978-1-5090-3502-1/16/$31.00 ©2016 IEEE
Crowds vs Swarms, a Comparison of Intelligence
Louis Rosenberg and David Baltaxe
2443 Fillmore Street, #116
San Francisco, CA. USA
University of Oxford
Christ Church College
Abstract— for well over a century, researchers in the field of
Collective Intelligence have shown that groups can outperform
individuals when making decisions, predictions, and forecasts. The
most common methods for harnessing the intelligence of groups
treats the population as a “crowd” of independent agents that
provide input in isolation in the form of polls, surveys, and market
transactions. While such crowd-based methods can be effective,
they are markedly different from how natural systems harness
group intelligence. In the natural world, groups commonly form
real-time closed-loop systems (i.e. “swarms”) that converge on
solutions in synchrony. The present study compares the predictive
ability of crowds and swarms when tapping the intelligence of
human groups. More specifically, the present study tasked a
crowd of 469 football fans and a swarm of 29 football fans in a
challenge to predict 20 Prop Bets during the 2016 Super Bowl.
Results revealed that the crowd, although 16 times larger in size,
was significantly less accurate (at 47% correct) than the swarm (at
68% correct). Further, the swarm outperformed 98% of the
individuals in the full study. These results suggest that swarming,
with closed-loop feedback, is potentially a more effective method
for tapping the insights of groups than traditional polling.
Keywords— Swarm Intelligence, Artificial Intelligence, Human
Swarming, Wisdom of Crowds, Collective Intelligence
In the field of Collective Intelligence, one story is told more
often than any other – the tale of Sir Francis Galton and his
pioneering use of a crowd to estimate the weight of an ox at a
county fair in England in 1906. 800 people tried their luck at
guessing the ox’s weight – not experts, but a mix of farmers and
butchers and regular fairgoers. Galton’s goal was to show that
average voters were not very smart, but he was shocked to find
that the average of the 787 estimates (800 minus 13 that were
illegible) was nearly perfect. This was true despite the fact that
the vast majority of the individual estimates were way off .
And thus was born the field of Collective Intelligence.
In the century since, the most common methods used for
tapping the wisdom of crowds have not changed much, still
focusing on the averages of independently contributed votes and
estimates. In some cases, the methods for tallying votes are
dependent upon the prior accuracy of individual participants.
This has been used to improve crowd averages [2,3,4] but such
methods require task-relevant history and then use that history
to alter the make-up of the crowd. This is not the same as
improving the methods by which the crowd’s intelligence is
tapped. This begs the question – is there a fundamentally better
way to harness the intelligence of human groups?
To answer this question, researchers have looked to Mother
Nature for guidance, finding that many species have evolved
methods for tapping the intelligence of groups [5,6,7,8,9,13].
What nature does not d o is collect independent samples and then
aggregate the data after the fact, the way Galton did in his
famous experiment. Instead, nature forms real-time closed-loop
systems with continuous feedback, enabling large groups to
work in synchrony and converge on solutions together. Known
as Swarm Intelligence (SI), this is the reason why birds flock,
fish school, and bees swarm – they make better decisions
together than the independent organisms could make on their
own [10, 11, 12]. Of course we humans didn’t evolve the ability
to swarm, for we lack the innate connections that other species
use to establish feedback-loops among members. Ants use
chemical traces. Fish detect ripples in the water around them.
Bees use vibrational gestures. Birds detect motions propagating
through the flock. This said, new technologies have enabled
groups of online human users to form real-time closed-loop
systems modeled after natural swarms.
Known as Artificial Swarm Intelligence (ASI), or more
simply “Human Swarming”, these computational methods
enable online human groups to work together in real-time by
forming a unified system that can answer questions, make
predictions, reach decisions, or take actions. As a system, human
swarms can collectively explore a decision-space and converge
upon preferred solutions. Prior studies have shown that by
working in swarms, human groups can outperform their
individual members as well as outperform groups taking
traditional votes or polls [5,6]. For example, in a prior study, a
randomly selected human group was tasked with predicting the
2015 Oscars, both by taking a poll and by forming a swarm.
Across 48 participants, the average poll result achieved 6 of 15
correct predictions (40% success). When taking the most
popular prediction in the poll as a crowd aggregate, the group
improved, achieving 7 of 15 correct predictions (47% success).
But when working together as a swarm, the group improved far
more, achieving 11 of 15 correct predictions (73% success). This
suggests that human swarming may be a superior method for
tapping the wisdom of crowds as compared to traditional votes,
polls, and surveys. The present study aims to explore the power
of crowds vs. swarms further, by comparing a poll of nearly 500
people with a swarm of only 29 participants in a more
challenging prediction task.
To evoke a real-time Artificial Swarm Intelligence (ASI)
among groups of networked humans, technology is required to
close the loop among members. To address this need, an online
platform called UNU was developed to allow distributed groups
of users to login from anywhere around the world and participate
in a closed loop swarming process. As shown in Figure 1, users
answer questions by collectively moving a graphical puck to
select among a set of alternatives. The puck is modeled as a
physical system with a defined mass, damping and friction.
Users provide input by manipulating a graphical magnet with a
mouse or touchscreen. By positioning their magnet, users impart
their personal intent as a force vector on the puck. The input
from each user is not a discrete vote, but a stream of vectors that
varies freely over time. Because the full set of users can adjust
their intent at every time-step, the puck moves, not based on the
input of any individual, but based on the dynamics of the full
system. This results is a real-time physical negotiation among
the members of the swarm, the group collectively exploring the
decision-space and converging on the most agreeable answer.
Fig 1. A human swarm comprised of user-controlled magnets.
It’s important to note that users can only see their own
magnet during the decision, not the magnets of others users.
Thus, although they can view the puck’s motion in real time,
which represents the emerging will of the swarm, they are not
influenced by the specific breakdown of support across the
available options. This limits social biasing. It’s also important
to note that users don’t just vary the direction of their input, but
also the magnitude by adjusting the distance between the magnet
and the puck. Because the puck is in motion, to apply full force
users need to continually move their magnet so that it stays close
to the puck’s rim. This is significant, for it requires all users to
be engaged during the decision process. If they stop adjusting
their magnet to the changing position of puck, the distance
grows and their applied force wanes.
To compare the predictive ability of crowds and swarms, a
formal prediction experiment was conducted using an easily
verifiable set of prediction events – a set of twenty independent
“Proposition Wagers” on Super Bowl 50, which took place on
Sunday, February 7, 2016. Known generally as “Prop Bets”
these are simple binary wagers aimed at the general public rather
than sophisticated sports fans, and are defined by Vegas odds-
makers to have equal probabilities. In other words, if Vegas sets
the bets correctly, the average person placing the 20 wagers
would get 10 correct and 10 incorrect. Of course, the attraction
for betting on Prop Bets is that most people believe they can
outsmart the odds makers and choose the more likely alternative.
In practice, this not the case – most people perform at the
expected odds which is why Vegas makes a healthy profit.
To compare the predictive ability of Crowds and Swarms,
two experimental groups were fielded – Group A and Group B.
Group A was comprised of 469 self-identified football fans who
were randomly assembled and asked to make predictions for
each of the 20 prop bets by working together as a crowd. This
entailed each of the 469 participants filling out an online survey
to indicate their individual picks for each of the 20 bets. A set
of “crowd-based wagers” were then generated by taking the
most popular answers across the full set of participants. Group
B was comprised of 29 self-identified football fans who were
randomly assembled and asked to make predictions for each of
the 20 prop bets by working together in real time as a swarm
using the UNU software platform. A set of “swarm-based
wagers” were then generated by the group converging in
synchrony as a real-time dynamic system. Prior to participating
as a swarm, the 29 members of Group B also recorded their
individual predictions by completing surveys.
It should be noted that Prop Bets for the Super Bowl are a
mix of sports predictions and pop-culture predictions, making
them a good target for a random sampling of casual sports fans.
For example, one of the target questions was firmly sports
related – “Which team will score first, the Broncos or the
Patriots?” while another of the target questions was more about
predicting the culture of football – “Who will the MVP thank first
when interviewed after the game?” Figure 2 below shows the
swarm in the process of answering that question.
Fig 2. A human swarm answers a Super Bowl prop bet.
Looking first at crowd-based wagers, the group of 469
football fans collectively achieved 9 correct picks out of 19
wagers placed. It should be noted that the 20
canceled by Vegas at the end of the game because it involved an
event that did not transpire. Thus, the 9 out of 19 success rate
for the crowd-based wagers translated into a 47% accuracy and
a small gambling loss. In this case, tapping the wisdom of the
crowd by taking an average of poll results did not allow the
participants to beat the Vegas odds makers and did not achieve
a profit on their wagers. In fact, the crowd-based bets performed
at the expected odds distribution for individual bets.
Looking next at the swarm-based wagers, the group of 29
randomly selected football fans who worked together as a real-
time closed-loop system, collectively achieved 13 correct picks
of the 19 wagers placed. Thus by working together as a swarm,
this group of 29 individuals produced a 68% accuracy rate. This
translated into a 36% gambling gain on the placed wagers. In
other words, the swarm of randomly selected football fans were
able to defy Vegas odds by pooling their knowledge and
intuition in real-time as a Swarm Intelligence. This conforms to
prior studies that show similar results. [6, 7]
It appears that by forming a Swarm Intelligence using the
UNU platform, the group of 29 randomly selected football fans
were able to significantly amplify their group intelligence with
respect to forecasting Super Bowl prop bets and thereby provide
deeper insights. The question remains, was the amplification of
gambling insight shown by the Swarm statistically significant as
compared to the 29 individuals, if they had each worked alone
and made their own wagers. Similarly, the question remains,
was the amplification of gambling insights shown by the Swarm
statistically significant as compared to the much larger pool of
469 individuals who participated in the crowd.
To answer these questions, the swarm’s performance was
compared to the statistical distribution of performance of
individual members of Group A (the crowd) and Group B (the
swarm). In Figure 3 below, the swarm’s performance is
represented by the red line on each graph. The x-axis represent
how many questions the individuals answered correctly. The y-
axis represent how many people answered correctly to that
particular number of questions. For both groups, most people
answered correctly 10 questions out of 19.
Fig 3. Swarm Performance vs Group Members
On the left side of Fig 3, the performance distribution of the
469 member crowd is shown. As indicated by the red line, the
swarm outperformed the individuals in the crowd by 2 standard
deviations (Z=1.99). In fact, out of the 469 randomly selected
people, only 4 individuals did better than the swarm. In other
words, the swarm outperformed 99% of the individual members
of the crowd. This suggests a significant amplification of
intelligence resulting from the swarming process.
On the right side of Fig 3, the performance distribution of the
29 members of the swarm is shown. In other words, this
compares the swarm as a unified system with the individual
participants in the swarm itself. As shown by the red line, the
swarm outperformed the individuals in the group by 1.7 standard
deviations (Z=1.72). In fact, out of the 29 randomly selected
people, only 3 individuals did better than the swarm. This
corresponds with the swarm outperforming 90% of the
individual swarm members. This suggests a clear amplification
of intelligence resulting from the swarming process.
As a further statistical test, the swarm’s performance was
compared to the distribution of correct answers expected by
chance. Using a random sampler, 19 individuals were selected
at random (with replacement) and the n-th individual answer
taken as the answer of the sample for the n-th question; that is,
the first randomly picked individual provides the answer for the
1st question, the second individual provides the answer for the
2nd question and so on. This process was repeated 10,000 times
until 10,000 nominal groups were created. Figure 4, below,
represents how many times these groups answered correctly
exactly x questions.
Fig 4. Swarm Performance vs Groups of Randomly Selected Members
In Figure 4, the chart on the left depicts the distribution of correct
answers when individuals were randomly selected from among
the 469 member crowd; the chart on the right shows the
distribution when selections came from the members of the
swarm itself. Again, the swarm’s performance is represented by
the red line. This analysis shows that the swarm outperformed
chance significantly, falling almost 2 standard deviations from
the mean in both comparison groups (Z=1.90 and 1.81,
Can human swarming amplify the intelligence of groups,
enabling a population of individual forecasters to perform better
together than the vast majority could perform alone? The results
of this study, along with prior studies, suggest this is the case.
This bolsters the premise that by working together in closed-
loop systems, with real-time feedback control, groups can more
effectively explore a decision-space and converge on optimal
solutions. Although football wagers were used as the testing
framework for this study, we believe the result are applicable to
a wide range of applications where groups of individuals can
contribute a diverse set of opinions, insights, and intuitions.
Is human swarming a more effective method for harnessing
the intelligence of groups than traditional “Wisdom of Crowd”
methods for aggregating forecast data? The results of this study,
along with prior studies, suggest this is the case. The 29
members of the swarm performed significantly better as a
synchronous system than as a collection of independently
surveyed participants. In addition, the swarm outperformed the
crowd-based poll despite the fact that the poll had a sample size
that was 16-times larger. This suggests that swarming not only
provides more accurate insights, it enables insights to be attained
from human groups with a much smaller sample sizes than polls
or surveys. Future work is needed to quantify the effective size
differences between crowds and swarms.
Should we be surprised by the effectiveness of Artificial
Swarm Intelligence for harnessing group insights? If we look to
Mother Nature as our guide – probably not. After all, the results
of this study parallel the benefits of swarming among honeybees
and other social organisms, where the decisions are reached in
real-time synchrony as closed-loop dynamic systems [5, 12]. In
fact, the swarming algorithms used by the UNU platform on
which this study was run, were modeled specifically after the
decision making processes of honeybees [7,13], so it’s
reasonable to expect a similar amplifications of intelligence.
This work was directly supported by Unanimous A.I., the
maker of the UNU platform for real-time human swarming. For
more information about UNU, visit http://UNU.ai.
 Galton F (1907) Vox populi. Nature 75, 450–451
 Whitehill J, Wu TF, Bergsma J, Movellan JR, Ruvolo PL (2009) Whose
vote should count more: Optimal integration of labels from labelers of
unknown expertise. Adv. Neur. In. 2035–2043
 Lee MD, Steyvers M, De Young M, Miller B (2012) Inferring expertise
in knowledge and prediction ranking tasks. Top. Cogn. Sci. 4(1), 151–
163. doi: 10.1111/j.1756-8765.2011.01175.x PMID: 22253187
 Madirolas G, de Polavieja GG (2015) Improving Collective Estimations
Using Resistance to Social Influence. PLoS Comput Biol 11(11):
 Seeley, Thomas D. Honeybee Democracy. Princeton University Press,
 Rosenberg, L.B., “Human Swarms, a real-time method for collective
intelligence.” Proceedings of the European Conference on Artificial Life
2015, pp. 658-659
 Rosenberg, Louis. "Artificial Swarm Intelligence vs Human Experts",
Neural Networks (IJCNN), 2016 International Joint Conference on. IEEE.
 Palmer, Daniel W., et al. "Emergent Diagnoses from a Collective of
Radiologists: Algorithmic versus Social Consensus Strategies." Swarm
Intelligence. Springer International Publishing, 2014. 222-229.
 Eberhart, Russell, Daniel Palmer, and Marc Kirschenbaum. "Beyond
computational intelligence: blended intelligence." Swarm/Human
Blended Intelligence Workshop (SHBI), 2015. IEEE, 2015.
 J.A.R. Marchall, R. Bogacz, A. Dornhaus, R. Planque, T.Kovacs, N.R.
Franks, On optimal decision making in brains and social insect colonies,
J.R. Soc Interface 6,1065 (2009).
 I.D. Couzin, Collective Cognition in Animal Groups, Trends Cogn. Sci.
 Seeley, Thomas D., Visscher, P. Kirk. Choosing a home: How the scouts
in a honey bee swarm perceive the completion of their group decision
making. Behavioral Ecology and Sociobiology 54 (5) 511-520.
 L. B. Rosenberg, "Human swarming, a real-time method for parallel
distributed intelligence," Swarm/Human Blended Intelligence Workshop
(SHBI), 2015, Cleveland, OH, 2015, pp. 1-7.