self-replicating programs emerge from simple interaction - https://news.ycombinator.com/item?id=40820022 - June 2024 (17 comments)
> Indeed, many of the team’s experiments ran for millions of steps [...]. Laurie says that one instance, running on his laptop, involved processing about 3 billion instructions a second and it still took around half an hour for self-replication to emerge.
This almost reads as satire. In this day and age, 30 minutes on a single laptop doesn't quite trigger my sense of "it's going to be really hard to scale further than this".
Napkin math says the team in the next building over at Google is routinely running experiments that use roughly 20 billions time more compute (90 days on 25k A100 GPUs).
I'm sure they could, at least, leave the trusty laptop on for a week end?
> Only way I could see this being surprising is if I was an 18 year old intern who has never played with something like this before.
Is this just like running the game of life [1] on a giant random board and seeing some generators emerge? Or something actually significant?
"This research is far from the first attempt at mimicking life digitally: for example, simulations such as the Game of Life, which has a grid of cells that are either “alive” or “dead” and are governed by simple rules, have shown self-replicating behaviour. Laurie says what makes this work unique is that the system had no formal rules, goals or processes to encourage or kick-start artificial life – it simply arrived. “It all fizzes around and then suddenly: boom, they’re all the same,” he says."
Still not sure if that means it's significant or not. Cool none the less though.
"This research is far from the first attempt at mimicking life digitally: for example, simulations such as the Game of Life, which has a grid of cells that are either “alive” or “dead” and are governed by simple rules, have shown self-replicating behaviour. Laurie says what makes this work unique is that the system had no formal rules, goals or processes to encourage or kick-start artificial life – it simply arrived. “It all fizzes around and then suddenly: boom, they’re all the same,” he says."
Another user posted a link to the full (non paywalled) article, it's worth a read. It also presents a few critic views.
Many others mentioned here: https://en.wikipedia.org/wiki/Life_simulation_game
There are older games (don't have my steam account on my work computer) that have much simpler starting states but they are abandoned or just not worth the price.
"I wrote a program to evolve life" is a super popular toy programming challenge as well, and you can find good and bad examples on youtube.