Your experience of the next solar eclipse, on 12 August, will change dramatically with latitude and observing location. With the help of AI and using NASA eclipse data, I simulate it across Italy: my birthplace in the south, the city where I now live in the north, and, of course, the capital.

Given the news from openAI, Anthopic, Meta and so on, I had to make this one. It was a moral duty.

Given the news from openAI, Anthopic, Meta and so on, I had to make this one. It was a moral duty.

Look at this one, just some seconds before crossing the sun light.

Only during July 2026 the burnt area in EU due to mega wildfires has been enough to match the worst year in record (2025). And that's almost 2x the average over the past 20 years. In Sicily, almost 40 years ago, I used to grow with wildfires during summers. Now it's the whole EU 😔

Finalizing the 4th chapter of my series on decoding the architecture of living systems from this beautiful corner of planet 🌏 I hope you can have a nice day too.

And this leads to the next question: How can living systems organize many interacting loops w/o allowing every local event to become a global one? This will bring us to modularity and hierarchy in the next chapter. Enjoy! /fin

This reveals another trade-off. Loops can generate robustness, but too many connections can overcouple the system, allowing local perturbations to propagate globally. Robust architectures must remain connected enough to sustain function, but localized enough to contain damage. 7/

So a loop is not useful simply b/c it exists. It becomes architecturally relevant when it sustains a biologically meaningful process at an affordable energetic, material and informational cost. 6/

We must also ask what the loop costs. Redundancy requires additional enzymes and substrates. Feedback requires sensing, regulation and appropriate timing. Productive closure requires continuous flows of matter and energy. 5/

These mechanisms share a “loopy” topology, but they do not perform the same function. A loop may redirect matter, regulate a signal, stabilize a variable, store memory or sustain the components that keep a process operating. Topology alone is not enough. 4/

In metabolism, an alternative route can preserve flux when one reaction fails. In sensory systems, feedback can restore sensitivity after stimulation. In autocatalytic systems, molecular products can help regenerate the processes required for their own production. 3/

A possible answer starts from loops: minimal structures that can support redundancy, feedback and productive closure. But a loop in a graph is only a structural possibility. Its biological function depends on what moves through it and on the dynamics it generates. 2/

Is this the effect of AI? Same class, same 59 students: mean score 95.7 → 48.8 when assessment moved from non-in-person to in-person. This does not establish causality, but it is worrying. Is generative AI already changing what exam scores mean?

Next-chapter spoiler: if robustness cannot come from adding links without limit, which structures make it possible? /7

This shifts what “architecture” means in living systems: a constrained generative process, compact enough to be transmitted, flexible enough to tolerate variability, stable enough to preserve function. 6/

So inheritance is less like copying a finished object and more like reconstructing a constrained process. A brain, immune system or regulatory network can be reproducible in function, while remaining variable in microscopic detail. 5/

The answer is that evolution does not need to write every connection. It can transmit generative rules, then let development, physics, chemistry, feedbacks and constraints unfold them into a functional structure. 4/

This is the information-storage paradox: If finite genome + costly information + astronomical connectivity = no explicit blueprint, why do living systems reliably reconstruct functional architectures? 3/

The tempting answer is: in the genome, as usual. But scale makes this answer problematic. The human genome contains ~3 billion base pairs, while the brain has ~86 billion neurons and ~10¹⁴ synaptic connections. No genome can explicitly list the full wiring diagram. 2/

Thanks heavens, it's eventually raining. #Heatwave #ClimateChange

To understand functionality, we have to follow perturbations, flows, time scales and the diversity of possible responses. Following-up my latest seminar, invted by @kayson.bsky.social & Petra Vertes, I summarize the main findings and the discussion here: manlius.substack.com/p/what-netwo... 1/

The Internet is an emblematic example that robustness is about preserving function while perturbations keep moving and propagating through the system. Topology gives a map, while functionality drives the pulse. Upcoming #ComplexityThoughts: manlius.substack.com/p/what-netwo... #ComplexSystems

Perhaps because evolution does not need to write down every connection. It can operate on rules, constraints and circuitries that make some organizations more accessible than others. That is the starting point, and it would be great to hear what you think about this during the journey. 7/

In this series, I will use “architecture” in this more precise sense: regularities of organization that sit between microscopic details and macroscopic behavior, shaped by constraints rather than specified by a central blueprint. 5/

But selection acts on substrates: regulation, memory, signaling and adaptation require circuitry. They require matter arranged so that information can move, perturbations can be absorbed, and responses can remain coordinated under changing conditions. 4/

Why living systems often display architecture w/o being designed like machines? A reductionist view misses that many system-level properties live in interactions rather than isolated components. An adaptationist view risks treating selection as if it were already an implementation mechanism. 3/

I know, the word is risky b/c it suggests design: a building has an architect and a machine has a blueprint, while living systems have neither in that sense. Yet they are not amorphous collections of parts: a cell repairs itself, an immune system coordinates molecular and cellular responses. 2/

MINERVA is moving forward, with the current snapshot: - 90 expert-group core entries - 50 expert-sourced additions - 270 community recommendations (45 unique candidate papers under evaluation) Read more and contribute: manliodedomenico.com/complexity_m... #ComplexSystems #NetworkScience