Hybridation
How can a mushroom talk to a robot? What's it like inside the brain of a fruit fly?
We live in an age of unprecedented frontier interactions.
With unparalleled intelligence and computing available at our fingertips, there seem to be an obvious avenue for new natural science. Just collect more granular data, feed it to the machine along with some interesting macroscopic variable you can observe, wait and let the codebreaking beast do its thing.
Whether it’s by mapping the genome, the proteome, or the brain’s neuronal wirings (the connectome), biologists are pursuing large data collection in an effort to pattern-match at scale. As more data is gathered and classified, AI can help discover exotic regularities over large those large complex networks. Every interaction, co-occurrence, or interference that is statistically significant would be tracked down.
Is this really all there is to it? What about our slow, explainable, story-based, very human hypotheses? Will they become a thing of the past? Or will our expressive means evolve with our understanding?
🍄 The mushroom that could control a robot
The March of the Mushroom Robots (Nautilus). Machines have always been designed with inspiration from the living kingdom. This time researchers are taking it one step further by growing mycelium into the electronics of a robot.
Turns out living beings are excellent sensors. While our conscious systems elaborate information at a very slow pace, our peripheric sensory systems process raw data very efficiently. What if we could plug this rich input stream into a robot’s brain?
Enter the field of biohybrid robotics. Plants and animal cells may be hard to work with for now, but fungal mycelia is highly sensitive to light and chemicals. It’s been described as a single, long, wiry nose.
In August, researchers from Cornell and Florence Universities published promising experimental results (Science Robotic). Their prototype has an electrical interface that interprets the mycelia’s activity and translates it into signals to move the robot parts.
When the mycelia detects UV light, it reacts. Then, the robot picks up the surge of electrical activity in the mycelium and starts moving.
Wouldn’t you say it’s now the mushroom that controls the robot? What if it could start learning from the outcomes of the integrated system, made of the environment plus the robot’s program? What if it adapted, modulating its responses to command its robotic extension?
🧠 What’s inside the brain of a fruit fly?
Neuronal wiring diagram of an adult fruit fly brain (Nature). Not unlike mapping the genome or the proteome, mapping the set of all neuronal connections (the connectome) in a brain is gargantuan task. It took over five years and hundreds of scientists expert annotations, even leveraging AI-augmented processes.
The research of the FlyWire consortium resulted in “a complete map of the female fruit fly brain” - the biggest and most detailed to date 140,000 neurons, which connect through 55 million synapses to permit vision, movement, and memory formation.
Scientists classified over 8.400 unique neuronal cell type, specified by their connectivity quality (“rich club, broadcaster or highly reciprocal”), or their superclasss role (afferent bringing information in, intrinsic elaborating information, or efferent bringing motor signals out).
Now, scientists can access not only wiring diagrams of individual neural circuits, or sinapse-level connectome, but also their aggregation, to map projections between brain regions. Such maps are called projectomes or mesoscale connectome. These structures are believed to be key for tracing complete pathways from sensory inputs to motor outputs.
It’s a brave new world for neuroscience.
💉The story your plasma proteins want to tell you
Longevity researchers have more data than ever to work with. Recent advances of proteomic mapping capabilities allow scientists to sample thousands of proteins from tiny blood samples. If you think it sounds just like what Theranos had promised its investors and feel skeptical, you’d be justified.
However, the number of protein target we can map and identify has been growing steadily. These results allow for better understanding of human aging process, identified many organ-specific decay paths (“organ clocks”) and how they may be favourably modulated.
This has the potential of using proteomic scores for assessing risk of various diseases. However, costs are still elevated. A high-throughput proteomic assessment ranges from $500 to $1000 per individual. The service is carried out by SomaLogic, which was acquired in Jan ‘24 by Standard BioTools (Nasdaq: LAB 0.00%↑ ), previously known as Fluidigm Corp. in an all-shares merger. Each SomaLogic stockholder got 1.11 of the acquirer’s common stock for each of their shares.
The proteinomics race advances.
🧩 Industrial-scale mathematics
We're Entering Uncharted Territory for Math (The Atlantic). Award-winning mathematician Terence Tao has recently described GPT-o1 as a moderately “competent graduate student”. What did he mean?
Tao described a kind of AI-enabled, “industrial-scale mathematics” that has never been possible before: one in which AI, at least in the near future, is not a creative collaborator in its own right so much as a lubricant for mathematicians’ hypotheses and approaches.
Some years ago, AI surprised the mathematics community coming up with a faster algorithm for matrix multiplication (Quanta). DeepMind researchers formulated it as an adversarial puzzle and unleashed AlphaTensor at it. The Chess world-champion AI rediscovered mathematical techniques (Strassen’s algorithm, 1969) from first principles and further innovated on them.
What Tao highlights in his The Atlantic interview, though, is slightly different:
“The type of math that I’m most interested in is math that doesn’t really exist. The project that I launched just a few days ago is about an area of math called universal algebra, which is about whether certain mathematical statements or equations imply that other statements are true.”
He points out the challenges of collaboration in mathematical research. Every sufficiently advanced mathematics is an island of its own. Very few mathematicians can understand and proof-read a colleague’s work. It’s rare for projects to count more then five academics.
Universal algebra is a fascinating project, launched by Tao’s blog post (“A pilot project in universal algebra to explore new ways to collaborate and use machine assistance?”) at the end of August.
It studies the most possibly bland algebraic object: a magma, which is just a set A, with a binary operation described from AxA → A. No further constrains are imposed on the operation (such as inverses, identity, associativity, commutativity). Then it asks the question: which of all the axioms you can formulate with up to four operations are logically implied by another axiom?
There are 4694 equational laws involving at most four magma operations, hence 4694*(4694-1) ~ 22m implications that need to be proven (or disproven).
The project is now up and running on GitHub, with contributions of over 30 mathematicians. It employs transitive rewriting or Vampire an automatic theorem prover for first-order logic from University of Manchester.
As of yesterday, Oct 6th, it is 99.995% complete with only 1001 open implications to be proven or disproven.
Quite the result for little more than a month’s work!
🤝 Thought-partners needed
It takes two to think (Nature). Artisanal human-made science, meanwhile, is very much a conversation game.
Itai Yanai, biomedical scientist at NYU, writes a rather romantic piece defending the thesis that the best creative chats are the ones where you reason with just another person. Too many peers around the table and it becomes a brainstorm, or worse an exercise of groupthink.
It seems reasonable that good science needs intimacy. Vulnerability is key for new ideas. A “Yes, and” mentality is what opens the door of creativity and collaboration. But I disagree that only two can make for a good chat. The evidence presented does not seem conclusive of a categoric result. I believe it all boils down to trust and unique viewpoints. When you bring in new further contributions that are not just pro or against something but genuine, trusting, unique viewpoints, why would they limit the scope of your conversation?
I’ll leave you the article below so you can enjoy it, and judge for yourself.
👩🍳 The curse of the limelight
The curse of the Michelin star (Economist). On another planet, economists are researching the adverse effects of Michelin stars to NYC restaurants with linear regressions.
Are we better off the spotlight? Restaurants that are awarded the star are apparently more likely to fail. Being in the limelight seem to raise everyone's expectations. Employees, suppliers, customers and landlords all turn greedier. Expectations are the true happiness killers. After all, your happiness is only measured against your expectations’ baselines.
Daniel B. Sands from UCL “found that, of this promising group, establishments which went on to get a Michelin star were more likely to close down in the years that followed than those which did not. The relationship remained even when factors such as location, price and type of cuisine were taken into account. All told, 40% of restaurants awarded Michelin stars in 2005-14 had closed by the end of 2019.” (Strategic Management Journal, 6th August 2024). The study was limited to establishments in New York City. Could this be a specific feature of its market?
I have mixed feelings about this result. On the one hand, I like it. It seems to confirm a very narratologic truth: after every excessive success, comes ruin. Not very far off from some Eastern wisdom! It also echoes protestant ethics, which looks with suspicion at any man-awarded gratification. Don’t stop working, they’d say, salvation is only granted from above. Never think you’ve made it.
On the others, the methodology and the sample seems limited. What could the story of a few hundreds (276!) restaurants tell us? The findings are valuable yet anecdotical.
The author’s find that third-party evaluations are double-edged is commendable, yet what isn’t double-edged?
Mind celebrating your wins too soon,
David