Conference PaperPDF Available

Prioritizing Policy Objectives in Polarized Groups using Artificial Swarm Intelligence


Abstract and Figures

Groups often struggle to reach decisions, especially when populations are strongly divided by conflicting views. Traditional methods for collective decision-making involve polling individuals and aggregating results. In recent years, a new method called Artificial Swarm Intelligence (ASI or Swarm AI) has been developed that enables networked human groups to deliberate in real-time systems, moderated by artificial intelligence algorithms. While traditional voting methods aggregate input provided by isolated participants, Swarm-based methods enable participants to influence each other and converge on solutions together. In this study we compare the output of traditional methods such as Majority vote and Borda count to the Swarm method on a set of divisive policy issues. We find that the rankings generated using ASI and the Borda Count methods are often rated as significantly more satisfactory than those generated by the Majority vote system (p<0.05). This result held for both the population that generated the rankings (the "in-group") and the population that did not (the "out-group"): the in-group ranked the Swarm prioritizations as 9.6% more satisfactory than the Majority prioritizations, while the out-group ranked the Swarm prioritizations as 6.5% more satisfactory than the Majority prioritizations. This effect also held even when the out-group was subject to a demographic sampling bias of 10% (i.e. the out-group was composed of 10% more Labour voters than the in-group). The Swarm method was the only method to be perceived as more satisfactory to the "out-group" than the voting group.
Content may be subject to copyright.
Prioritizing Policy Objectives in Polarized Groups
using Artificial Swarm Intelligence
Gregg Willcox
Unanimous AI
San Francisco, CA
Louis Rosenberg
Unanimous AI
San Francisco, CA
Mark Burgman
Imperial College London
London, UK
Alexandru Marcoci
University of North
Carolina, Chapel Hill
Chapel Hill, NC
Abstract Groups often struggle to reach decisions, especially
when populations are strongly divided by conflicting views.
Traditional methods for collective decision-making involve polling
individuals and aggregating results. In recent years, a new method
called Artificial Swarm Intelligence (ASI) has been developed that
enables networked human groups to deliberate in real-time
systems, moderated by artificial intelligence algorithms. While
traditional voting methods aggregate input provided by isolated
participants, Swarm-based methods enable participants to
influence each other and converge on solutions together. In this
study we compare the output of traditional methods such as
Majority vote and Borda count to the Swarm method on a set of
divisive policy issues. We find that the rankings generated using
ASI and the Borda Count methods are often rated as significantly
more satisfactory than those generated by the Majority vote
system (p<0.05). This result held for both the population that
generated the rankings (the “in-group”) and the population that
did not (the “out-group”): the in-group ranked the Swarm
prioritizations as 9.6% more satisfactory than the Majority
prioritizations, while the out-group ranked the Swarm
prioritizations as 6.5% more satisfactory than the Majority
prioritizations. This effect also held even when the out-group was
subject to a demographic sampling bias of 10% (i.e. the out-group
was composed of 10% more Labour voters than the in-group). The
Swarm method was the only method to be perceived as more
satisfactory to the “out-group” than the voting group.
KeywordsArtificial Swarm Intelligence, Human Swarming,
Artificial Intelligence, Voting Methods, Borda Count, Majority
Voting, Brexit.
Groups often struggle to reach satisfactory decisions, in the
sense that most participants approve of the decision, especially
when the population is divided by conflicting views. This is
particularly true in the realm of governmental decision making,
as deeply held political and ideological opinions often prevent
groups from reaching decisions that satisfy the whole, or even
most of, the population.
Traditional approaches to group decision-making solicit
isolated opinions. The results are then aggregated using a voting
algorithm. The Majority Voting algorithm, in which each polled
respondent votes for only one candidate, and the candidate with
the most votes wins, is the most widely voting algorithm in the
Anglo-Saxon world [20].
Other widely studied systems include ranked voting
methods, such as the Condorcet [22] and Borda Count [23]
methods, both developed in the 1700’s as alternatives to
Majority Voting. In these methods, participants rank all
candidates from the most to least preferable, and the candidates
that are ranked with the highest average preference, or that are
ranked higher than other candidates most often, win the vote.
These traditional voting systems have often faced criticism
because all fail to pass simple tests of “fair” aggregation
algorithms [26]. As one example, the Borda and Majority
algorithms are very open to manipulation in real voting systems
[21, 27, 28]. Research that tries to distinguish between these and
other methods in real-world practice and to find the best voting
method for a given context, often tries to calculate the utilitarian
value for each voting method’s outcome, which is unrealistic in
real-world scenarios [25].
In recent years, a new method called Artificial Swarm
Intelligence (ASI) has been developed that enables networked
human groups to deliberate in real-time systems moderated by
artificial intelligence algorithms. Whereas most existing voting
methods focus on collecting isolated input from a group of
participants in parallel, ASI collects input from participants in
real-time: the group engages in a real-time deliberation to vote
on a set of alternatives, and participants can switch their
response at any point.
A recent study [19] attempting to measure the utilitarian
optimality of groups with conflicting opinions found that ASI
reaches decisions that are significantly better, as measured by
the monetary amount won by the group, than the Borda and
Majority methods. In another study [24], groups made political
prioritizations using both ASI and Majority Voting protocols.
The group later rated it’s own prioritizations made via ASI as a
more accurate representation of the group’s opinions (74% of
responses) and a more accurate representation of individual
priorities (66% of responses) than the Majority Voting protocol.
This study aims to identify the conditions under which
groups benefit most from using ASI to facilitate decision-
making (when compared to the Borda and Majority methods).
In traditional voting schemes, participants provide input in
isolation. In swarm-based methods, groups think together in
systems modeled after biological swarms and converge on
collective solutions. As shown in Figure 1, a typical ASI system
includes a group of users connected in real-time over a network.
Each computer, which may be a desktop, tablet, or phone, runs
a unique software interface designed to capture and stream the
user’s real-time input to a cloud-based processing engine. The
engine runs swarming algorithms and sends back real-time
output to each user, creating a closed-loop.
Fig.1. System Diagram for a real-time ASI System
The present study uses Swarm AI technology, which is
modeled largely on the dynamic behaviors of honeybee swarms.
The decision-making process that governs honeybee swarms has
been researched since the 1950s and has been shown at a high
level to be quite similar to decision-making in neurological
brains [13,14]. Both employ populations of simple excitable
units (i.e., neurons and bees) that work in parallel to integrate
noisy evidence, weigh competing alternatives, and converge on
decisions in real-time. In both brains and swarms, outcomes are
arrived at through competition among sub-populations of
excitable units. When one sub-population exceeds a threshold
level of support, the corresponding alternative is chosen. In
honeybees, this enables hundreds of scout bees to work in
parallel, collecting information about their local environment,
and then to converge together on a single optimal decision,
frequently picking the best solution to complex multi-variable
problems [15-17].
The similarity between “brains” and “swarms” is apparent
when comparing the decision-making models that represent
each. The decision process in primate brains is often modeled
as mutually inhibitory leaky integrators that aggregate incoming
evidence from competing neural populations [18]. A common
framework for primate decision-making is the Usher-
McClelland model in Figure 2 below.
Fig. 2. Usher-McClelland model of neurological decision-making
This neurological decision model can be compared to
swarm-based decision models, for example the honey-bee
model represented in Figure 3. As shown below, swarm-based
decisions follow a very similar process, aggregating input from
sub-populations of swarm members through mutual excitation
and inhibition, until a threshold is exceeded.
Fig. 3. Mutually inhibitory decision-making model in bee swarms
Thus, while brains and swarms are very different forms of
intelligence, both enable decisions to emerge from the
interactions among collections of processing units. The goal of
the present study is to apply this decision-making model to
human groups deliberating on divisive political issues and
investigate the satisfaction of group members with the collective
To enable swarming among groups of networked humans,
ASI technology allows distributed groups of users to form
closed-loop systems [5-7] and (a) integrate noisy evidence, (b)
weigh competing alternatives, and (c) converge on decisions in
synchrony, while also allowing all participants to perceive and
react to the changing system in real-time.
As shown in Figure 4, networked human groups can answer
questions as a “swarming system” by collaboratively moving a
graphical puck to select among a set of alternatives. Each
participant uses a mouse or touchscreen to manipulate a
graphical magnet. By positioning their magnet with respect to
the moving puck, participants impart their personal intent on the
system as a whole. The input from each user is not a discrete
vote, but a stream of real-time vectors that varies freely. Because
all users can adjust their intent continuously in real-time, the
puck and the participant swarm around it move, based on the
dynamics of the full system. This enables a complex negotiation
among all members simultaneously, empowering the group to
collectively explore the decision-space and converge on the
most agreeable solution in synchrony.
Fig. 4. A human swarm answering a question in real-time
It is important to note that participants freely modulate both
the direction and magnitude of their intent by adjusting the
distance between their magnet and the puck. Because the puck
is in continuous motion across the decision-space, users need to
continually adjust their magnet so that it stays near the puck’s
outer rim. This is significant, for it requires participants to
remain continuously engaged throughout the decision process,
evaluating and re-evaluating the strength of their opinions as
they convey their contribution. If they stop adjusting their
magnet with respect to the changing position of the puck, the
distance grows and their imparted sentiment wanes. A more
complete description of the algorithm can be found in [31, 32].
Thus, like bees vibrating their bodies to express sentiment
in a biological swarm, or neurons firing activation signals to
express conviction levels within a biological neural-network, the
participants in an artificial swarm must continuously update and
express their changing preferences during the decision process,
or lose their influence over the collective outcome. In addition,
algorithms monitor the behaviors of all swarm members in real-
time, inferring their implied conviction based upon their relative
motions over time. This reveals a range of behavioral
characteristics within the swarm population and weights their
contributions accordingly, from entrenched participants to
flexible participants to fickle participants.
A study was conducted to evaluate a political constituency’s
satisfaction with the output of three prioritization methods
Majority voting, Borda Count voting, and ASI.
A. Pilot Study: Question Divisiveness Measurement
The questions in this study needed to be highly politically
divisive for British participants, so a pilot study was conducted
to evaluate the divisiveness of six policy questions among
Labour and Conservative members of the UK public. With help
from political science experts, six policy ranking questions
were designed that were likely to be divisive. These six
questions were sent to 42 UK citizens: 22 Conservative and 20
Labour voters.
The divisiveness of each question was measured as the average
difference between the Labour and Conservative rankings of
the question’s political objectives. The three questions that
were the most divisive were selected for inclusion in the
following experiment, and are described in Appendix A.
B. Participants
To evaluate the satisfaction of a political constituency with
various prioritization methods we recruited N=237 participants
using a market research company. 119 participants were female.
Participants were compensated for the 45 minutes or less session
in line with the practices of the market research company at a
rate of approximately £36 / hour. All participants signed an
informed consent form.
C. Materials and Methods
Participants were randomly assigned into two conditions: the In-
Group and the Out-Group. Participants in the former condition
were randomly assigned to four groups of between 8 and 20
individuals, and were tasked with prioritizing three sets of policy
objectives from least to most important and then ranking their
satisfaction with each prioritization method. These four groups
were considered the “in-group”, as they contributed to the
prioritizations that they later scored for satisfaction. One “out-
group” of 170 participants was also convened that did not
contribute to any prioritization, and that only ranked their
satisfaction with each prioritization from the in-groups.
To give a more complete overview of the demographics of
participants in this study, table 1 lists the demographic
composition of each group, split by Political Affiliation, Brexit
Stance, and Gender.
Table 1: Demographic Breakdown of Groups
Participants in the in-groups first provided their answers
independently using a standard online survey to prioritize each
set of objectives. Upon completion, the groups congregated on
the Swarm AI platform (an online tool, purposefully built to
facilitate ASI decision-making) to answer the same set of
While the participants were completing the ordering tasks using
ASI, their survey results were analyzed using the Majority and
Borda Count algorithms to generate two ordered lists. For the
Majority algorithm, the objectives were ordered by the number
of participants that ranked each objective as the “most
important”, and ties were broken randomly. In the Borda Count
Number of
Labour /
Leave /
Remain /
Male /
In-Group 1
4 / 4
3 / 5 / 0
4 / 4
In-Group 2
10 / 10
9 / 11 / 0
10 / 10
In-Group 3
12 / 8
8 / 12 / 0
6 / 14
In-Group 4
7 / 12
10 / 8 / 1
12 / 7
87 / 83
70 / 88 / 12
86 / 84
algorithm, each participant’s ranking was converted into a score
for each objective: 1 point for the “most important” objective,
2 points for the second most important objective, 3 points for
the third most important, etc., and the sum of these points across
all participants in the group was calculated for each objective.
The objectives were ordered from least points (most important)
to most points (least important), with ties broken randomly.
When moving to the Swarm AI platform, the groups prioritized
the objectives using an iterative elimination approach: the
groups started by selecting the LEAST important objective out
of the 6 objectives listed, then this objective was eliminated
from consideration, and the group repeated the process, until
there were two objectives left. For the final elimination, the
question was flipped, and the group was asked which of the
remaining objectives was the MOST important. The ranking
generated in this way using the swarm platform was considered
the group’s ranking of the objectives.
After completing all questions on the Swarm AI platform,
participants were redirected to a follow-up survey, where they
were presented with the three questions they just answered,
along with three anonymized rankings for each questionthe
Majority, Borda Count, and ASI rankings. For each question,
participants were asked to rank each of the three lists based on
their level of satisfaction with the list, from 1-most satisfied to
3-least satisfied. The three anonymized lists were presented in
a different order for each question to eliminate ordering bias.
Finally, an out-group of participants was assembled to represent
a public constituency that did not directly vote on the policy
objectives. Members of the out-group were not a part of any of
the four group ranking exercises. The satisfaction of the out-
group with each of the in-group’s rankings was measured using
a standard survey. This survey contained 12 questions: each
responding to a set of rankings created by one of the four in-
groups in response to one of the three questions. Each question
required the participant to rank their satisfaction with each of
the three prioritizations that the in-group made using the
Majority, Borda Count, and ASI methods.
As an example, one question in this survey asked out-group
participants to rank their satisfaction with each of in-group 1’s
question 1 Majority, Borda Count, and ASI prioritizations.
All participants completed the survey fully before joining the
groups on the Swarm AI platform; no survey data were missing
at the time of analysis. All questions were answered in between
10 and 60 seconds.
A. Question Divisiveness
Overall, each of the three questions proved to be highly
divisive when segmenting by the Political Affiliation of
participants (Conservative or Labour). Significant and important
differences in ranking (p<0.05) were observed in each of the
three questions when segmenting this way, as shown in Table 2.
Table 2 summarizes the number of statistically significant
differences between the Labour and Conservative average
ranking of each of the 6 items in each question. These
differences were calculated using a 2-sample t-test for each item.
Notably, the Political Affiliation of participants significantly
impacted their answers to all three questions.
The average effect size for all significant differences,
calculated as the average difference between Conservative and
Labour rankings of each of the significantly different items, is
reported. Significant differences were observed when the
average ranking difference was at minimum 0.83 ranks, or
16.6% of the maximum observable difference, which is a
considerable difference in ranking.
Question Number
Number of Significant
Differences (p<0.05) by
Political Affiliation
1: Objectives
2: Issues
3: Immigration
Table 2: Number of Significant Differences Observed by demographic when
ranking priorities as individuals via online survey.
The average ranking difference of each item, and the
confidence interval of each difference, is shown in Figures 1-3.
The fact that half of the six items in each question were ranked
differently by Labour and Conservative voters in this pool
indicates that the questions formulated by the research team
were meaningfully divisive and therefore were suitable for use
in this experiment.
B. In-Group Satisfaction
Next, each in-group’s satisfaction with each ranking was
calculated, as shown in Appendix B as the average ranking that
was given to each list by participants in that group. A score of
1.0 indicates the most preferred list, while a score of 3.0
indicates the least preferred list. In Question 2, Group 1’s
Majority and ASI lists were the same, so only two independent
lists were ranked in this instance.
Analyzing the average satisfaction ranking for each method
across all groups, we find that the Swarm method was preferred
to the Majority Voting method for all three questions and was
significantly preferred to the Majority Voting method on the
second question alone (p<0.01). ASI also received a higher
average satisfaction ranking than the Borda Count method on
two of the three questions, though there were no significant
differences between the two methods on any of the three
Figs. 4-6: Average Ranking Differences between Conservative and Labour
Voters across each question in the study. 95% confidence intervals are shown.
C. Out-Group Satisfaction
Next, the average out-group satisfaction ranking with each of the
three decision-making methods was calculated for each
question. Then, the average rank of each of the three decision-
making methods was calculated across questions. As shown in
Appendix C, the ASI method outperformed the Majority
method, resulting in superior satisfaction ranking for three of the
four questions.
Interestingly, the ASI and Borda Count methods were the
most-favored method in two of the groups considered. In the
first group, which consisted of 4 Labour and 4 Conservative
members, the ASI method outperformed the majority and Borda
count methods, though the difference between the satisfaction
rankings was not significant when measured with a paired t-test.
In the second group, which consisted of 10 Labour and 10
Conservative members, the Borda count significantly
outperformed the ASI and Majority methods (p=0.015,
p=5.9E-5 respectively). In the third group, which consisted of 8
Conservative and 12 Labour members, the Borda count
significantly outperformed the ASI and Majority methods
(p=0.0026, p=0.0040 respectively). In the fourth group, which
consisted of 12 Conservative and 8 Labour members, the ASI
method significantly outperformed the Borda Count and
Majority methods (p=0.047, p=0.0087 respectively).
Over all groups, the rankings generated by the ASI and
Borda count methods both significantly outperformed those
arising from the Majority method (p=2.28E-5, p=1.66E-7
respectively). When comparing the ASI and the Borda Count
methods, the Borda Count method marginally outperformed the
Swarm method, though this effect was not significant (p=0.233).
D. In-Group vs Out-Group Satisfaction
It is instructive to compare the satisfaction of the in-group with
that of the out-group: the participants who took part in ranking
the policies (the in-group) may be more satisfied with the
rankings they generated than the participants who had no say
(the out-group). The in-group and out-group satisfaction
rankings of each decision method can be compared by
subtracting each decision method’s average ranking in the out-
group from the same ranking in the in-group, as shown in
Appendix D. The ASI was the only decision-method that was on
average ranked as more preferable by the out-group than the in-
group; however, the size of this effect was small.
E. Sampling Bias
Often, policy decisions are not made with a representative
group of decision-makers: the elected officials that make policy
decisions may be composed of different demographics than the
electorate whom the officials represent. The effect where one
group is sampled from a wider population, but ends up with a
different demographic composition, may be referred to as
sampling bias.
To investigate the impact of sampling bias of the in-group
on these results, the satisfaction of the out-group was measured
when composed of varying ratios of Labour and Conservative
participants. Ratios between 25% Labour / 75% Conservative
and 75% Labour / 25% Conservative were tested in 5%
increments. To create a robust estimate of the out-group’s
satisfaction with a particular political composition, participants
in the out-group were resampled with replacement 1000 times.
The average ranking of each method’s lists across all
questions is shown for groups 1-4 in Figures 4-7. The 60%
confidence intervals of the average ranking of each method
were calculated to give a sense of the uncertainty of the result,
and is shown as the shaded interval around each line.
Figures 7-10: Bootstrapped Average Satisfaction Ranking of Out-Group, by
the Percent of the Resample Identifying as Labour Voters. Shaded areas
indicate 60% confidence intervals.
The 90% confidence intervals of these same graphs are
included in Appendix E.
In these graphs, a decision method with a positive slope
indicates that Labour voters favored that method’s lists more
than Conservative members - so including more Labour voters
as part of the Out-Group sample led to a higher average group
satisfaction with that method’s rating. There’s no clear trend
from these charts that any decision method was favored by either
Labour or Conservative members across the board.
There is, however, weak evidence that the preference of the
out-group with the results of the ASI and Borda Count methods
(as compared to the Majority method) are stable in the face of
out-group sampling bias: the optimal decision method for each
question is unchanged even when the Out-group’s composition
is changed by 25% or more. In addition, in only one case--when
group 3’s resampled out-group had fewer Labour voters than the
in-group--did the Majority method outperform the ASI method.
A two-sample t-test was used to measure the effect that these
demographic labels had on the in-group’s ranking of each of the
items in each of the three questions in this study. Table 3 below
shows a breakdown of the number of items which were ranked
significantly differently (p<0.05) for each question. For the
Brexit Stance column, only the Leave and Remain demographic
labels were used, since there was only one person in the in-group
that self-reported a Brexit Stance of “Undecided”.
1: UK Government Objectives
2: Government Issues
3: Immigration Policy
Table 3: Number of Significant Differences Observed by demographic when
ranking priorities as individuals via online survey.
Although Political Affiliation was the demographic that showed
the largest number of significant ranking differences overall,
Brexit Stance showed the most significant differences on a
single question: the Immigration Policy question found 5 of 6
(83%) of items were ranked differently by the Leave and
Remain members of the in-group. Gender was not meaningfully
related to the ranking of items on this test. Only one item
(“Globalization” in the Government Issues questions) was
ranked significantly differently by men and women in this test.
A Chi-squared analysis was then conducted to measure whether
each self-reported demographic was indeed independent of one
another. Over the 67 members of the in-group, Gender was not
significantly correlated with either Political Affiliation or Brexit
Stance. Political Affiliation, however, was significantly
correlated with the Brexit Stance of participants (p=0.018):
Conservative participants made up 66% of participants who
identified as wanting to Leave the EU, while Labour participants
made up 75% of participants who identified as wanting to
Remain in the EU.
Over the three politically divisive questions in this study, and
over the four groups that answered these questions, the Majority
voting method was regularly the least preferred voting method.
This was a surprising finding, as the Majority voting method is
the most common method of aggregating voters preferences in
modern democracies. The Borda Count and ASI methods were
similarly preferable in most groups, though the ASI method had
slightly higher satisfaction levels among the out-group. This
result suggests that, of these three methods, the Borda Count
and ASI methods may be promising candidates for general
public votes and for representative voting structures, which by
their nature are subject to a degree of sampling bias.
Future work could collect more data with greater power to
detect differences between the methods. Other studies could
take steps beyond this study, investigating the group-level and
individual psychological effects of using ASI and survey-based
voting methods to make group decisions or prioritizations: does
ASI enable individuals to feel a higher level of buy-in to the
group’s decisions than survey methods? Are the decisions
executed satisfactorily more frequently when decisions are
made on the Swarm AI platform? Do people feel like their
views are more represented in the group’s decision when using
ASI vs traditional methods? Anecdotal evidence suggests this
may be the case, but a rigorous study has yet to be conducted.
Another interesting avenue for future work includes comparing
Swarm AI to a real-time iterative Majority voting framework
[26]since Swarm AI can be seen as a continuous voting
framework. It would also be interesting to consider the effect
on the results of deliberation before voting, since deliberation
would open the doors up to both collective reasoning and
strategic voting of many kinds. Other voting systems that take
account of the full ordering of preferences, such as Preferential
Voting, may generate equivalent levels of satisfaction.
Finally, future work may compare ASI to a Delphi
method. Delphi methods use iterated rounds of deliberation and
voting to refine group consensus [29]. While on the surface
Delphi and ASI methods may sound similar, there are deep
structural differences in the two methods that make the
comparison interesting for future research: Delphi methods
often take weeks to perform, but enable groups to directly
deliberate and discuss problems [30], while ASI requires less
than a minute for a group of any size to reach an answer, and
features less language-based deliberation.
Thanks to Chris Hornbostel for his efforts in coordinating
the swarms. Also, thanks to Unanimous AI for the use of the
Swarm platform for this ongoing work. Contact Unanimous AI
or the authors for more information on Swarm AI and to use
this technology in a research capacity. This work was funded
[1] Galton, F. (1907). Vox Populi. Nature, 75, 450-451.
[2] Steyvers, M., Lee, M.D., Miller, B., & Hemmer, P. (2009). The Wisdom
of Crowds in the Recollection of Order Information. In Y. Bengio and D.
Schuurmans and J. Lafferty and C. K. I. Williams
[3] Philip E. Tetlock and Dan Gardner. 2015. Superforecasting: The Art and
Science of Prediction. Crown Publishing Group, New York, NY, USA.
[4] J Dana, P Atanasov, P Tetlock, B Mellers (2019), Are markets more
accurate than polls. The surprising informational value of “just asking.”
Judgment and Decision Making 14 (2), 135-147
[5] Rosenberg, L.B., “Human Swarms, a real-time method for collective
intelligence.” Proceedings of the European Conference on Artificial Life
2015, pp. 658-659
[6] Rosenberg, Louis. “Artificial Swarm Intelligence vs Human Experts,”
Neural Networks (IJCNN), 2016 International Joint Conference on. IEEE.
J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol.
2. Oxford: Clarendon, 1892, pp.6873.
[7] Rosenberg, Louis. Baltaxe, David and Pescetelli, Nicollo. "Crowds vs
Swarms, a Comparison of Intelligence," IEEE 2016 Swarm/Human
Blended Intelligence (SHBI), Cleveland, OH, 2016, pp. 1-4.
[8] Baltaxe, David, Rosenberg, Louis and N. Pescetelli, “Amplifying
Prediction Accuracy using Human Swarms”, Collective Intelligence
2017. New York, NY ; 2017.
[9] Willcox G., Rosenberg L., Askay D., Metcalf L., Harris E., Domnauer C.
(2020) Artificial Swarming Shown to Amplify Accuracy of Group
Decisions in Subjective Judgment Tasks. In: Arai K., Bhatia R. (eds)
Advances in Information and Communication. FICC 2019. Lecture Notes
in Networks and Systems, vol 70. Springer, Cham
[10] L. Rosenberg, N. Pescetelli and G. Willcox, "Artificial Swarm
Intelligence amplifies accuracy when predicting financial markets," 2017
IEEE 8th Annual Ubiquitous Computing, Electronics and Mobile
Communication Conference (UEMCON), New York City, NY, 2017, pp.
[11] L. Rosenberg and G. Willcox, "Artificial Swarm Intelligence vs Vegas
Betting Markets," 2018 11th International Conference on Developments
in eSystems Engineering (DeSE), Cambridge, United Kingdom, 2018, pp.
[12] L. Rosenberg, M. Lungren, S. Halabi, G. Willcox, D. Baltaxe and M.
Lyons, "Artificial Swarm Intelligence employed to Amplify Diagnostic
Accuracy in Radiology," 2018 IEEE 9th Annual Information Technology,
Electronics and Mobile Communication Conference (IEMCON),
Vancouver, BC, 2018, pp. 1186-1191.
[13] Seeley T.D, Buhrman S.C 2001 “Nest-site selection in honey bees: how
well do swarms implement the ‘best-of-N’ decision rule?” Behav. Ecol.
Sociobiol. 49, 416427
[14] Marshall, James. Bogacz, Rafal. Dornhaus, Anna. Planqué, Robert.
Kovacs, Tim. Franks, Nigel. “On optimal decision-making in brains and
social insect colonies.” Soc. Interface 2009.
[15] Seeley, Thomas D., et al. "Stop signals provide cross inhibition in
collective decision-making by honeybee swarms." Science 335.6064
(2012): 108-111.
[16] Seeley, Thomas D. Honeybee Democracy. Princeton Univ. Press, 2010.
[17] 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.” Behavioural Ecology and Sociobiology 54 (5) 511-520.
[18] Usher, M. McClelland J.L 2001 “The time course of perceptual choice:
the leaky, competing accumulator model.” Psychol. Rev. 108, 550–592
[19] L. Rosenberg, G. Willcox, "Artificial Swarms find Social Optima: (Late
Breaking Report)", 2018 IEEE Conference on Cognitive and
Computational Aspects of Situation Management (CogSIMA), pp. 174-
178, 2018.
[20] Laslier JF. (2012) And the Loser Is… Plurality Voting. In: Felsenthal D.,
Machover M. (eds) Electoral Systems. Studies in Choice and Welfare.
Springer, Berlin, Heidelberg
[21] Poundstone, W. (2009): “Gaming the Vote: Why Elections Aren’t Fair
(and What We Can Do About It)” pp. 79, 206, 230.
[22] Condorcet, Jean-Antoine-Nicolas de Caritat, marquis de, 1743-1794
Essai sur l'application de l'analyse la probabilit des dcisions rendues
la pluralit des voix", Pre-1801 Imprint Collection (Library of
Congress) DLC
[23] Merijn Van Erp and Lambert Schomaker (2000) Variants Of The Borda
Count Method For Combining Ranked Classifier Hypotheses. In the
seventh Intenational workshop on frontiers in handwriting recognition
[24] L. Rosenberg, D. Baltaxe, “Setting Group Priorities - Swarms vs Votes”
(2016) IEEE Conference on Swarm Human Blending
[25] Kolm SC. (1993) The Impossibility of Utilitarianism. In: Koslowski P.,
Shionoya Y. (eds) The Good and the Economical. Studies in Economic
Ethics and Philosophy. Springer, Berlin, Heidelberg
[26] Burgman, M. A., Regan, H. M., Maguire, L. A., Colyvan, M., Justus, J.,
Martin, T. G., & Rothley, K. (2014). Voting systems for environmental
decisions. Conservation biology, 28(2), 322-332.
[27] Gibbard, Allan (1973). "Manipulation of voting schemes: A general
result". Econometrica. 41 (4): 587601. doi:10.2307/1914083. JSTOR
[28] Satterthwaite, Mark Allen (April 1975). "Strategy-proofness and Arrow's
conditions: Existence and correspondence theorems for voting procedures
and social welfare functions". Journal of Economic Theory. 10 (2): 187
217. CiteSeerX doi:10.1016/0022-0531(75)90050-2.
[29] Pill, Juri (Febuary 1971). “The Delphi method: Substance, context, a
critique and an annotated bibliography”. Socio-Economic Planning
Sciences. 5 (1): 57-71.
[30] Boulkedid, R., Abdoul, H., Loustau, M., Sibony, O., & Alberti, C. (2011).
Using and reporting the Delphi method for selecting healthcare quality
indicators: a systematic review. PloS one, 6(6), e20476.
[31] Rosenberg, L., Willcox, G. (2019). Artificial Swarm Intelligence.
Unanimous AI,
[32] How does Swarm work? -. (n.d.). Retrieved from
1) Rank the following UK Government objectives in order of their importance (1=most important, 2=next most important, etc.).
*Address climate change
*Drive economic growth
*Fix immigration policy
*Reduce crime
*Reduce poverty
*Solve the housing crisis.
2) Rank the following issues in order of the priority the UK government should give them. (1=highest priority, 2=second highest
priority, etc.)
*Fake News (misinformation in media)
*Gender inequality
*Income inequality
3) Rank the following proposed [immigration] policies in order of your preference. (1=most preferred, 2=next most preferred,
*Allow free movement from the EU
*Allow free movement internationally (non EU)
*Improve how migrants are integrated
*Reduce all net immigration to under 100,000
*Reduce net immigration from the EU
*Reduce net immigration from the non-EU world
*=(p<0.05) as compared to the Majority Rating, **=(p<0.01) as compared to the Majority Rating
In-Group Satisfaction Ranking by Question Number
Question 1
(UK Govt. Objectives)
Question 2
(Government Issues)
Question 3
Group 1
Group 2
Group 3
Group 4
All Groups
# Conservative
# Labour
Group 1
Group 2
Group 3
Group 4
Ω = significant at the 0.05 level relative to the Swarm
Ψ = significant at the 0.01 level relative to the Swarm
* = significant at the 0.05 level relative to the Borda Count
**=significant at the 0.01 level relative to the Borda Count
Group 1
Group 2
Group 3
Group 4
... In parallel with the engineering of brain interface technologies, there have been impressive developments in the field of mapping collective behaviours in animals, and in programming algorithms to mimic these behaviours to improve outcomes in collective tasks. Sometimes referred to as "swarming" or "swarm intelligence", applied tools utilizing these algorithms have been shown to improve decisionmaking across a diverse range of domains in experimental studies, ranging from predicting outcomes of sports events to improving pneumonia diagnosis, and organizing policy priorities [41,46,59]. These developments are part of a broader line of research into "hybrid intelligence" -networks of humans and computers thinking together. ...
Full-text available
A growing number of technologies are currently being developed to improve and distribute thinking and decision-making. Rapid progress in brain-to-brain interfacing and swarming technologies promises to transform how we think about collective and collaborative cognitive tasks across domains, ranging from research to entertainment, and from therapeutics to military applications. As these tools continue to improve, we are prompted to monitor how they may affect our society on a broader level, but also how they may reshape our fundamental understanding of agency, responsibility, and other key concepts of our moral landscape. In this paper we take a closer look at this class of technologies – Technologies for Collective Minds – to see not only how their implementation may react with commonly held moral values, but also how they challenge our underlying concepts of what constitutes collective or individual agency. We argue that prominent contemporary frameworks for understanding collective agency and responsibility are insufficient in terms of accurately describing the relationships enabled by Technologies for Collective Minds, and that they therefore risk obstructing ethical analysis of the implementation of these technologies in society. We propose a more multidimensional approach to better understand this set of technologies, and to facilitate future research on the ethics of Technologies for Collective Minds.
Full-text available
Psychologists typically measure beliefs and preferences using self-reports, whereas economists are much more likely to infer them from behavior. Prediction markets appear to be a victory for the economic approach, having yielded more accurate probability estimates than opinion polls or experts for a wide variety of events, all without ever asking for self-reported beliefs. We conduct the most direct comparison to date of prediction markets to simple self-reports using a within-subject design. Our participants traded on the likelihood of geopolitical events. Each time they placed a trade, they first had to report their belief that the event would occur on a 0-100 scale. When previously validated aggregation algorithms were applied to self-reported beliefs, they were at least as accurate as prediction-market prices in predicting a wide range of geopolitical events. Furthermore, the combination of approaches was significantly more accurate than prediction-market prices alone, indicating that self-reports contained information that the market did not efficiently aggregate. Combining measurement techniques across behavioral and social sciences may have greater benefits than previously thought.
Conference Paper
Full-text available
In the natural world, Swarm Intelligence (SI) is a commonly occurring process in which biological groups amplify their collective intelligence by forming closed-loop systems. It is well known in schools of fish, flocks of bird, and swarms of bees. In recent years, new AI technologies have enabled networked human groups to form systems modeled after natural swarms. Known as Artificial Swarm Intelligence (ASI), the technique has been shown to amplify the effective intelligence of human groups. This study compares the predictive ability of ASI systems against large betting markets when forecasting sporting events. Groups of average sports fans were tasked with predicting the outcome of 200 hockey games (10 games per week for 20 weeks) in the NHL. The expected win rate for Vegas favorites was 62% across the 200 games based on the published odds. The ASI system achieved a win rate of 85%. The probability that the system outperformed Vegas by chance was extremely low (p = 0.0057), indicating a significant result. In addition, researchers compared the winnings from two betting models - one that wagered weekly on the Vegas favorite, and one that wagered weekly on the ASI favorite. At the end of 20 weeks, the Vegas model generated a 41% financial loss, while the ASI model generated a 170% financial gain.
Full-text available
New technologies enable distributed human teams to form real-time systems modeled after natural swarms. Often referred to as Artificial Swarm Intelligence (ASI) or simply “human swarming”, these real-time systems have been shown to amplify group intelligence across a wide range of tasks, from handicapping sports to forecasting financial markets. While most prior research has studied human swarms with 20–100 members, the present study explores the ability of ASI to amplify accuracy in small teams of 3–6 members. The present study also explores if conducting multiple swarms and aggregating by taking a “vote of swarms” can further amplify the accuracy. A large set of 66 small teams were engaged in this study. Each team was given a standard subjective judgement test. Participants took the test both as individuals and real-time swarms. The average individual scored 69% correct, while the average swarm scored 84% correct (p < 0.001). In addition, aggregation of multiple swarms revealed additional amplifications of accuracy. For example, by randomly selecting sets of 3 swarms and aggregating by plurality vote, average accuracy increased to 91% (p < 0.001). These results suggest that when small teams make subjective judgements as real-time swarms, they can be significantly more accurate than individual members, and that their accuracy can be further amplified by aggregating the output across small sets of swarms.
Conference Paper
Full-text available
Swarm Intelligence (SI) is a biological phenomenon in which groups of organisms amplify their combined brainpower by forming real-time systems. It has been studied for decades in fish schools, bird flocks, and bee swarms. Recent advances in networking and AI technologies have enabled distributed human groups to form closed-loop systems modeled after natural swarms. The process is referred to as Artificial Swarm Intelligence (ASI) and has been shown to significantly amplify group performance. The present research applies ASI technology to the field of medicine, exploring if small groups of networked radiologists can improve their diagnostic accuracy when reviewing chest X-rays for the presence of pneumonia. Performance data was collected for individual radiologists generating diagnoses alone, as well as for small groups of radiologists working together to generate diagnoses as a real-time ASI system. Diagnoses were also collected from a state-of-the-art deep learning system (CheXNet) developed at Stanford University. Results showed that small groups of networked radiologists, when working as a real-time ASI system, were significantly more accurate than the individual radiologists on their own, reducing diagnostic errors by 33%. Results also showed that small groups of networked radiologists, when working as an ASI system, were significantly more accurate (22%) than a state-of-the-art deep learning system (CheXNet).
Conference Paper
Full-text available
In the natural world, many social species amplify their collective intelligence by forming real-time closed-loop systems. Referred to as Swarm Intelligence (SI), this phenomenon has been rigorously studied in schools of fish, flocks of birds, and swarms of bees. In recent years, technology has enabled human groups to form real-time closed-loop systems modeled after natural swarms and moderated by AI algorithms. Referred to as Artificial Swarm Intelligence (ASI), these methods have been shown to enable human groups to reach optimized decisions. The present research explores this further, testing if ASI enables groups with conflicting views to converge on socially optimal solutions. Results showed that “swarming” was significantly more effective at enabling groups to converge on the Social Optima than three common voting methods: (i) Plurality voting (i) Borda Count and (iii) Condorcet pairwise voting. While traditional voting methods converged on socially optimal solutions with 60% success across a test set of 100 questions, the ASI system converged on socially optimal solutions with 82% success (p<;0.001).
Conference Paper
Full-text available
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.
Swarm Intelligence (SI) is a natural phenomenon that enables social species to quickly converge on optimized group decisions by interacting as real-time closed-loop systems. This process, which has been shown to amplify the collective intelligence of biological groups, has been studied extensively in schools of fish, flocks of birds, and swarms of bees. This paper provides an overview of a new collaboration technology called Artificial Swarm Intelligence (ASI) that brings the same benefits to networked human groups. Sometimes referred to as "human swarming" or building "hive minds," the process involves groups of networked users being connected in real-time by AI algorithms modeled after natural swarms. This paper presents the basic concepts of ASI and reviews recently published research that shows its effectiveness in amplifying the collective intelligence of human groups, increasing accuracy when groups make forecasts, generate assessments, reach decisions, and form predictions. Examples include significant performance increases when human teams generate financial predictions, business forecasts, subjective judgments, and medical diagnoses.