Recall the end of week 2#

Sensing is not reception. A sensory system is a commitment about which distinctions are worth paying for, and more channels can mean worse discrimination.

The question we left open

Everything so far holds nothing. Can the past constrain the present without a nervous system?

Where this week sits#

The third part is not a tour. It is the only place a bound is visible from outside, which is what the fourth part needs.

Memory outside the head#

SLOP3233 · Week 3 · Rosalind Achebe

What week 2 left us with#

A system that measures, compares, and holds nothing.

E. coli has a memory of seconds and no representation of anything.

Today: the past starts to constrain the present. Without neurons.

Physarum polycephalum#

A single-celled organism, though one that can cover a square metre.

No neurons. No nervous system. One enormous multinucleate cell, flowing through a network of tubes it builds and dismantles continuously.

The U-shaped trap#

A standard test in robot navigation.

Place a chemoattractant behind a U-shaped barrier. Gradient-following alone gets you into the U and cannot get you out — every local improvement leads back in.

Reid et al., 2012#

Physarum solves it. And the mechanism is not internal.

It refuses to cross its own extracellular slime trail. Explored ground is marked as explored, on the ground.

PNAS 109(43), 17490–17494. [E]

Work the trap#

Gradient-following is a rule with no memory:

rule:  move toward higher attractant concentration

Inside a U-shaped barrier, every direction that raises concentration leads further into the U, because the barrier blocks the short path and the gradient points through it.

step 1:  concentration rising  →  advance
step 2:  concentration rising  →  advance
step 3:  at the barrier, all forward moves blocked
step 4:  retreating lowers concentration  →  rule forbids it

The rule has no exit. It is not that escape is slow; it is unreachable.

Now add one bit of state#

rule:  move toward higher concentration, EXCEPT onto your own trail

Retreat is no longer forbidden, because the trail marks the path back as already-explored rather than as lower-value.

Sanity check: this is exactly why disguising the trail restores the failure. Remove the extra bit and you are back to the rule with no exit.

The moral#

The organism did not get cleverer. It got one extra distinction — visited against unvisited — and that distinction is stored outside its body.

Which is the cheapest possible upgrade, and it converts an unsolvable problem into a solvable one.

How they established that it is the trail#

Not by observing avoidance, by removing the cue.

Disguise the trail and the behaviour breaks. Performance on the U-trap collapses to gradient-following.

That is the difference between a correlation and a claim.

What this architecture buys, and costs#

And there is a follow-up worth knowing#

Smith-Ferguson, Reid, Latty & Beekman found the externalised memory helps in open or simply-bounded environments —

and in complex bounded environments it confers no advantage and can actively hinder navigation.

A memory you cannot switch off is not always an asset.

The theoretical suggestion#

That externalised spatial memory may be the functional precursor to internalised memory.

Attractive, and speculative. Grade [S] — it is a plausible ordering, not an established lineage.

Jennings, 1906#

Stentor roeseli, a sessile trumpet-shaped ciliate that builds itself a tube and feeds by drawing water in.

Jennings subjected it to repeated irritation and described something nobody expected in a single cell.

An ordered hierarchy of avoidance#

Why this is a strong claim#

Each response is more costly and less reversible than the last. Detaching abandons the tube entirely.

And the order is fixed. The cell does not sample randomly from a repertoire; it escalates.

In one cell. One nucleus. No neurons.

Then it was discredited#

During the behaviourist era the result was treated as non-reproducible, and it stayed that way for most of a century.

Which, on the face of it, is science working correctly. A surprising claim failed to replicate, so it was set aside.

Dexter, Prabakaran & Gunawardena, 2019#

They replicated it. Video microscopy, statistical analysis of response sequences, and the hierarchy is there.

Current Biology 29, 4323–4329.

And they found why the replications had failed.

The mismatch#

Why the mismatch matters mechanically#

S. coeruleus is motile and builds no tube.

An organism that can simply leave has no reason to escalate through increasingly costly attempts to stay. The hierarchy exists because roeseli is anchored and detaching is expensive.

So the failed replications were not testing a weaker version of the claim. They were testing a different animal with a different problem.

Sit with this one#

No misconduct. No p-hacking. No motivated reasoning.

Everyone in the chain behaved reasonably, and a correct result was thrown away for ninety years because a species name went unchecked.

One methodological detail worth having#

Dexter and colleagues also found Jennings’s original protocol, carmine powder in pond water — rarely elicited the behaviour in their hands.

They had to adjust the stimulation to get it at all.

Which is its own lesson about what “following the published method” secures.

The same year. The same plant. Different grades.#

Both 2016. Both Pisum sativum. Both claiming something cognitive.

One has held up. One has not. The seminar is about why.

Gagliano et al., 2016 — associative learning#

Pavlovian design adapted for a plant. Pea seedlings in Y-mazes, blue LEDs as the unconditioned stimulus, a fan as the neutral stimulus.

Three days of training with wind reliably preceding light from one arm.

Result: at the choice point, the fan alone influenced growth direction. Untrained plants were unaffected by it.

Scientific Reports 6, 38427.

Markel, 2020 — the replication#

Same paradigm. One crucial addition: blinded scoring.

Plants were removed from the chamber with fans and lights detached, placed in random order, and scored by an independent observer while the experimenter was in another room.

Result: no effect. eLife 9, e57614.

The exchange, and neither side conceded#

Gagliano et al. (eLife 9, e61141): the protocol was unsuitable. Their own unconditioned stimulus produced a 100% phototropic response; Markel’s produced only a slight bias, so there was nothing reliable to condition.

Markel (eLife 9, e61689): the differences cannot account for the failure. And it is surprising to claim plants always grow toward a one-hour light presentation that followed presentations from the opposite side.

Grade [C]#

Not “we don’t know yet”. Not a shrug.

The disagreement is the content, and both positions are on the record with their reasoning. This is what a live methodological dispute looks like from the outside.

Older attempts at plant conditioning — Holmes & Gruenberg 1965, Haney 1969, Levy et al. 1970, Armus 1970 — mostly failed to observe it, and those that claimed it lacked controls.

Dener, Kacelnik & Shemesh, 2016 — risk sensitivity#

A split-root design. One root system, two pots.

Both pots have the same mean nutrient concentration. One is constant, one varies over time. The measure is relative root mass in each.

The result, and why it is the interesting one#

That is the energy budget rule from animal foraging theory. Gamble when you cannot survive the average; don’t when you can.

Current Biology 26(13), 1763–1767. [E]

Kacelnik’s own framing#

Not that plants are intelligent.

That an adaptive response to risk does not require a nervous system, and that complex behaviour can be predicted as a biological adaptation without invoking anything cognitive.

The restraint in that claim is why it survived.

So: why do these two get different grades?#

Ask this in the seminar before you answer it.

“One replicated” is the start of the answer. The rest involves: which claim was more surprising relative to prior work, which result depended on unblinded judgement, and which framing made the smaller demand on the reader.

What week 3 established#

1. Memory can live outside the organism, and doing so changes what it can and cannot distinguish.
2. A single cell can run an ordered, escalating response hierarchy.
3. A correct result can be discarded for ninety years by people behaving reasonably.
4. Two claims about the same organism can deserve different grades, and saying why is a skill.