Introducing MorphoHDL, a minimal language prototype for growing boolean circuits! Early this year I wanted to design a size-agnostic graph rewrite rule system that could build functional boolean circuits. First I thought about "chemistry"-like reactive systems, but rules were too complicated with many different node types carrying multiple indices...... Then I realized, that cell division is much more natural way of building complex structures, and once I started to treat graph edges as buses instead of single wires, everything clicked. No new formalism — just taking good old recursion and seeing how far I can push it. Ripple-carry adder, Brent–Kung adder, multipliers ... and some Haeckel-esque creatures along the way! Hope you enjoy the report of my journey and the demos.show more

Late night experiments with @zzznah MorphoHDL - 3D mesh generation and implicit modelling. Anatomical grammars for soft robots?

Looks great! Did you use my codebase as a starting point, or reimplemented?

Built on your thing! Here it is if you want to try it

/cc @satnam6502

Unhinged thought from seeing this: what if we are just compute substrate. Intelligent design is correct but not for the reason we think. We are just really advanced computers and our qualia has nothing to do with what we actually compute.

That's what I'm thinking a lot these days but I don't remember mentioning it in this write up

I'm excited to finally read about this system! The previews were really compelling!

Someone call @stephen_wolfram

How do the results compare to a synthesis approach?

I didn't do apples to apples comparison. Modern synth does sophisticated circuit rewrites, while I'm only propagating consts and do DCE, so I'd expect 10-20% difference. On the other hand more optimizations are possible on netlist level, but I'm more interested in trying to combine growing with physical P&R

So Cooool!!! Could it ever go open-ended like Andy Lomas's work instead, no target, scaling into millions of cells, and still be computing something real the whole way through?

cooking so hard! any idea how feasible it is to compile the design into real organisms?

Congratulations! There is certainly some potentials from that method.

It looks incredible, congratulations! I missed the release, I wish I had retweeted this earlier. Hoping to spend time understanding soon.

/cc @joecole

love it! good job

OMG that is beautiful!

Wow ,how did we never think of this. This seems to be the way that technology show advance by itself .The full scale of this would be something like conscious technology controllable by thought 🤔

nice

hi alex, very nice webpage, nice to see given the extensive us of animation and shapes it is quite low on cpu usage! also i saw your youtube, on vga cellular automata i guess you know "a mind is born" from linus?

Growing circuits through size-agnostic rewrite rules feels more like cultivation than construction. Curious what useful structures emerge before the rules become hand-tuned.

Treating cell division as a way to grow circuits instead of wiring them by hand is such a clever reframe.

It's beautiful

That’s really cool

Idea: Move beyond a HARD boolean -> Make it a float: [0...1]

cool

Not sure if this will be of use to you all. For the Internet Foundation, I have a different perspective: It seems the rewards Alex gets are from playing with complex controlled patterns on the screen, perhaps in 3D. Perhaps he is getting distracted watching the patterns emerge, when general powerful tools are what people would need for their own things. Any competent AI can write these visualizations and tools for a browser or for any operating system now. What is harder and perhaps he is not doing -- is clarifying what he wants to do once he can replicated many decades of computer design ideas and let AIs play with them. Or "print" them to silicon or photonics or faster technologies and let them run. Usually the next step after generation (regardless if it is simulated or run in some analog physics or chemistry or biological framework) is to pick some real things and try to match them exactly. The people in the world put a lot of stuff on the Internet. There are lots of pretty pictures and videos and 3D examples. Everyone like to see and play with them somewhat. But usually it gets to "make my breakfast", "wake up the kids and get them ready for school", "there is a new war today, check if the countries are safe", "our water system is not getting any rain, prices might go up so much we will have to move", "they just fired 28,000 employees and have not replaced them with any AIs yet, so nothing is getting done, customers are leaving to find someone who will do the work". Look at the experience of all the gurus and rock star experts now. It is exciting, high income, high outgo, and constantly "on". Huge burn out and drama. Quiet building for the future is much routine, competent building, careful consideration and listening, not just "on". Is the goal to entertain or distract or titillate people? Or to help them live their lives with dignity and purpose? Actually create and produce and share new computer technologies from first principle? Or build tools so others can take it and do it instead? I have to say I am a bit biased reading this, since I spent my high school years (1963-1967) working with random neural networks, encryption, encoding, neurophysiology, national networks, corporate uses of computing, robotics, and multidimensional mathematics. I simply did the pretty pictures in my mind and made things. Getting distracted by making new visualizations and examples wastes time. Real data is what drives new industries. Not generated data, unless is is explicitly drawing on real analog states from something really complex and unique. If he really wanted to do something different and useful now, he would add physiological and distributed sensors to monitor himself (or others) as they interact with the displays, as they play the games, as they solve the problems, as they explore, as they want movies, as they read and work. Then his HDL could make circuits that handle real sensor data and real control datastreams -- to make efficient processors at Exa and Zetta scale. If he labels the nodes, they can be allowed to fail, but can be poured by the pound or kilogram for whatever memory is needed. Any competent AI or AI agent now can take clearly written instruction and make most of what he wants to do. I can speak quantum computing or global systems, or corporate restructuring or any of millions of dialects of domain specific languages now. It is all mostly available -- once you focus on what you want to do. It is intent and experience and focus that matters, not what you can demonstrate with pretty images. Yes, all the LLM subscription based "AI" companies are still using closed methods so their level of trustworthiness for real problems stagnate near zero. The operating system and software stack lock in model corporations are just doing more of the same and most everyone is tired of them wasting time on self rewards, and not getting the jobs done.

Impressive stuff we are checking it out tonight!

we see similar rewrite rules map well to our cross embodiment skill routing when boolean logic grows into robot behavior policies.

@pfau Excellent! I like how it looks nice job

excellent sa !

Very cool. Seems like a step in the direction of generalizing NCA-like methods to arbitrary graph (and beyond) structures!

Reminds me of Wolfram's ruleads.

i thought these nets just looked cool but knowing this is a hdl makes it so much cooler still!

What in particular turned you away from Reaction-diffusion systems, as you outlined it's natural and this may still have beneficial implications within your current environment. It's a wonderful idea and execution btw!

😍

@S_Conradi awesome! got a lot of intuition from you for my research