Below is an email exchange I had with Stephen Wolfram about Strange Binary in September '20.
Hi Stephen -
You changed and enriched my worldview when I read NKS over ten years ago in college, and I'm excited for your latest project. I imagine you don't have time to answer emails like this, but wanted to share a side-project, and ask three questions.
Strange Binary is a 'game' that allows you to explore the computational universe in an aesthetically pleasing way. It's not clear what the goal of the game is for the player - but it has the side-effect of building some small amount of computational intuition in the player.
Essentially you navigate through 2D toroidal cellular automata, by toggling the system rule up or down randomly in 'dynamism' (a parameter that reflects how many system rules mandate a change).
As you navigate, the system state (ie 'on' cells) is ported into the new system. This is aesthetically fun, but probably un-useful. It's sort of like having objects that couldn't have formed in universe A show up in universe B.
If you have a chance to read this, or look at Strange Binary, I'd love to know if you have any heuristics for thinking about how to make discovered systems useful beyond fun and aesthetics. From the point of view of building intuition, what other system parameters would you expose to players? Have you ever trained something like a neural net with many steps of system data - with a view to something like pockets of reducibility across similar systems?
Thank you if you were able to read!
Myles
Myles ---
Thanks for being in touch, and apologies for a slow response…
About your game. I think it’d be useful to be able to see the effect of a “simple seed” for each rule. It’d also be nice to see the “spacetime” structure, rather than just the 2D+time version movie. I’m not sure that your universe A -> B thing is bad; it’s kind of reminiscent of successive layers in neural nets.
Plucking systems from the computational universe and having them be useful is, I think, a core challenge of future technology. In a sense, my lifelong effort on computational language is about making a bridge between human goals and what’s computationally implementable. But how best to go from “computation in the wild” [e.g. in your game] to things we humans find “useful" is a big (and important) challenge.
We’ve done quite a few experiments on finding pockets of reducibility using neural nets. The results haven’t been too impressive so far. Though as I think about it, I don’t believe we’ve really done cross-system studies of the kind you’re suggesting. In other words, are there “generically useful” abstractions/reductions that work across lots of systems? Or, conceivably, if you have an “oracle” for one neural net, that can compute with one system, can “just the neural net” be useful for another system? But viewed this way, one’s approaching the question of whether inside a neural net one can have something like a CA. And usually the more sophisticated the computational behavior, the more difficult it is to train. You should try to figure this out!
Looks like you’re quite busy with tech projects, but I’d encourage you to consider our Summer School if you can spare the time. By the way, rather to my surprise, our physics project may provide a new, very practical approach to blockchain-like systems … so there may yet be a way to merge your tech and science interests….
Creator here. There is no defined objective. Perhaps that disqualifies it from being a 'game'. One hope I have had is that the game could be used to discovering interesting (turing complete or otherwise) rules that may have information processing capabilities beyond the game. Vague I realize. The timeseries at the bottom is just measuring the number of 'on' cells.
I understand that it's visually unpleasant (not to mention potentially dangerous for epileptics). That said it's literally just a rendering of what the system rule dictates. Lower dynamism systems tend not to be as flashy.
> This kills my back button. What are you storing in the URL?
The URL stores the system rule - the idea being that you can share it. The URL changes over time as you play because playing edits the system rule.
J and K toggle down and up the 'dynamism' in the system. It's semi-random in the sense that you can't control what new system you get when you press J or K, but you will get a system that has either less or more dynamism than the current system. Dynamism is a property of the system - roughly how often 'things change' per the system rule. EG a conditional in the system rule that necessitates an off cell becoming an on cell is a 'thing changing'.
I built the first version of Cusp about a year ago. It was a tool to help people host fasts (as in not eating).
Quickly I came to the conclusion that ephemeral micro-communities are bigger than just the fasting use-case.
Cusp is exploring ephemerality in communities. Can we have more meaningful real-time connections when the experience is ephemeral by design? Can we have hyper-granular topics when the experience is ephemeral by design?
I also wanted to explore the thesis that micro-community curators/hosts should be able to get compensated for the value they provide. Thus you can optionally charge people to join groups.