My Notes on: When Ants Are Smarter Than People

Disclaimer: these are little more than reading notes. I'll list some excerpts and highlights from a paper I've loved reading recently, with only marginal explanations. Rougher than usual, but this shortcut allows me to share things I wouldn't have the time to cook otherwise. Let me know how it landed.

The paper: Comparing cooperative geometric puzzle solving in ants versus humans, by Dreyer et al. (2025)

There are so many alternative definitions of intelligence that it's hard to pick one, but in general they all tend to be about one's ability to solve new problems. We know that people are intelligent, and that dogs, say, are intelligent but less so, and other animals (and even plants, fungi, etc.) all have some level of intelligence, some much higher than others. It may be hard or impossible to measure in most cases, but the idea that there are "degrees of intelligence" seems rather uncontroversial. But what about groups?

If intelligence is about solving problems, and groups (of people, of fish, of beech trees...) are known to solve problems together, it stands to reason that we should be able to talk about "group intelligence". For example, how smart is a team of researchers compared to a single one? It seems obvious that the team is smarter, but is that a jump in cognitive abilities similar to that from a chimpanzee to a human, or more, or less? Does the group's intelligence grow steadily with the number of members? And what factors determine how much the intelligence grows, if at all?

All of these are super important questions, if you think about how much of our modern world relies on the cooperation of many people. While it is impossible to answer them rigorously yet, this paper by Dreyer et al. does an excellent job at shedding some light.

The Puzzle

Early on, the paper acknowledges the main difficulty in assessing "group intelligence":

While tempting, direct comparisons between the cognitive abilities of wholes versus individuals are often meaningless because different group sizes tend to interact with the environment at different scales. For example, there is no meaningful way to compare the cognitive capacity of a single neuron to that of the brain.

Another way to put it is that the problems faced by an individual agent are usually different from those faced by groups. The comparison would be apples to oranges.

Usually, that's the case. But here's where the authors' ingenious idea comes in: find one problem or task that remains the same regardless of the number of participants.

What's more, they wanted a problem so scalable that it could be applied equally well to people and ants! Ants are known to achieve amazing feats when they work shoulder to shoulder (do they even have shoulders?), so I can see why the researchers would be curious to compare them with people. But what kind of problem could fit the bill here?

Of all the social animals, only ants and people excel at cooperative transport.

That's the perfect task to level the playing field, then: have them carry things around.

Dreyer et al. devised the following scalable puzzle: carry a (relatively) large, T-shaped object through two doorways. For the humans, the object had wheels and handles to be pulled around, while the ants (longhorn crazy ants, to be precise) relied on their pincers and disproportionate strength to pull their 3D-printed, food-smelling Ts directly. The rooms of the puzzle were oriented so that the ants' nest was beyond the second door, which motivated the insects to pull the object in that direction.

Finally, they did the experiment both with individuals (single ants, single people) and with groups.

Of course, the size of the T and of the room layout was varied in each case to make the subjective conditions as similar as possible, but the proportions were the same in every case.

The task sounds easy, but the proportions of the maze are such that the T won't fit through the doors except in a very specific and indirect sequence of maneuvers. The most obvious answer—plunging the T's long end into the first door, is wrong, and so are most of the other obvious-seeming steps. Simple and hard, as every good puzzle should be: how quickly would people and ants, individuals and groups be able to solve it? That's the experiment's measure of (a form of) intelligence.

The experiment was repeated for the following five "solvers", each with the object and room layout scaled to keep the proportions right:

1. Solver: individual ants
2. Solver: groups of ants
3. Solver: individual people
4. Solver: groups of people who were allowed to communicate
5. Solver: groups of people who were prohibited from communicating

The division of groups of people into "unlimited communication" and "restricted communication" is another nice and important touch. The latter is where the playing field is really level, because ants can't speak. People in the restricted communication group wore masks and eyeglasses, and were not allowed to speak or gesture to each other. The only means of communication they could use was by pulling at levers attached to the object, i.e. letting the others know that they intend to pull in this or that direction. This mimics the limitations faced by ants.

The insects, on the other hand, were unable to use their own unique superpower:

For longhorn crazy ants, communication in the context of cooperative transport is naturally mediated by both haptic sensation and pheromone communication. However, since in the context of our puzzle, pheromones are practically useless, this primarily leaves the ants with force-based communication. This makes comparisons between ant groups and restricted communication human groups especially compelling.

Ant Performance

The first (unsurprising) finding was that "large ant groups perform significantly better than individual ants and small groups of ants."

Very few of the individual ants were able to solve the puzzle at all, no matter how much they tried. With their tiny brains, it would be difficult to expect otherwise.

The story was different for groups of ants, though. Small groups of them (about seven individuals) were able to eventually solve the puzzle about half the time. Large groups (about 80 individuals) were able to solve it 100% of the time given enough tries.

That is a striking increase in intelligence. Since no single ant has the brainpower to mentally solve the problem, the strategy they use to solve it as groups must be distributed among them, and not located in any one of them. To understand how this might work, the researchers created an ant simulator: an agent based model (ABM) where each virtual ant follows really basic but specific rules not requiring an overall understanding of the puzzle's configuration.

This empirically verified model assumes that when an ant attaches to the load, she transiently acts as an “informed leader” by pulling it in the direction of the nest. After about ten seconds, the newly attached ant switches her state to that of an “uninformed follower” and tends to align her pulling effort with the direction in which the load, at her point of contact, is moving. This tendency is larger for larger group sizes.

This simple rule seems to capture what is happening in the actual ants, because the computer simulations are able to closely mimic their performance.

Pause for a minute and ponder this fact. The groups of ants are much, much better at solving the puzzle than any single one of them, which means that they have a form of collective intelligence. The group is intelligent. But none of them understands what is happening at all. None of them thinks "oh, this puzzle is tough, we need to find a different approach!" All they can think is "pull towards the nest, pull towards the nest!"

This is a case of what Daniel Dennett calls competence without comprehension: the ability of agents (and machines) to accomplish impressive feats without having a clue as to how they're doing it.

In practice, Dreyer's team hypothesizes, the key to the increased intelligence of the ant groups is in how their movement of the T is less chaotic and more systematic. Instead of bumping around, rotating left and right and generally following an erratic wiggling motion, the object carried by larger groups of ants glides smoothly and, upon hitting against a wall, slides against it until it finds an opening. This greatly decreases the number of attempts they have to make before finding the right sequence of steps to get through the puzzle.

But why the smoother motion? With more ants pulling, each individual participant has less of an effect on the whole. If the T is already moving in one direction, a lot of ants would need to decide to change direction in unison in order to sway it.

The high persistence of large groups translates into short-term collective memory.

To summarize, emergent cognitive faculties allow large ant groups to employ a heuristic that is reminiscent of the well-known “right-hand rule,” in which, upon entering a maze, the solver slides their right hand along the wall and proceeds forward without changing their direction. Moreover, the fact that the ants occasionally move away from the wall and collide again at a new location allows them to avoid infinite loops that may plague strict right-hand-rule followers. On the other hand, small ant groups exhibit random-walk-like dynamics that include futile searches and trapping in dead ends.

Human Performance

Needless to say, people did better than the ants in all cases, but there is a lot of juicy nuance here. Let's look at the "no talk" group first.

While restricted-communication groups of people communicate via forces as ants do, they did not show a corresponding improvement with numbers. In fact, the opposite was true as these groups performed significantly worse than individuals.

Now, that is surprising. The conditions are exactly the same as the ants, and yet the collective intelligence of the groups of people was lower than individuals. When working alone, people solved the puzzle in under 13 attempts about 60% of the time, yet groups of non-communicating people did so approximately 17% of the time. That is pretty abysmal, if you ask me.

Things were different when talking was involved:

Groups which were allowed to communicate reverse this effect and marginally outperform individuals.

The group intelligence did increase in these cases, but only by a little. They solved the problem in under 13 attempts some 75% of the time (fifteen percentage points above individual solvers). On the other hand, the talking groups were considerably slower than all others, because "communicating groups of people spend significant time discussing and deciding on their next move and, by this, display similar performance to individuals." I'll touch on the possible reasons for this underwhelming group performance later.

Clearly, things work very differently here than in the ants' case. People have enormous brains capable of mentally simulating future movements, and they have the major advantage of a long-term memory.

The motion exhibited by people, either as individuals or in groups, displays even higher efficiency as these solvers tend to take the direct path [from one phase of the puzzle to the next.]

Yet even people seem to follow relatively simple rules of thumb, at least in the "restricted communication" case. The pull-handles had force sensors, so the scientists could track exactly when and how each person pulled:

The individual force-meters show that similar to ants, once motion begins, all group members tend to align their pulling efforts with it

and

Summing up the force vectors of all initial pullers is equivalent to a majority vote among this leading subgroup (or oligarchy). Once motion commences, all other group members quickly comply with this direction and transport the load to the chosen target node.

As we've seen, this strategy didn't work out very well.

Why Did Humans Proportionally Suck?

People have big brains, big enough to contain thousands of whole ants (~650,000 average-sized ants to be precise, according to Kimi K2), so I find the difference in intelligence measured in this experiment pretty disappointing.

Individual people did vastly better than individual ants—that checks out—but when they got into groups, the ants got dangerously close to people:

We find that, on average, human solvers perform better than ant solvers. However, the full performance distributions do display a small overlap as the best ant solvers outperform the worst human solvers.

Sure, we were smarter overall, but we can't be too proud of ourselves. (I imagine the referee of these sessions looking at their sensors after a while and going, "folks, I just wanted to let you know that some ants have already finished the puzzle at this point; just sayin'.")

What gives? Well, it's complicated, but the authors of the paper identify an interesting tendency, especially in the non-communicating group of people: their strategies tended to be "greedier", in the sense that they often attempted the most easy- and obvious-looking state transition first. Unfortunately, this puzzle is designed to require indirect, non-obvious steps to solve, so the greedy choices were usually wrong. But why the extra greediness?

One potential factor is that, when they can't tell each other what's around the corner, people in a group have a limited visibility of the overall situation from where they are standing. That is a physical factor. But there seems to be a psychological side, too:

Greediness may also be related to the empirical observation that when debate is prohibited, people tend to reach a consensus quickly. In this case, in line with the notion of groupthink, people tend to forsake their personal opinion and promote a different one, not because they think it is the better option, but because it is the option they believe is most likely to be independently chosen by others.

Follow the majority, they must know what they're doing, right?

It may be the case that this assessment of the majority opinion is, in fact, the minority opinion, which is reminiscent of the social phenomenon of pluralistic ignorance.

Here you might be wondering how this differs from the ant teams, since we've seen that following the majority is the core of their strategy, too. There can't be a greedier strategy than "pull towards the nest, pull towards the nest!"

The short answer is yes, ants also gingerly rush in wrong directions all together, and this is confirmed by the fact that ants still perform (mostly) worse than even the noncommunicating human groups. But at least the groups of ants do strictly better than their individuals. The paper's explanation for this is nuanced.

We assume that while longhorn crazy ants discern the context of cooperative transport, they make no distinctions regarding the geometry of the specific problem and always apply the same individual scale behavioral rules.

I.e., ants are equally greedy at all levels but, instead of being their doom, this alignment and simplicity of behavior is the key to their collective success. By dumbly pulling in one direction at a time, the large groups of ants create a sort of momentum in the movement, and this creates a form of collective memory. None of them is clever enough to remember the direction they were going a moment earlier, but the group as a whole does. So they effectively implement the right-hand technique for solving mazes, and this happens to be rather effective.

Every single human, on the other hand, can stop and reflect, imagine different options, and prune those that seem to be clearly hopeless.

People are more flexible in selecting tools from their cognitive repertoire and can finely adjust their problem-solving tactics to suit the particular task at hand. While this flexibility can enhance individual performance, it inevitably results in interpersonal differences that may require more advanced communication to avoid worsening collective performances and allow for effective cooperation.

Inside single brains, there is a sort of internal debate—what we call "thinking". This debate can be partly "scaled up", so to speak, when group members are able to talk:

Communicating groups reach consensus in a very different manner [compared to non-communicating groups]: at the beginning of a solution attempt, and before any motion commences, they tend to spend tens of seconds conversing. ... The ability to discuss frees the group from the urge to make a single-shot decision and they can, instead, take their time to advocate for less obvious, but more accurate choices toward a joint decision.

Just like lone humans, the groups that can talk are able to do this thing called "thinking", and it happens between people, not only within each skull. This option is precluded from the groups that are not allowed to communicate, and that is why their performance is more ant-like. The saying that "two heads are better than one" should come with the fine print, "if they can talk to each other."

That said, the performance of talking groups is only slightly better than single people. Apparently people are able to correct each other's ideas—someone might point out a mistake that another participant had overlooked—but we're not seeing the dramatic leaps in group intelligence that we saw in ants.

The researchers' reasonable hypothesis to explain this is that unlimited communication groups informally or implicitly elect one person as their leader. But, since those groups perform worse than the best individual solvers, we must infer that they are not very good at choosing leaders.

Therefore, communication does not significantly help the group distinguish a competent member from the rest and follow her lead.

My Takeaways

What can we learn from all this? Does understanding this paper make us better at collective action?

The temptation is to infer that we should act more like the ants: find a set of simple rules governing individual behavior, then have everyone just do their job without worrying about the big picture. Have someone draft the perfect formula, distribute the instructions to everyone top-down, and the resulting harmony will boost our overall intelligence, right? Of course not! That is exactly what the non-talking human groups did—be greedy and too eager to act—and they became dumber, not smarter!

A more cautious takeaway might be that the ants demonstrated the tremendous potential for increased intelligence in groups. If humans are still unable to increase their intelligence dozens-fold like the ants did, it may be just a matter of removing the procedural obstacles, whatever they are.

I don't think you can generalize these results into concrete operational advice. This is such a specific kind of problem, and rather simple. The best strategies for solving this sort of puzzle—even assuming we know them—will probably be useless in most other contexts. Also, the experiment was explicitly designed to remove most of the planning and thinking time for people. The unlimited communication groups had the opportunity to talk at length, but what if they had let them reflect on it for a whole day? What if they had given them pen and paper, computers, books about geometric puzzles, and training for such situations? A lot of the combined intelligence of us homo sapiens resides in our capacity to form castles in the mind, and to access external and shared memory. None of that was allowed to shine in this experiment, for obvious and good reasons.

Still, I find this work by Dreyer et al. fascinating for the way it demonstrates that the interpersonal dynamics make all the difference. It also shows what I call a fruitful shifting of boundaries: groups can be treated as their own black boxes with behaviors, strategies, and intelligence that don't directly map to those of their constituents.

Here are some questions that surfaced in my mind while reading this:

• How does the gap between individual and group intelligence change for more complex problems, especially those too complex to fully model in any single mind?
• Assuming the ability to communicate, when does this gap reverse and become negative (i.e. individuals fare better), if ever?
• Can people achieve the same leaps in collective intelligence as ants do?
• Can we have both human-level unit intelligence and ant-level concerted synergy? ●

You can watch other fun videos from the experiment here.