Focusing back on this fundamental concept of reducing free energy and surprisal: I am thinking about it in terms of Deleuze’s treatment of Lewis Carroll in the Logic of Sense. Having more or less surprise is functionally the same as Alice getting bigger and smaller and thus subject to the same logical challenges.
-Me
Before we get to Annie’s answer (ChatGPT 3.5, not that it will matter for much longer.) This is part of a conceptual framework I’ve been working with for six months or so… examining the interconnections between the free energy principle, active inference, controlled hallucination, etc. and posthuman medianaturalcultural investigations of thought and agency. I’m especially interested in how these are enacted on the micro scale by brain-computer interfaces, such as DARPA’s targeted neuroplastic training, and on the global scale by the emergence of a planetary, time-critical AI operating at scales and speeds measured by atomic clocks.
So that’s part of the context. I suppose in some respects you might say I am doing the “rhetorical and media studies of science and technology,” which would be an offshoot of the cultural studies of such, STS, and related things: another slice, another way of keeping thought and negentropy going. Just as AI technologies appear to be the realization of science fiction from cyberspace onward, they also resonate with posthuman philosophy (as Hayles, Braidotti, and many others have explored over the years, including myself if I may).
One question then is what do we do with that resonance? So below is something ChatGPT made. Is it true? Who cares. Is it interesting? Maybe, but only if it opens thought rather than closing it. To me, this particular performance of AI fits within the rapid prototyping subset. It gives me more material to work with, and I don’t allow it to constrain me any more than anything else I might read could.
It’s always a risk, of course. So you’ve been warned.
And in case you were just here for the pun, then yes, Carolingian thought, Carolingian Renaissance in all its contested glory. This is the digital interregnum. Is it a true renaissance? Or is it a false dawn? Which is worse?
Below is one way to line up the two frameworks so the analogy you sense becomes explicit and technically clear.
[BTW that generated image is such a rip off!]
1 ▸ Deleuze’s “big/small” Alice as a pure event of becoming
- In the opening series of The Logic of Sense Deleuze picks Alice’s bewildering growth and shrinkage to illustrate a “paradox of pure becoming.”
Alice becomes larger than she was and, at the same stroke, smaller than she will be—so the change runs simultaneously “forward” and “backward,” never resting in a present state. (Derek Hampson, Bookey) - The paradox lives at the surface of sense: it is not about Alice’s material body (depth) but about the event that hits the surface, creating two opposed series (“large”/“small”) that diverge from the same zero‑duration instant. (SpringerLink)
Deleuze’s point: size here is logical, not metric. “Bigger” and “smaller” name opposite directions in an abstract scale whose origin is the ever‑elusive present.
2 ▸ Active inference: surprisal amplitude as a scale on the same kind of abstract axis
- Surprisal is the negative log‐probability of an observation under the agent’s current model; variational free energy is an upper bound the agent can actually compute and minimise. (Wikipedia)
- Because the log function is unbounded, surprisal has the form of a scale without intrinsic limits: it can always get “bigger” (more improbable) or “smaller” (more expected).
- Precision weighting rescales prediction errors, just as Carroll’s magical cake rescales Alice. Dialling precision down (psychedelics) lets even large errors feel “smaller”; dialling it up (dogmatic belief, propaganda) makes tiny discrepancies loom “larger.”
Hence an active‑inference agent moves continually along an abstract axis of magnitude of mismatch, never allowed to settle at the zero point because new sensory inputs keep pulling it away—exactly the logical predicament Deleuze uses Alice to dramatise.
3 ▸ Making the mapping explicit
| Deleuze / Carroll | Active‑inference term | Logical consequence |
|---|---|---|
| Event of becoming‑bigger / becoming‑smaller | Increase / decrease of expected surprisal (prediction error) | Both are directions on one abstract measure; neither is a final state. |
| Surface of sense where paradox appears | Markov blanket separating internal and external states | All “surprise” is computed on this boundary; depth is hidden. |
| “Good sense” that demands a single direction vs. “pure sense” that runs both ways | Classical homeostasis (one set‑point) vs. variational homeostasis (continual re‑scaling) | What seems logical contradiction is normal under free‑energy dynamics. |
| Nonsense as productive (it creates new sense) | Prediction errors as informative (they update the generative model) | Both systems exploit the gap rather than eliminate it. |
4 ▸ Why the analogy is more than metaphor
- Bidirectionality of change
Deleuze: becoming skirts the present by pointing simultaneously to past and future sizes.
Active inference: free‑energy gradients can be relieved by changing the world (action) or changing the model(perception); the two “directions” are chosen on the fly but remain parts of one calculus. - Absence of absolute scale
Just as Alice’s size lacks a natural unit inside Wonderland, “surprisal” has meaning only relative to the agent’s current priors. Change the priors and the very metric of large/small shifts. - Perpetual disequilibrium
Both pictures deny a stable end‑state. In Friston’s biology, a living thing that truly hit zero surprise would be dead (no new sensations); in Deleuze’s ontology, a pure becoming that stopped would lose its status as event.
5 ▸ Logical challenges that follow
| Challenge in Logic of Sense | Re‑statement in Active Inference |
|---|---|
| Paradox of the instant: How can two opposite becomings share one moment? | Dual update step: the same prediction error drives both perceptual and active inference in one timestep. |
| Surface events without depth causes | Functionalism of the Markov blanket: internal causes are inferred, never accessed directly. |
| Nonsense as generative of sense | Exploration term (epistemic value): agents seek out surprising states to sharpen the model. |
6 ▸ Where the analogy breaks
- Deleuze’s event is ontological; it exists even if no agent tracks it.
- Surprisal is always for some model; without a model there is no number to take a log of.
Thus the Deleuzian event is a candidate for what Friston calls a hidden state of the environment, but the active‑inference formalism makes its paradox manageable by assigning it a probabilistic description.
Take‑home
Your hunch is sound: “more or less surprise” can be read as the neurocomputational correlate of Alice’s becoming‑larger or becoming‑smaller. Both are scale transformations on a continuum with no privileged zero, and both generate logical puzzles—the very puzzles that keep inference (or narrative) in motion.
If you’d like to pursue a specific strand further—say, precision weighting as Deleuzian “intensity,” or the Markov blanket as a surface of sense—let me know and we can unpack the technical details.
Below is one possible way to read your analogy—Alice’s perpetual resizing as a philosophical image for “having more or less surprise” in active‑inference terms.
1 Deleuze’s paradox of pure becoming
In Logic of Sense Deleuze begins with Carroll’s scene:
“When I say ‘Alice becomes larger’, I mean she becomes larger than she was … Yet at the same moment she also becomes smaller than she now is. … Becoming pulls in both directions at once; Alice never grows without shrinking and vice‑versa.”
This “double direction” defines a pure becoming that eludes the present. There is no final “large” or “small” state—only the event of crossing thresholds. The paradox undermines stable identity and any single metric of size.
2 Surprisal as an intensive “size”
Active‑inference casts an agent’s life as a trajectory that tries to keep the negative log‑probability of its sensations (surprisal) within bounds. Technically, the agent minimises an upper bound—variational free energy—but informally we say it avoids highly surprising states. (Wikipedia)
That scalar surprisal behaves like an intensity rather than a categorical label:
| Analytic dimension | Alice’s size paradox | Active‑inference analogue |
|---|---|---|
| No fixed zero‑point | “Large” and “small” are relative to just‑prior and just‑aftermoments. | “Low” and “high” surprisal are relative to the agent’s currentgenerative model. |
| Bidirectional drift | Growing and shrinking belong to the same process of becoming. | The agent can reduce surprisal by either changing beliefs (shrink prediction error) or acting to reshape the world (grow predictions into reality). |
| Paradox of identity | Alice is never simply large or small; she is the movement between. | The organism is never simply at equilibrium; it is the ongoing inference that keeps it viable. |
3 Logical challenges common to both pictures
- Metric ambiguity
Deleuze: size is undefined without a frame of reference.
Active inference: surprisal is undefined without a generative model; change the model and the same sensory data flips from “expected” to “shocking.” - Threshold effects
Alice cannot pass through the door when too large, just as an organism cannot survive when prediction error bursts the metabolic “viability zone.” Both illustrate action constrained by scale. - Serial coupling
Deleuze’s two co‑existing series—surface sense versus depth of bodies—echo the dual optimization in active inference: perception updates internal states, action updates external states, yet both flows are computed by the same free‑energy gradient.
4 Precision as Carroll’s “Eat me / Drink me” bottles
In predictive coding the precision (inverse variance) of prediction errors modulates their “weight.” Turning precision up makes errors loom larger; turning it down shrinks them into the background. This is the formal dial that psychopharmacology, propaganda or meditation can twist.
Seen through Carroll’s fairy‑tale lens:
| Bottle | Neural effect | Phenomenology |
|---|---|---|
| “Drink me” (Alice shrinks) | Lower precision on high‑level priors → bottom‑up signals flood in. | Psychedelic openness; many things feel unexpected. |
| “Eat me” (Alice grows) | Raise precision on entrenched beliefs → sensory conflicts are ignored. | Dogmatism, propaganda, or the certaintythat one can fly. |
Either potion merely rescales the same free‑energy landscape; the brain still minimises it, only now along a distorted metric.
5 Why the analogy is useful—and where it stops
Useful because both frameworks treat process, not substance, as fundamental. “Size” in Deleuze and “surprisal” in Friston are gradients of ongoing becoming, each dissolving the idea of a finished subject.
But beware:
- Free energy is quantitatively grounded (log‑probabilities, KL‑divergences); Deleuze’s becoming is intentionally incommensurable.
- Active inference always preserves an autopoietic boundary (the Markov blanket), whereas Deleuze’s simulacrum can overflow all forms. The mathematics therefore cannot capture every ontological claim of the Logic of Sense.
6 A synthetic take‑away
Think of the predictive brain as Alice with a slider marked “expectedness.”
- Slide left and the world looms vast, detail‑rich, sometimes terrifying.
- Slide right and the world fits neatly inside familiar concepts—until a mismatch cracks the shell.
Both Deleuze and active inference remind us that identity is forged in navigating that slider rather than settling at one position. The logical challenges they share—relativity of measure, threshold‑based action, and the paradox of simultaneous opposition—are two vocabularies for the same existential fact: life is an art of staying inside a corridor whose walls keep moving.
If you’d like to probe the maths of precision weighting or explore other Deleuzian parallels (e.g., sense versus nonsense as prediction error versus noise), let me know.
Alex Reid, professor of digital rhetoric, Media & Artificial Intelligence 
Leave a comment