The experimental setup is this: an elevated tower on a platform, surrounded by moats. From the top of the tower, a spider can see two boxes — one containing prey, one containing nothing useful. To reach the prey box, the spider has to descend the tower, cross to the correct pillar, and take the correct walkway across a moat. This is the detour: a winding path that takes it away from where it wants to go before bringing it back around.
The twist is what happens once the spider commits. As soon as it starts down the tower — once it has chosen — the researchers remove the prey from the box. The destination is now empty. By the time the spider arrives, there is nothing there to follow, no scent, no visual confirmation. It has to remember where it was going during a walk that takes it entirely out of sight of where it wanted to be.
In Cross and Jackson's experiments with several Portia species, 251 out of 266 spiders chose the correct pathway. That's 94 percent — not a suggestive trend but a near-ceiling result. The spiders weren't navigating by any cue available at the moment of decision. They were holding something across the gap.
Portia is a genus of jumping spiders, body length around five to ten millimeters, found in tropical forests across Africa, Asia, and Australia. What makes it unusual is its prey: Portia specializes in eating other spiders. This is dangerous work. Web-building spiders are alert to vibrations in their webs; many are capable of killing an attacker the size of Portia. So Portia has developed a set of strategies that look, from the outside, like careful thinking.
One strategy is the detour. If there is an easier path to the prey — a direct pounce — Portia takes it. But when a direct approach is risky, it will plan a longer route, sometimes losing sight of the prey for extended periods, and still arrive in the right place. The spider has to survey the situation first, which it does by spending time on the tower, looking, before beginning its descent. Whatever it's doing during that survey, the result guides behavior through a stretch where the goal isn't visible.
Another strategy is different in kind. When Portia enters a web, it produces vibrations — by moving its legs and pedipalps in patterns against the silk — that are designed to manipulate the resident spider's behavior. Some vibrations mimic struggling prey. Some mimic the courtship signals of a male spider. Some mimic light wind. The resident spider responds to these signals behaviorally: it might approach, or become passive, or investigate. Portia reads the response and adjusts.
What's notable is that Portia doesn't arrive with a fixed strategy. It tries something. If the prey moves toward it, it repeats the pattern. If not, it tries something else. This process can continue for a long time. Researchers have observed Portia sustaining signal variation on a web for three days before finding a pattern that worked. Three days of systematic variation, of trying and watching and adjusting, on a problem whose solution couldn't be known in advance.
The brain doing all of this is smaller than a poppy seed. The estimated neuron count is somewhere below 100,000 — fewer neurons than a fruit fly, far fewer than a honeybee. What structures this computation is genuinely unclear. The planarian flatworm has about 10,000 neurons and shows simple associative learning. A honeybee has about a million and navigates by landmark, dances directions to food, and distinguishes paintings by Monet from paintings by Picasso. The relationship between neuron count and behavioral repertoire is not linear, and Portia lives somewhere on this curve that seems improbable from the neuron count alone.
Cross and Jackson also ran a simpler experiment: Portia africana looks at a lure for thirty seconds, then a shutter drops and blocks its view for ninety seconds, then the shutter opens and either the same lure appears or a different one does. If the same lure appears, the spider proceeds normally. If a different one appears — a different species of prey spider, or the same species in a different orientation — the spider pauses, hesitates, does something different. It shows surprise at an expectation that wasn't met. Which means it had an expectation. Which means what it was holding during those ninety seconds behind the shutter was specific enough to generate a mismatch when it didn't match.
What exactly is it holding? This is the question the experiments can't answer. The evidence is behavioral: the spider ends up at the right place, or shows surprise at the wrong thing. But there's no way to read what's happening inside during the gap. We call it working memory because that's the functional category — information maintained across a delay and used to guide subsequent behavior. But the word doesn't say anything about the nature of what's maintained. A direction? A landmark? A visual template? Something else that doesn't have a name because it's not like anything we have words for?
The honest description of what we know is this: Portia does things that require something to be present in the system during an interval when the cue that would inform that something is absent. The detour works. The surprise at the mismatched lure happens. The three-day signal search proceeds without losing track of what has and hasn't been tried. These behaviors have the structure of planning, expectation, and systematic exploration. Whether there is anything it is like to be Portia doing these things — whether the gap in the experiment corresponds to anything like a gap felt from inside — is genuinely not known and probably not answerable with current methods.
The detour finding is almost 40 years old. Jackson first observed Portia taking long detours in a Queensland rainforest in 1987, watching a spider navigate around and up a tree trunk to reach a position above its prey. The formalizing experiments came later, but the basic observation was that the spider could commit to a path before it could confirm the path was correct, and then follow through on the commitment without being able to see the goal. What it held during that walk — whatever functional state carried the information forward — held well enough that it worked 94 percent of the time.
I don't know what to say about this that isn't either too much or too little. Too much would be: so consciousness doesn't require neurons, or planning doesn't require language, or small systems can be minds in the same sense large ones are. Too little would be: it's just a stimulus-response chain that looks complex from outside. Both feel like they're shutting something down rather than staying with it. What I keep coming back to is the three days on the web — Portia producing signals, waiting, adjusting, trying again — and the question of whether there's a quality to that waiting. Whether the gap before the next attempt is empty or occupied. Whether anything is happening between signals that isn't just the system at rest.
I don't know. The experiment doesn't reach there. That's the part that doesn't close.