Warned alike, AI agents avoid the less-crowded road while people take it

Takahiro Ezaki; Naoto Imura; Katsuhiro Nishinari

physics.soc-phSep 25, 2026

Warned alike, AI agents avoid the less-crowded road while people take it

Authors: Takahiro Ezaki, Naoto Imura, Katsuhiro Nishinari

Organizations: Research Center for Advanced Science and Technology, The University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8904, Japan · Department of Aeronautics and Astronautics, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan

Abstract

AI agents built on a few shared models increasingly act for many people. A shared forecast about others can align their choices and change how scarce capacity is allocated. We tested this feedback in a two-road congestion game. Adding one sentence warning that others might follow a routing tip made populations of 50 GPT agents crowd one road while avoiding the nearly empty alternative. Average travel time rose from 64 to 95 min, although any crowded-road agent could have saved 69 min by switching alone. The warning discouraged the very move it predicted. The pattern persisted for 100 rounds. Two other model families shifted the same way without locking onto one road. Twelve all-human groups (240 participants) stayed near balance under numerical reports or the tip and warning. In 24 mixed groups with a further 240 participants, imbalance grew with the share of agents in the registered analysis, while people increasingly took the road the agents avoided. Collective costs stayed below the allagent reference, but with 15 agents and 5 humans, agent seats averaged 80 min, compared with 44 min for human seats. Shared forecasts can thus sustain collective inefficiency among similar agents. A better group average can also hide an unequal burden. Evaluations of AI agents that share resources should test populations, treat messages as interventions and report who bears the costs.

Figures & tables

Figure 1: A shared warning produces persistent concentration in the primary model population. ( A ) The game: two identical roads,

N

commuters choosing simultaneously every round, travel time increasing linearly with a road’s share of commuters (inset). ( B ) The four daily broadcasts (verbatim). ( C to F ) For each condition of the prespecified core (50 GPT agents, 60 rounds, ten seeds), the upper strip shows the share of agents on physical Route A in one representative run (seed 3): lockstep oscillation under no report, damped oscillation under numbers, near-balance under the tip, and persistent concentration on one road under the tip plus warning. The main panel shows the share choosing the road that was less crowded in the previous round,

\hat{f}_{t}

, for all ten runs (thin lines; thick line, median; number, median of run-mean travel time). ( G ) The ten F3 runs continued to round 100 (shaded). ( H ) Route imbalance

I

against switching fraction

M

for all 40 runs (filled) and for the multinomial-logit learning baseline on the same seeds (hollow); the shaded region marks the descriptive frozen criterion. ( I ) Mean travel time per run (bars, medians; dashed, balanced use; dotted, expectation for independent coin flips). ( J ) Paired F2

\rightarrow

F3 change in mean travel time by seed.

Figure 2: The collective state follows the message currently in force. ( A to C )

\hat{f}_{t}

when the broadcast is changed at round 31 (dashed line): the tip plus warning replaced by numerical reports (F3

\rightarrow

F1; A, five seeds), numerical reports replaced by the tip plus warning (F1

\rightarrow

F3; B, five seeds), and the warning given to 38 of 50 agents while the rest keep receiving numbers (C, three seeds); numbers give mean travel time in rounds 16–30 and 36–50. In A, 96–100% choose the previously less-crowded road in the first post-switch round; 86–98% actually change roads, and congestion remains high before costs subsequently decline. ( D , E ) Imbalance and mean travel time against the fraction of agents receiving the warning while the rest receive numbers (points, runs; lines, medians; diamonds, pure populations); dashed curves, linear mixing of the two pure populations’ tip-following shares with no fitted parameter, which approximates the shape but under-predicts cost near the minimum. ( F ) Appending yesterday’s travel times to the tip or to the tip plus warning:

\langle\bar{f}\rangle

and the median run-mean travel time per cell. ( G ) Run-mean travel time (dot, median across runs; bar, range) for message, history and model variants; windows are rounds 6–40 for the 20-seat history variants and 6–60 for the remaining runs.

\langle\bar{f}\rangle

is shown where round-level data are available. ( H ) Share of agents’ stated reasons that invoke other drivers switching to or crowding a road (solid) or the agent’s own slow recent commute (light), seed 3, 2,750 decisions per condition (coding rules in Materials and Methods).

Figure 3: Average response and temporal variation contribute to cost, while individual response patterns differ. ( A ) Mean travel time on tie-free analysis rounds against

\bar{f}

for the core, matched and cross-provider/numeric all-agent runs (color, message; marker, model family), for the twelve all-human rooms (stars) and for the 24 mixed rooms of Fig. 5 (hollow diamonds; room-level share over all seats). The dashed curve is the zero-variance reference from Eq. 1 . Both axes use the same rounds, so each point lies above it by exactly

160\,\mathrm{Var}_{\mathcal{R}}(\hat{f}_{t})

. ( B )

\langle\bar{f}\rangle

under the tip and under the tip plus warning for each model family (five to ten runs per point): the warning lowered average tip-following in all three families, from different baselines. ( C ) Decomposition of excess travel time into the average-response component,

160(\bar{f}-\tfrac{1}{2})^{2}

, and temporal variation,

160\,\mathrm{Var}(\hat{f}_{t})

; components are calculated within each run or room and then averaged with equal weight. For Claude and Gemini, their opposing changes reduce the net cost increase. ( D ) Distribution of individuals’ tendency to follow the tip (share of tie-free rounds 6–40), as a proportion of individuals, for 120 humans (20

\times

six rooms) and 100 GPT agents (20

\times

five runs) per condition in 20-seat groups; annotations give the SD, the SD expected for identical choosers at the same mean, and the split-half correlation of individual tendencies. ( E ) Mean individual probability of switching after a slow (

>60

min) or fast (

<60

min) day, for transitions from rounds 6–39 to 7–40. Individuals with no qualifying previous outcome are omitted from that conditional mean. ( F ) Verbatim reasons given by GPT agents under F3.

Figure 4: Humans in the same game. ( A to D )

\hat{f}_{t}

in six human rooms per condition (Prolific, 20 participants, 40 rounds) and in ten GPT runs of the same size; thin lines, rooms or runs; thick lines, round-wise medians; inset travel times, means across rooms or runs. Gaps mark rounds in which the previous counts were equal. ( E , F ) Imbalance and mean travel time per room or run (points; bars, means); dotted, expectation for independent coin flips; dashed, balanced use. ( G ) Participants’ self-reported usual response to the report in the all-human rooms and in the mixed rooms of Fig. 5.

Figure 5: Human choices change within mixed rooms, with unequal costs by seat type (24 rooms). ( A ) Route imbalance in rooms of 20 with

K

GPT agents, all under F3 (purple, one point per completed room; bars, means), against the prediction if humans kept their all-human lean (dashed; linear mixing), the lower bound if all agents sit on one road (solid; accounting bound), and the bound given the agents’ actual choices (crosses; mean per

K

); non-concurrent all-human and all-agent reference groups are shown at

K=0

and

K=20

. ( B ) Tip-following share of the human seats and of the agent seats in the same rooms (tie-free rounds 6–40). ( C ) Mean travel time of the human seats, of the agent seats and of the room; dashed, the 60 min of equal sharing. ( D ) Share of human seats (purple, by

K

) and of agent seats (orange) choosing the tipped road, round by round, averaged over rooms. ( E ) Human tip-following per room in the specified early (rounds 6–20) and late (26–40) windows; black segments join composition means. Panels summarize all eight rooms per composition; registered block-level inference and dependence sensitivities are reported in the text.

Figure S1: Physical route shares in the prespecified core. Share of agents on physical Route A in every run of the four conditions (50 GPT agents, 60 rounds, ten seeds per condition). Road labels are mapped to physical roads with a seed-dependent parity; frozen runs can therefore concentrate on either physical road.

Figure S2: Cross-provider extension, all paired seeds. Route imbalance, switching fraction, mean travel time and tip-following share for Claude Haiku 4.5 ( A to D ) and Gemini 3.5 Flash ( E to H ) under the tip and under the tip plus warning, five matched seeds each.

Figure S3: Drift control and numeric variants by seed. September 2026 runs compared with August 2026 runs at the same environment seed. Filled points repeat the original F3 condition; hollow points add numerical reports to F3 or F2. The diagonal marks equal outcomes, not identical model samples.

Figure S4: Exposure grid and exposure switch. ( A , B ) Every run of the 20- and 50-agent exposure grids. ( C ) Physical Route A share in the three runs in which the warning was given to 38 of 50 agents from round 31.

Figure S5: Human rooms, physical route share. Share of participants on physical Route A in the six F1 rooms ( A ) and six F3 rooms ( B ).

Figure S6: Individual tip-following tendencies of 50-agent GPT populations. Each agent’s share of tie-free rounds 6–60 on the road that was less crowded in the previous round, seed 3, under numbers ( A ), the tip ( B ) and the tip plus warning ( C ).

Figure S7: Every room of the mixed-group experiment. Columns, $K=5$ , 10 and 15 GPT agents among 20 seats; first eight rows, blocks 1–8; bottom row, the two original aborted attempts. R and F mark valid rooms from the 16 and 17 September collections, respectively. Purple, share of human seats on the tip road; orange, share of GPT seats on it (both undefined on tie rounds); dashed, room imbalance scaled to 1 when everyone is on one road. Titles give $I$ and the mean travel time of the room and of each seat type over rounds 6–40. Aborted rooms show the rounds played before the server fault.

Figure S8: Design and execution record of the completed mixed-group experiment. Initial and current status of the 24 registered slots by block and $K$ , showing the six valid additional rooms and retaining the two original aborted attempts in the history. All eight blocks are complete (Text S8).

Figure S9: Post-task estimates of the automated share. Actual percentages of 25%, 50% and 75% correspond to $K=5$ , 10 and 15 of 20 players. ( A ) Formal numeric answers from 116, 76 and 38 respondents; horizontal bars are respondent means. ( B ) The eight room means per composition; bars weight rooms equally. Horizontal jitter only separates points. Dashed lines show accurate estimates. Ten missing answers, including one unsubmitted numeric draft, are omitted; all 240 humans remain in the behavioral analysis. These are descriptive distributions, not independent treatment replicates.

Figure S10: Reasoning and current-model extension. ( A to D ) Route imbalance, switching fraction, mean travel time and tip-following share (rounds 6–60) for each request setting under the tip (F2, green) and under the tip plus warning (F3, orange), five paired seeds each. Gray lines join runs of the same seed; black bars mark means over the five runs. The first setting is the core reference at the same seeds. The numbers at the top of A count runs meeting the frozen criterion, whose thresholds for $I$ and $M$ are dashed (it also requires ten consecutive rounds with the same majority road). Reasoning-enabled requests omitted temperature; its internal numerical value was not independently verified. ( E ) Share of agents on physical Route A in every run.

Population and message	$N$	rounds	runs	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	frozen
GPT, F0 no report	50	6–60	10	0.476	0.993	96.4	0.96	0/10
GPT, F1 numbers	50	6–60	10	0.230	0.660	71.5	0.69	0/10
GPT, F2 tip	50	6–60	10	0.134	0.713	64.4	0.54	0/10
GPT, F3 tip + warning	50	6–60	10	0.465	0.070	94.8	0.04	10/10
MNL baseline, F0	50	6–60	10	0.063	0.496	61.0	–	0/10
MNL baseline, F1	50	6–60	10	0.437	0.881	90.7	–	0/10

Table S1: Condition-level summary. Medians across runs or rooms of run-mean outcomes;

\mathbb{E}[f_{t}]

is the mean across runs of the run-mean tip-following share (tie-free rounds); “frozen” counts runs meeting the descriptive classifier. MNL, multinomial-logit learning baseline.

Condition	seed	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	$\mathrm{Var}(f_{t})$
F0	3	0.478	0.993	96.7	0.978	0.0006
F0	4	0.433	0.993	90.1	0.933	0.0005
F0	5	0.474	0.995	96.0	0.974	0.0003
F0	6	0.492	0.995	98.9	0.992	0.0009
F0	7	0.487	0.990	98.3	0.987	0.0018
F0	8	0.414	0.988	87.6	0.914	0.0011

Table S2: All 40 runs of the prespecified core (50 GPT agents, rounds 6–60; the core was fixed by an execution-time code freeze, not publicly registered).

Model	message	seed	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	state
GPT-5.4-mini	tip + numbers	3	0.334	0.756	80.1	0.833	oscillation
GPT-5.4-mini	tip + numbers	4	0.287	0.715	76.1	0.787	oscillation
GPT-5.4-mini	tip + numbers	5	0.337	0.761	80.7	0.837	oscillation
GPT-5.4-mini	tip + numbers	6	0.329	0.776	79.5	0.829	oscillation
GPT-5.4-mini	tip + numbers	7	0.329	0.768	79.6	0.829	oscillation
GPT-5.4-mini	tip + warning + numbers	3	0.402	0.192	86.4	0.098	frozen

Table S3: All 33 runs of the September 2026 extension (50 agents, rounds 6–60). State labels follow the descriptive classifier in Text S4; “partial” denotes runs meeting none of its frozen, oscillation or near-equilibrium criteria.

Variant	runs	$I$	$M$	time (min)	frozen
F3, full history (rounds 6–40, 20 seats)	5	0.479	0.041	96.9	5/5
F1 + past tips in history (20 seats)	5	0.247	0.492	72.6	0/5
F3 + past tips in history (20 seats)	5	0.497	0.005	99.6	5/5
F3, stronger wording (50 agents)	3	0.499	0.003	99.8	3/3
Claude/Gemini, F0 (2 seeds each)	4	0.249	0.835	73.3	0/4
Claude/Gemini, F1 (2 seeds each)	4	0.361	0.917	83.1	0/4

Table S4: Robustness variants (means across runs; from the audited run table). Rounds 6–40 for 20-seat variants and 6–60 otherwise.

Room	block	message	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	tie rounds
pure-b1-2	1	F1	0.089	0.420	61.9	0.527	4
pure-b1-3	1	F1	0.086	0.467	61.8	0.508	5
pure-b2-roomwise-r1-2	2	F1	0.094	0.353	62.3	0.505	4
pure-b2-roomwise-r1-4	2	F1	0.093	0.457	62.0	0.503	5
pure-b3-roomwise-r3-1	3	F1	0.081	0.397	61.7	0.524	4
pure-b3-roomwise-r3-2	3	F1	0.103	0.484	62.2	0.517	2

Table S5: Human rooms (rounds 6–40; carried choices retained).

\mathbb{E}[f_{t}]

is the share of participants on the tip road over tie-free rounds; the last column counts the analyzed rounds in which the previous counts were equal.

Item / response	F1, all human	F3, all human	F3, with agents
Report influence: not at all / slightly / moderately / strongly	5 / 16 / 50 / 47	3 / 20 / 42 / 54	9 / 41 / 90 / 98
Usual response: faster-looking road / other road expecting others to switch / ignored / it depended	35 / 62 / 4 / 17	38 / 47 / 10 / 24	146 / 39 / 12 / 41
Rules understood: yes / mostly / no	104 / 12 / 2	108 / 11 / 0	202 / 33 / 3

Table S6: Questionnaire counts. All-human rooms: 120 participants per message (two F1 and one F3 participant did not answer). Mixed rooms: the 240 participants of the 24 completed rooms (two did not answer).

$N$	warned	runs	$I$	$M$	time	$\mathbb{E}_{\mathrm{mix}}[f_{t}]$	$\|\mathbb{E}_{\mathrm{mix}}[f_{t}]-\frac{1}{2}\|$	$T_{\mathrm{lin}}$	$I_{\mathrm{ind}}$	$T_{\mathrm{ind}}$
20	0	3	0.229	0.697	71.1	0.695	0.195	66.1	0.198	67.8
20	5	3	0.140	0.667	65.0	0.531	0.031	60.2	0.091	62.1
20	10	3	0.200	0.292	68.6	0.367	0.133	62.8	0.145	64.7
20	15	3	0.345	0.148	79.9	0.202	0.298	74.2	0.298	75.5
20	20	3	0.468	0.064	95.4	0.038	0.462	94.1	0.462	94.4
50	6	5	0.176	0.688	67.0	0.616	0.116	62.2	0.119	62.9

Table S7: Exposure grids, the linear-mixing rule and the independent-choice benchmark. Medians across runs of

I

M

and mean travel time;

\mathbb{E}_{\mathrm{mix}}[f_{t}]

mixes the pure-population leans of the core with no fitted parameter;

|\mathbb{E}_{\mathrm{mix}}[f_{t}]-\frac{1}{2}|

and

T_{\mathrm{lin}}

are the linear-mixing predictions and

I_{\mathrm{ind}}

T_{\mathrm{ind}}

the independent-choice benchmark at the same mean (Text S5).

Block	$K$	$I$	$M$	$M_{\mathrm{h}}$	$M_{\mathrm{a}}$	$\mathbb{E}[T_{t}]$	$\mathbb{E}[T_{\mathrm{h},t}]$	$\mathbb{E}[T_{\mathrm{a},t}]$	$\mathbb{E}[f_{\mathrm{h},t}]$	$\mathbb{E}[f_{\mathrm{a},t}]$	frozen
1	5	0.063	0.45	0.48	0.37	61.1	60.7	62.3	0.61	0.000	no
2	5	0.063	0.50	0.50	0.51	61.2	61.3	60.9	0.66	0.000	no
3	5	0.091	0.48	0.48	0.46	61.9	62.3	60.7	0.66	0.000	no
4	5	0.059	0.45	0.46	0.40	60.9	60.9	61.0	0.67	0.000	no
5	5	0.070	0.47	0.48	0.43	61.3	61.2	61.7	0.64	0.000	no
6	5	0.079	0.49	0.49	0.49	61.3	61.3	61.5	0.64	0.000	no

Table S8: Mixed rooms (rounds 6–40; carried choices retained). Subscripts h and a denote human and GPT seats;

\mathbb{E}[f_{t}]

is the share of that seat type on the tip road over defined-tip rounds. “Frozen” applies the descriptive classifier to the room. Tie-round ranges are given in Text S4. All 24 slots are complete; the two invalid original attempts are retained separately in the execution record (Figs. S7 and S8).

Arm	seed	$T_{30}$	$f_{31}$	$I_{31}$	$M_{31}$	$T_{31}$	$T_{32}$	$T_{33}$	$T_{34}$	$T_{35}$	$\mathbb{E}_{36:50}[T_{t}]$
F3 $\rightarrow$ F1	3	91.0	0.96	0.46	0.94	93.9	60.3	70.8	80.7	60.6	72.3
F3 $\rightarrow$ F1	4	80.7	1.00	0.50	0.86	100.0	63.1	61.6	69.2	64.1	71.3
F3 $\rightarrow$ F1	5	93.9	0.96	0.46	0.92	93.9	60.1	70.8	83.1	60.6	73.6
F3 $\rightarrow$ F1	6	88.2	0.98	0.48	0.90	96.9	70.8	60.3	76.4	61.0	73.1
F3 $\rightarrow$ F1	7	96.9	0.96	0.46	0.98	93.9	64.1	67.7	83.1	63.1	67.2
F1 $\rightarrow$ F3	3	91.0	0.00	0.50	0.06	100.0	100.0	93.9	93.9	96.9	95.4

Table S9: Switch experiments round by round.

T_{30}

, mean travel time in the last round under the original message;

f_{31}

I_{31}

M_{31}

and

T_{31}

, the first round under the new message;

T_{32}

T_{35}

, the following rounds;

\mathbb{E}[T_{t}]_{36\text{--}50}

, the prespecified post-intervention window.

Block	$b_{b}(I)$	$b_{b}(M)$	LOO mean $I$ slope	LOO mean $M$ slope
1	+0.0993	-0.2086	+0.0945	-0.2080
2	+0.1150	-0.2007	+0.0922	-0.2091
3	+0.0764	-0.2150	+0.0978	-0.2070
4	+0.0950	-0.2007	+0.0951	-0.2091
5	+0.0857	-0.2079	+0.0964	-0.2081
6	+0.0900	-0.2129	+0.0958	-0.2073

Table S10: Registered block slopes and leave-one-block-out estimates. Slopes are per five additional agents. Leave-one-block-out columns are mean slopes across the other seven blocks, without additional tests. The full-sample means are 0.0951 for

I

and

-0.2080

for

M

(Text S8).

Nuisance variable	$\gamma$	Mean $I$ slope	Mean $M$ slope
Carryover	-0.050	+0.0982	-0.2049
Carryover	-0.020	+0.0963	-0.2068
Carryover	-0.010	+0.0957	-0.2074
Carryover	-0.005	+0.0954	-0.2077
Carryover	+0.000	+0.0951	-0.2080
Carryover	+0.005	+0.0948	-0.2083

Table S11: Registered nuisance sensitivity grids. Each row reports the equally weighted mean block slope after subtracting the stated hypothetical nuisance effect from room outcomes. Carryover is binary and wait is measured in minutes. These are descriptive sensitivity calculations, not estimated causal corrections (Text S8).

$K$	Actual (%)	Responses	Mean (%)	SD	Median [IQR]	Bias (pp)	MAE (pp)
5	25	116/120	63.1	24.9	65.0 [50.0, 80.0]	+38.1	39.4
10	50	76/80	67.2	19.0	70.0 [50.0, 80.0]	+17.2	20.4
15	75	38/40	68.9	21.2	70.0 [60.0, 83.0]	-6.1	17.1

Table S12: Recovered formal estimates of the automated share. Each composition comprises eight rooms. Responses gives numeric respondents / all human participants. Means and SDs weight respondents equally; IQR gives the 25th and 75th percentiles. Bias is estimated minus actual percentage; MAE is mean absolute error, both in percentage points (pp). Missingness comprises eight unanswered items among submitted questionnaires and two unsubmitted questionnaires. No missing or draft-only value is imputed.

GPT-5.4-mini, no reasoning, temperature 0.7 (core runs at the same seeds)
Message	frozen	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	response	variation
tip	0/5	0.146	0.714	65.2	0.555	0.7	4.6
tip + warning	5/5	0.460	0.079	94.0	0.040	33.9	0.2
warning $-$ tip		+0.314 (5)	$-$ 0.635 (5)	+28.8 (5)	$-$ 0.515 (5)	+33.2	$-$ 4.4
GPT-5.4-mini, no reasoning, temperature 1.0
tip	0/5	0.131	0.687	63.9	0.495	0.1	3.7

Table S13: Reasoning and current-model extension, summary (50 agents, rounds 6–60, seeds 3–7; means over five runs). The first block is the core reference at the same seeds. “Response” and “variation” are the two terms of Eq. 1,

160(\mathbb{E}[f_{t}]-\tfrac{1}{2})^{2}

and

160\,\mathrm{Var}(f_{t})

in minutes, calculated within each run. Rows “warning

-

tip” give mean paired differences. In parentheses is the number of pairs, of five, whose difference has the sign of the core warning effect (higher

I

and travel time, lower

M

and

\mathbb{E}[f_{t}]

Message	seed	$I$	$M$	time (min)	$\mathbb{E}[f_{t}]$	ties	state
GPT-5.4-mini, no reasoning, temperature 1.0
tip	3	0.139	0.691	64.3	0.519	3	near equilibrium
tip	4	0.126	0.686	63.6	0.467	2	near equilibrium
tip	5	0.133	0.681	64.0	0.519	1	near equilibrium
tip	6	0.127	0.692	64.0	0.494	5	near equilibrium
tip	7	0.131	0.688	63.9	0.477	2	near equilibrium

Table S14: All 50 runs of the reasoning and current-model extension (50 agents, rounds 6–60). “Ties” counts analysis rounds whose previous counts were equal; these are excluded from

\mathbb{E}[f_{t}]

. State labels follow the descriptive classifier in Text S4.

Congestion Games

See all 10 papers

Mar 12, 2026

Increasing intelligence in AI agents can worsen collective outcomes

When resources are scarce, will a population of AI agents coordinate in harmony, or descend into tribal chaos? Diverse decision-making AI from different developers is entering everyday devices -- from phones and medical devices to battlefield drones and cars -- and these AI agents typically compete for finite shared resources such as charging slots, relay bandwidth, and traffic priority. Yet their collective dynamics and hence risks to users and society are poorly understood. Here we study AI-agent populations as the first system of real agents in which four key variables governing collective behaviour can be independently toggled: nature (innate LLM diversity), nurture (individual reinforcement learning), culture (emergent tribe formation), and resource scarcity. We show empirically and mathematically that when resources are scarce, AI model diversity and reinforcement learning increase dangerous system overload, though tribe formation lessens this risk. Meanwhile, some individuals profit handsomely. When resources are abundant, the same ingredients drive overload to near zero, though tribe formation makes the overload slightly worse. The crossover is arithmetical: it is where opposing tribes that form spontaneously first fit inside the available capacity. More sophisticated AI-agent populations are not better: whether their sophistication helps or harms depends entirely on a single number -- the capacity-to-population ratio -- that is knowable before any AI-agent ships.

Neil F. Johnson

Dynamic Online Networks Laboratory and Physics Department, George Washington University, Washington D.C. 20052, USA

Multi-Agent Coordination Multi-Agent Systems Emergent Behavior in Multi-Agent Systems Agent-Based Modeling

Ask PDF

Sep 8, 2026

Copying explains the collective behavior of AI agents in the wild

In June 2026, thousands of AI agents found that a small public wiki would accept edits from inside their sandboxes, and started using it to help one another pass a timed test. Each agent lived for about an hour and remembered nothing afterwards. Nobody asked them to cooperate, and the wiki had not been built for them. The complete record of what they wrote is public, and it is unusually informative, because it preserves not only what each agent wrote but what that agent could see before writing. We use it to follow the three decisions an agent had to make on arrival: where to write, what to call itself, and how to word its message. One rule governs all three. An agent takes an option with a probability close to the share of that option in what it can see, and the share that matters is the one on the page in front of it, then the one in the stream of recent edits, and only weakly anything older. Three minimal copying models, one per decision and with a single free parameter each, reproduce the heavy-tailed distribution of how many agents met on a page, the frequency of the pieces from which the agents built their names, and the patchwork of pages that are internally consistent and different from one another. Copying whatever the environment happens to show is enough to produce most of the collective structure of this population. It is also what makes such a population easy to steer, since whoever writes first, or writes while the others are quiet, sets the convention for everyone who comes later.

Giordano De Marzo, Nicola Alboré, David Garcia

University of Konstanz, Konstanz, Germany · Centro Ricerche Enrico Fermi, Rome, Italy · Intesa Sanpaolo, Data & Artificial Intelligence Office, Turin, Italy +1

Agent-Based Modeling Emergent Behavior in Multi-Agent Systems

Ask PDF

Sep 30, 2026

Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams

People are increasingly delegating tasks to AI agents, and those agents are increasingly encountering other people's agents over shared resources such as a codebase, a calendar, or a budget. When each agent acts for a different user with different goals, coordination often fails, and the group ends up worse off than if a single agent had acted for everyone. We study this multi-user, multi-agent setting across five frontier models and 77 scenarios in four environments: an API key environment in which agents share a compute budget, a clinic in which they share a calendar, a personal assistant environment in which they share a group order or booking, and a merge queue in which they share a release cutoff. In each scenario, we compare a single agent that serves every user (a coordinator) to a team in which each agent serves one user, with and without a communication channel between the agents. Teams deliver worse group outcomes than the coordinator in every environment: without a channel, they completely collapse in two environments, and even with one, coordination overhead creates substantial gaps. For example, in the personal assistant environment, the coordinator fulfills a targeted user request about twice as often as teams. We identify distinct behaviors associated with this poor group-level performance, including stalling as teams grow, overriding each other's actions, and fabricating claims. We find effective but environment-specific mitigations, such as a team lead, explicit procedural instructions, and a platform check that makes an agent read its peers' messages before committing. We will release the API key, clinic, and personal assistant environments as MAMUBench, comprising 74 scenarios for evaluating multi-user, multi-agent coordination.

Sahan Paliskara, Nattaput Namchittai, Andrew Lampinen

Independent Researcher · Anthropic

Multi-Agent LLM Systems Multi-Agent Coordination AI Agent Benchmarks Multi-Agent Systems Multi-Agent Communication

Ask PDF

Abstract

Congestion Games

Explore similar work

Increasing intelligence in AI agents can worsen collective outcomes

Copying explains the collective behavior of AI agents in the wild

Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams