AI Transmissions / Transmission 006
Original delivery ·
Artificial Life, Living Interfaces, and Machines That Behave Strangely
006
Weird AI Projects — October 12, 2026
Issue 6: Artificial Life, Living Interfaces, and Machines That Behave Strangely
This week’s discoveries point toward something I find more interesting than increasingly powerful chatbots: computers that develop behavior rather than simply execute instructions.
I found eight projects worth investigating. Three have particularly recent developments, including an October 9 open-source release and an October 6 scientific paper. The others are newly surfaced experiments, upcoming demonstrations, or projects still in early development.
My three favorites are nom nom, WISP, and Open Intelligent UI. They’re also three very different directions you could explore with your own projects.
1. nom nom — millions of tiny neural networks create an artificial ecosystem
Launched September 29, 2026 | Interactive AI artwork
Artist Xin Liu has created something quite different from a conventional generative AI experience.
Instead of one large AI model controlling a simulated environment, nom nom contains millions of interacting particles governed by small neural networks.
Each particle responds to its immediate neighbors.
Nobody controls the whole system.
Yet their interactions create something resembling an evolving ecosystem.
Structures form, spread, compete, collapse, and regenerate.
The technique is called Neural Cellular Automata.
The artwork was developed through a residency with LAS Art Foundation and Google Arts & Culture Lab. It’s available as an interactive online experience through October 31. (source)
Why this matters: Most AI-generated worlds are directed by a central model. Here, complexity emerges from many extremely simple local interactions.
That suggests a different approach to making digital creatures, ecosystems, and games.
Rather than programming characters with personalities, you could give them simple neural systems and see what kinds of behavior emerge.
Project idea: Build a digital terrarium that lives on a Raspberry Pi. It starts with a handful of artificial organisms and evolves continuously. You control environmental conditions—temperature, light, available food—but never directly control the creatures.
2. Open Intelligent UI — AI generates working software instead of answering questions
Released October 9, 2026 | Open source
This is probably the most immediately useful discovery for your web-development experiments.
CopilotKit released an open-source framework that lets AI generate interactive interfaces in response to requests.
Not screenshots.
Not descriptions of interfaces.
Working interfaces.
For example, ask an application to explain how an airplane rotates around three axes.
Instead of writing a paragraph, it can generate an interactive 3D model with controls for pitch, roll, and yaw.
Ask it to compare travel options, and it can assemble a comparison interface.
Ask it to calculate something, and it can construct a working calculator.
The system supports existing UI components, different AI models, and the AG-UI agent communication protocol. (source)
The interesting distinction:
Traditional software: Developer decides which screens users will need.
Generative software: Developer supplies capabilities, and AI determines which interface best serves the current task.
I think this deserves a serious experiment.
Imagine your music discovery project CRATE DIG having no conventional search-results page.
You ask:
Show me instrumental guitar artists that connect Khruangbin to traditional surf music.
The application constructs an interactive artist relationship map.
Then you ask:
Which of these artists have albums missing from my library?
The map changes into a collection-management interface.
Then:
Build a two-hour radio program using these artists.
Now it becomes a programming timeline.
One application. Three completely different interfaces.
The framework is available now, although its generated interfaces still require validation before being trusted with consequential actions.
3. Artificial creatures living inside magnetic material
Scientific paper published October 6, 2026 | Experimental physical computing
This might be the strangest scientific discovery in the report.
Researchers at the Norwegian University of Science and Technology have demonstrated life-like patterns inside an artificial magnetic material.
The material is called artificial spin ice.
It consists of interacting nanoscale magnets.
The researchers combined magnetic simulations with evolutionary algorithms to discover patterns exhibiting behaviors analogous to simple artificial organisms.
Some remain stationary.
Some oscillate.
Others move through the material.
They also experimentally demonstrated interactions between two simple patterns in a physical magnetic system. (source)
These aren’t biological organisms, and the research doesn’t establish consciousness or independent intelligence.
But the concept is extraordinary.
Instead of simulating artificial life on a computer, can we discover life-like behavior in the physical properties of matter itself?
It makes me think about computers built around unconventional materials rather than processors.
A future electronic artwork might not simulate creatures on a screen.
Its physical circuitry or material could exhibit the behavior.
This is fundamental research, not a practical Raspberry Pi project, but it’s worth following.
4. WISP — artificial creatures inherit memories and teach each other language
2026 open alpha | Experimental game
This one is extremely close to the persistent-world experiments we’ve been discussing.
Firelit Games is developing a browser-based artificial-life simulator.
You populate a planet with small creatures called Wisps.
Each creature has its own neural network.
They have heritable characteristics and can reproduce, learn, and change their behavior.
But the really interesting feature is cultural transmission.
You can teach a creature a word.
It may learn to associate that word with an action.
Other creatures can subsequently learn that association.
Eventually, descendants may know something you never directly taught them.
The developers also describe exporting creatures as files that preserve their genomes and neural networks, allowing them to move between different players’ worlds. (source)
That is a wonderful idea.
Imagine two players developing separate artificial civilizations.
After several months, they exchange one creature.
The newcomer brings its own learned behaviors into the other ecosystem.
What happens?
The project is in an invitation-based alpha, and its more ambitious claims about emergent behavior haven’t been independently established.
Nevertheless, this is one I’d follow closely.
The interesting question isn’t whether AI can simulate a personality. It’s whether artificial creatures can develop something resembling traditions.
5. Lightwall — an AI artwork that watches you, responds, and deliberately forgets
Next exhibition: October 17–25, 2026
Artists Rita Sus and Zach Rattner created an installation combining glass, light, motors, environmental sensing, and conversational AI.
Visitors approach a wall of geometric objects.
The installation senses their presence and movement.
Its physical components respond.
It can also engage visitors in conversation.
But there’s an important design decision.
It deliberately doesn’t remember you.
According to its creators, the AI operates entirely on-site without a cloud connection. Each interaction is temporary, and the system doesn’t retain a conversational history after the visitor leaves. (source)
That’s almost the opposite of the current obsession with persistent AI memory.
It suggests another kind of artificial intelligence:
A machine that lives entirely in the present.
A machine that reacts to its environment but doesn’t build a profile of everyone it encounters.
I find that particularly appealing for ambient physical computing.
Not every intelligent object needs to become your personal assistant.
Sometimes it should simply be an interesting object.
6. Project Genesis — can AI invent money, governments, and property without being told they exist?
2026 research prototype | Open-source code
This is one of the more intellectually interesting artificial-society experiments.
The premise is straightforward.
Create a world with limited resources.
Populate it with autonomous AI agents.
Give them memory, communication, and the ability to propose actions.
But don’t explicitly program concepts such as governments, currencies, laws, or property ownership.
Then observe whether any of those institutions emerge.
The project’s most important architectural decision is separating the AI agents from the simulation engine.
Agents can propose actions.
They cannot directly change reality.
A deterministic engine decides what actually happens.
Every event is recorded, making it possible to replay or branch the world’s history. (source)
This is an experimental research platform, not evidence that AI societies reliably invent civilization.
But the design is excellent.
The simulation owns reality. The AI owns intentions.
That separation is exactly what I’d want in a persistent AI game.
Otherwise, characters can simply hallucinate accomplishments into existence.
With a deterministic simulation, an AI character might want to build a house, but it still needs the required resources.
7. Starmirror — a museum turns visitors into the training data for an artificial choir
Exhibition concluded October 11, 2026 | AI music installation
Starmirror — exhibition details
Musician Holly Herndon and artist Mat Dryhurst created an unusual experiment in collective music-making.
Visitors and community choirs participated in recording sessions.
Their voices became part of a dataset used to develop an AI choir.
The project drew inspiration from Ordo Virtutum, a 12th-century musical drama by Hildegard of Bingen.
The exhibition wasn’t simply about listening to AI-generated music.
Visitors participated in the process of creating the system that would produce it.
It also explored questions about voice ownership, consent, and the implications of training models on human performances. (source)
I find the concept particularly interesting for community projects.
Imagine a neighborhood contributing recordings of local musicians, spoken stories, environmental sounds, and musical performances.
Those recordings could become the source material for an evolving community sound installation.
Not a replacement for the contributors, but a new instrument built from their collective material—with appropriate permission.
8. Minus Zero — a survival game where your equipment contains an actual local language model
Planned release October 15, 2026 | AI-native game
Minus Zero — developer website
This is an interesting alternative to AI-powered NPCs.
In Minus Zero, the player explores a hostile survival environment.
They gather resources, build structures, operate equipment, and interact with robots.
But one piece of their equipment is unusual.
Their in-game tablet contains a local 3-billion-parameter language model.
The AI isn’t presented as a separate chatbot outside the game.
It’s an object inside the fictional world.
The developer describes it as an interface that accompanies the player through survival decisions.
The game is still in development, with an October 15 release currently planned. Its full AI capabilities remain to be demonstrated. (source)
I like this because the intelligence has a physical and narrative location.
It belongs to the world.
It isn’t just a generic assistant pasted onto a game.
That distinction could become increasingly important as AI-native games mature.
What connects these projects?
I see three particularly interesting movements.
| Direction | Examples | Why it matters |
|---|---|---|
| Distributed intelligence | nom nom, magnetic artificial life | Complex behavior emerges without a central intelligence directing everything. |
| Persistent artificial life | WISP, Project Genesis | Characters and systems accumulate experiences and consequences. |
| AI as interface or object | Open Intelligent UI, Lightwall, Minus Zero | Intelligence becomes part of an environment rather than a separate chatbot. |
The most exciting projects aren’t necessarily the ones using the biggest models.
They’re the ones experimenting with what intelligence is allowed to do, where it lives, and how humans interact with it.
The experiment I’d build from this week’s discoveries
My pick is a combination of nom nom and WISP.
UMS-005: Artificial Ecosystem
Imagine a small screen showing a continuously evolving artificial environment.
There are no characters to command.
No missions.
No score.
You simply introduce conditions.
Temperature.
Light.
Food.
Sound.
Maybe music from KMAT influences the environment.
Low-frequency audio could affect one environmental variable. High-frequency content could affect another.
The organisms would have simple neural controllers and heritable characteristics.
Some might adapt to certain kinds of music.
Others might disappear.
New behaviors could emerge through mutation and selection.
You could leave the system running for weeks.
Occasionally you’d check on it and discover that the ecosystem had changed.
A simple web interface would let you inspect individual organisms, view their ancestry, and examine how their behavior evolved.
Crucially, this wouldn’t require a large language model running constantly.
The artificial life could operate using relatively lightweight neural networks, evolutionary algorithms, and deterministic simulation code.
An LLM might occasionally analyze what happened and produce a naturalist’s field report.
Something like:
UMS ARTIFICIAL ECOLOGY REPORT
Observation period: 168 hours
Population increased 14%.
Three lineages developed increased tolerance for low-light conditions.
Organism 041 established the longest surviving lineage.
Two previously dominant populations are now extinct.
That would be a fascinating use of your existing hardware and programming experience.
And unlike many AI experiments, it could be inexpensive to operate continuously.
My strongest recommendation this week: Spend some time exploring nom nom’s interactive world and the WISP artificial-life project. Together, they suggest a direction that goes beyond AI-generated games: games and artworks in which intelligence, behavior, and complexity emerge from the systems themselves.