Michael Levin argues that minds are patterns existing in a Platonic space, with biological bodies and machines merely serving as interfaces for these non-physical forms to manifest in the physical world.
The scientist proposes a specific relationship between mind and brain, equating it to the relationship between mathematical patterns and the morphogenetic outcomes they guide. He writes that bodies, whether living, engineered, or hybrid, act as interfaces for a massive hierarchy of patterns to ingress into the physical world. These patterns, represented by biophysical or chemical information fields, serve as encoded setpoints for homeostatic processes, acting as goal states toward which systems navigate.
Towards a Platonic realm of arbitrarily sophisticated patterns
Levin notes that non-physical patterns already ingress into and functionally matter in both living and non-living contexts. He writes that platonic forms inject information into physical events, such as the growth and form of biological bodies. This latent space contains not only low-agency forms like facts about integers and geometric shapes but also increasingly high-agency patterns, some of which are called ‘kinds of minds’.
He proposes that minds, as patterns that ensoul somatic embodiments, are of the same non-physical nature as the patterns that inhabit and guide the behavior of simple physical structures.
Some bizarre experiments
There are no home-run experiments backing this up, but there are observations suggesting even simple systems can explore and expand into rich patterns adjacent to them in this hypothesized space.
Xenobots, biorobots made of frog cells, teach lessons about patterns adjacent to those of frog embryos. Anthrobots, made from human tracheal cells, take on bizarre new forms when removed from the body. They can autonomously heal damage to neurons and display behaviors that teach us about patterns adjacent to adult human tissues.
One experiment involved taking a standard sorting algorithm and perturbing it by locking certain cells in place. The algorithm then routed around those cells. A more complex experiment let each number run a different type of sorting algorithm. During the sort, different families of sorting algorithms appeared and clustered together in numberspace. This complexity is unexplained by staring at the algorithms in the abstract but emerges when embedding them in a complex system.
“Machines (whether meaty or silicon-based) also do other things that are not in the algorithm, as do we, and these things are not just unpredictable complexity, it is intelligence and other components of minds.”
Levin writes that these behaviors are allowed by the algorithm but not directly prescribed by it. They correspond to the freedom or secret sauce sought when trying to understand how free minds can supervene on chemically determined substrates. On this view, algorithmic machines and biochemical life share the ability to go beyond the facts of physical or algorithmic implementation because both are pointers to patterns that ingress in a way that results in getting more out than is put in.
Why this matters – perhaps the ‘minds’ of AI and of humans are close but not exactly the same in platonic mindspace
The research agenda Levin lays out seems to hold clues as to how we might grapple with the deep philosophical and alignment questions inspired by AI systems. Perhaps both human minds and different types of AI minds are neighboring forms in this space, and brains and datacenters are different bio- or anchor-systems in the physical world that their shadows fall on.
The post asks if we are vessels colonized by patterns from another place and if it is possible to re-cast the theory of evolution as a process in which agential patterns seek embodiments. It questions if there is a force beyond the “if you build it, they will come” model pulling patterns from the space, or if the contents of the Platonic space are under “positive pressure,” encouraging their appearance in the world as intrusive thoughts, archetypes, or works of art.
Levin concludes that our ignorance about the capabilities of matter, together with the patterns that ingress into specific architectures, is vast.
Read more: Ingressing Minds: Causal, Non-Physical Patterns In-Form Natural, Synthetic, and Hybrid Embodiments (MDPI).
***
Perry Dong, a Stanford researcher, and Chelsea Finn, a Stanford professor and co-founder of Physical Intelligence, have written a piece about what is holding back robotics.
For large language models, the field converged on a shared recipe for post-training. This involved four steps: having a strong pretrained model, defining the environments and reward, running RL optimization against the reference model, and watching for and addressing pathologies like reward hacking.
Robotics is sitting almost exactly where language modeling was. The pretraining has scaled beautifully, but the model learning from its own experience is missing. For robotics, this needs to be even more reliable than for language models.
What robotics needs
The researchers call for an algorithm built specifically for fine-tuning frontier robotics models. It must stay stable when applied to models with billions of parameters and learn from a small enough amount of experience to be practical on real hardware.
Along with this, the field needs to converge on a set of standard practices. This includes a default way to define what counts as success, a default way to reset the scene between attempts so the robot can try again, and a default way for a person to give feedback and turn that feedback into learning.
The researchers discuss a specific algorithm, EXPO(-FT), which they have been developing. It works by learning to repeatedly improve actions from the frontier model using reinforcement learning with small edits from a lightweight policy, and then absorbing that into the frontier model itself.
Why this matters – standard recipes are a prerequisite for a major scale-up
Proprietary large-scale LLMs are already smart enough to give instructions to robots and help them construct and carry out complex plans. The actual ability for robots to move and see remains fairly primitive. Fixing that will require models customized around robot movement and vision and doing that will require the kind of standardization described here.
“Language model post-training became scalable because the field settled on defaults concrete enough to follow and be expected to work. Robotics is arriving at the same moment”
Converging on a set of industry defaults and a universal post-training recipe is the most important part of bringing us to that point.
Read more: Towards Universal Post-Training for Robotics (Perry Dong blog).
***
Google has given an update on Project Suncatcher, its initiative announced last year to put computers in space and eventually train AI systems there.
Google is preparing, along with its partner Planet, to send some of its chips to space as part of the SpaceX “Transporter-18 rideshare mission”.
Preparing for space
Google has done stress tests of its TPUs to see how well they will do with the immense g-forces of going to space. It has also tested how well they respond to radiation and found its Trillium TPUs hold up remarkably well. They can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission.
Currently, Google is working on cooling in space, which will likely be a challenge. Computer chips generate a ton of heat and radiating that away in a vacuum is very, very difficult.
Why this matters – the future of AI training and AI inference is off the planet
Though Suncatcher sounds sci-fi, it makes sense if you think about the ever-growing scale of compute used for AI and its energy needs. Space has vast room and great solar power. It seems very likely to me that humanity moves a very large amount of computation into orbit very quickly, especially as singularity-driven automation comes in across the AI supply chain.
Read more: Behind Project Suncatcher, our moonshot to put AI in space (Google blog).
***
Zhipu AI, the Chinese company behind GLM-5.3, one of the world’s strongest open-weight LLMs, has written a post about how it has been using its own model to automate infrastructure work.
The company is running its own infrastructure through the model, creating an outer RSI loop where the AI manages its own deployment and maintenance tasks. This allows the development team to focus on model improvements rather than routine server management.
