By admin

Study finds peculiar multi-generational collective learning in homing pigeons

Last month, Takao Sasaki and Dora Biro, both from the University of Oxford, published a curious paper in Nature Communications about homing pigeons. Sasaki and Biro had set out to examine the possibility of Cumulative Cultural Evolution (the ability of a population to accumulate knowledge or skills across generations) in homing pigeons. Previously, it was thought that such multi-generational learning was a trick unique to humans and some other primates with large brains. In their experiment, Biro and Sasaki not only showed that pigeons indeed can pass knowledge from one generation to the next, but something else much more surprising: repeatedly swapping out knowledgeable pigeons for clueless ones actually improved the overall efficiency of the group’s choices over time. Their findings have interesting implications for understanding collective intelligence and innovation in other animal groups — possibly including humans.

The Experiment

Experiment group (a), in which one member of each pair was replaced every 12 flights; and control groups (b), which had the same members throughout the experiment’s 60 flights.

Homing pigeons have are possesed of a uniquely measurable skill: finding their way back to a particular spot when released from a distant location. By fitting pigeons with minature GPS loggers and then repeatedly releasing them from the same spot, researchers can measure to what extent pigeons learn to improve their route over multiple trips.


In this experiment, pigeons were tested in three groups: one experimental group and two control groups. The first control group consisted of 10 individual birds. Each of them flew the 8.6 km route alone, a total of 60 times. The second control group of 20 pigeons was grouped into 10 pairs, who also each flew the route 60 times. The experimental group also flew primarily as pairs, but instead of the same pair for the whole experiment, every twelve flights one bird of the pair would be replaced with a new pigeon that had never flown the route before. The first twelve flights were done by a solo pigeon.


The researchers measured the flight path of each trip, and calculated how close it came to ideally optimized bee-line route.


As would be expected of an intelligent bird like a homing pigeon, the efficiency of the route for all groups improved over time.

And, supporting the hypothesis that intergenerational learning is possible even among birds, the newly introduced birds in the experimental group were able to quickly learn from experienced birds, and then pass on their route information to the next generation of inexperienced birds.

Over the course of the experiment, though, something else happened: by the end of each twelve-flight “generation”, the experiment group was consistently outperforming the navigational efficiency of the two control groups, even though the individual birds in the control groups had been flying the same route for much longer than those in the experiment group. This result was, to quote the study authors, “a phenomenon unexpected under current theories of route navigation in pigeons”.

It turns out that pigeons, like humans, form habits. By about the thirteenth flight, the pigeons in both pair and solo control groups had settled upon their own “idiosyncratic” route preferences. They had improved their routes dramatically over their initial flights, but then had picked a route that was, apparently, good enough, and their efficiency plateaued and remained at that point for the rest of the flights.

The experiment group, on the other hand, was in constant confusion. Every time an experienced bird was swapped out for a rookie, the pair’s efficiency would drop dramatically as the inexperienced newcomer influenced the flight path away from the beeline optimum. Inevitably, though, some of this newcomer’s suggestions turned out to be good ones. By the end of twelve flights, the bad suggestions had been weeded out and the good ones incorporated in the pair’s flight plan, and the pair was outperforming the old hands in the control groups and in most cases the previous generations of the experiment group as well.

Route efficiency for each of the sixty flights, by group.

Asynchronous CI, intergenerational churn, pigeon politics, and other speculative conclusions

Like any new and interesting study, it would be imprudent to draw too many conclusions until the results have been replicated and verified by other researchers. If it does turn out to be repeateable — and especially if it turns out to be consistent across species, which Sasaki and Biro are planning to test next — then we can draw some interesting general lessons about learning and group knowledge transmission that are surprisingly congruent with some common-sense notions about human groups. Let’s consider three of them:

Asynchronous collective intelligence
One way to think about the learning among pigeons is that the experimental group was able to benefit from the navigational diversity of the total number of birds that flew in its pairs. Even though there were only two birds in the flock at any one time, the total number of birds that contributed to its collective knowledge was much greater. Looked at this way, it’s not surprising that it outperformed the other groups. If unhelpful individual idiosyncracies are weeded out while beneficial ones are kept, then the total performance will increase with group size. Could the results of this experiment be explained as the cumulative contributions of a larger collective of pigeons? In order to answer that question, we would need to compare the experimental group’s performance with a group of the similar total size flying the same routes. Which, unfortunately, was not included in the study.

Is it newness, or just diversity?
Pop-sociology conclusions are hampered by an ambiguity in the experiment’s results: is the increased learning power a result of more birds, or newer birds? If it’s newer birds, we can look at this as an interesting microcosm of generational succession in humans and the curious usefulness of naive newcomers in a problem solving situation. They may not know a thing, but still, sometimes the presence of a mind without habitual answers to the given question will help bring the solution to light by one idiosyncratic means or another.

According to the eminent physicist Max Planck, human scientists bear some resemblance to pigeons in their habit-forming tendencies. “A new scientific truth,” he famously said, “does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it.” A clever piece of meta science from the National Bureau of Economic Research published in 2015 seems to bear out this adage, showing that there is a burst of progress in scientific fields where a reigning expert has just perished.

Known good configurations vs. experimental innovation
One experimental result isn’t highlighted in the paper: while the peak efficiency of the experimental group (at the end of each generation) was higher than the control groups, the average efficiency was far lower. Every time an experienced bird was replaced, efficiency plummeted. It wasn’t until near the end of the “generation” that it recovered. Experimental innovation has its costs.

A possible corollary is that in a more complex group structure — one in which knowledge can be shared between as well as within groups — there is value in sub-group experimentation. This minimizes the risk by limiting negative outcomes of experimentation to a small part of the population, while maximizing the potential reward of new knowledge that can be shared population-wide. In other words, Burning Man is potentially important to normal human society after all.

In homage to pigeons

All implication for human groups aside, let’s ponder this for a moment: imagine that you’re flying alongside a pair of pigeons in the experimental group on their first runs after a generational exchange. How do the two pigeons — one experienced, who presumably knows the flight path well, and one complete newcomer — negotiate a collective trajectory? Why does the knowledgeable pigeon follow the rookie on mistaken detours during the early flights, and how do the two pigeons — apparently without any complex language — negotiate the process of refining the route so that, after twelve flights together, they are collectively outperforming all of their peers?


For more on this, see the original paper (open access), and articles from Science and the University of Oxford.

Google (re)discovers collective intelligence

Google’s good at crunching data, and, like most data-oriented corporations, it’s also fond of applying its data collecting and processing habits to its own operations. Last year, Google announced the results of a two-year effort to understand what makes some Google teams more effective than others.

For those of you familiar with collective intelligence research, the results won’t be too surprising: After looking at 250 attributes of 180+ teams and conducting hundreds of interviews, Google’s analysts concluded that it’s not who’s on the team that matters most, but how they interact with each other.

The most important feature of successful teams was what Google’s researchers called “psychological safety”: the feeling among team members that they can take risks without insecurity or fear of embarrassment.

For more details, see these articles:

New York Times: What Google Learned From Its Quest to Build the Perfect Team

re:Work: The five keys to a successful Google team

Associated Press: Google searches itself to build more productive teams

Inc. : What Google’s New Emotional Intelligence Study Says About Teamwork and Success

This result aligns well with  experimental findings about collective intelligence from Alex Pentland and MIT’s Center for Collective Intelligence.

While Google has not discovered something entirely new, it has proved that collective intelligence principles can trump individual-centric notions of performance, even in a real-world setting.


What is Collective Intelligence?

This site is about Collective Intelligence in its broad definition: how can smaller pieces form larger wholes that are as smart, wise, and capable as possible? This question has big implications in many areas of human affairs. Anywhere where people work together for the same goal or in the same system — be it a corporation, a family, a network, or a planet — collective intelligence becomes important.

Collective intelligence is everywhere in nature. Schools of fish, flocks of birds, ant colonies, fungi, and bacteria all employ methods of aggregating the experience and instincts of their members to direct collective action.

We are each individually a deeply puzzling and mysterious case of collective intelligence: our brains are composed of some 100 billion neurons (plus many more other cells), each with its own metabolism, skeleton, organs, and channels of input and output. Yet somehow, this vast social network gives rise to a subjectively unified consciousness that is enormously more intelligent and capable than any of its constituents. How this occurs is one of the greatest unsolved mysteries in science.

This site will delve into the topic of collective intelligence in all of these fascinating aspects. It will feature ideas, theories, and speculations, but more importantly it will pull together research and examples that point to how to make collective intelligence work in practical, on-the-ground ways.

Let’s get smarter, together.