The Novelty Bottleneck: A Framework for Understanding Human Effort Scaling in AI-Assisted Work

Jacky Liang

cs.AIMar 28, 2026

The Novelty Bottleneck: A Framework for Understanding Human Effort Scaling in AI-Assisted Work

Authors: Jacky Liang

Abstract

We propose a stylized model of human-AI collaboration that isolates a mechanism we call the novelty bottleneck: the fraction of a task requiring human judgment creates an irreducible serial component analogous to Amdahl's Law in parallel computing. The model assumes that tasks decompose into atomic decisions, a fraction $ν$ of which are "novel" (not covered by the agent's prior), and that specification, verification, and error correction each scale with task size. From these assumptions, we derive several non-obvious consequences: (1) there is no smooth sublinear regime for human effort it transitions sharply from $O(E)$ to $O(1)$ with no intermediate scaling class; (2) better agents improve the coefficient on human effort but not the exponent; (3) for organizations of n humans with AI agents, optimal team size decreases with agent capability; (4) wall-clock time achieves $O(\sqrt{E})$ through team parallelism but total human effort remains $O(E)$ ; and (5) the resulting AI safety profile is asymmetric -- AI is bottlenecked on frontier research but unbottlenecked on exploiting existing knowledge. We show these predictions are consistent with empirical observations from AI coding benchmarks, scientific productivity data, and practitioner reports. Our contribution is not a proof that human effort must scale linearly, but a framework that identifies the novelty fraction as the key parameter governing AI-assisted productivity, and derives consequences that clarify -- rather than refute -- prevalent narratives about intelligence explosions and the "country of geniuses in a data center."

Figures & tables

Configuration	$\nu$	$p_{\text{routine}}$	$p_{\text{novel}}$	$r$	$c_{v}$
Low Novelty	0.1	0.95	0.3	0.0	0.05
Medium Novelty	0.3	0.95	0.3	0.0	0.05
High Novelty	0.8	0.95	0.3	0.0	0.05
High Capability	0.3	0.99	0.7	0.8	0.02

Table 1: Agent configuration parameters for the scaling experiment.

Configuration	$\alpha$	$H/E$ at $E{=}5000$
Low Novelty ( $\nu=0.1$ )	0.999	0.377
Medium Novelty ( $\nu=0.3$ )	1.004	0.841
High Novelty ( $\nu=0.8$ )	1.002	1.990
High Capability + Self-Correction	1.004	0.360

Table 2: Fitted scaling exponents

\alpha

where

H\sim E^{\alpha}

Mutual Information	$H/E$
0.0	1.050
0.3	0.751
0.6	0.450
0.9	0.151
0.99	0.060

Table 3: Effect of mutual information on

H/E

ratio at

E=5{,}000

Figure 1: Simulation results across configurations. Top left :

H

vs.

E

is linear for all configurations. Top center :

H/E

ratio converges to a constant. Top right : Agent trajectory divergence grows as

O(\sqrt{E})

, confirming the random walk model. Bottom left : Higher mutual information reduces the linear coefficient but not the scaling class. Bottom center : Even 1% novelty makes

H/E

converge to a positive constant. Bottom right : Stacked decomposition of

H

for the high-novelty case.

Figure 2: Left : Log-log plot confirming linear scaling (

\alpha\approx 1

) across all configurations. Reference lines for

O(E)

O(\sqrt{E})

, and

O(\log E)

shown. Right : Fitted scaling exponents; all configurations cluster tightly around

\alpha=1.0

Figure 3: March of nines analysis. Left : End-to-end reliability decays exponentially with task length. Center : Required per-step nines grow with

E

(log scale). Right : Human checkpoints required scale linearly with

E

Figure 4: Verifiability frontier. Left :

H/E

ratio across the novelty-verifiability plane. Center : Verifiability reduces the slope but does not eliminate linear scaling. Right : Concrete task types placed in the novelty-verifiability space.

Configuration	$c$	$\beta$	$\gamma$	$\tau$
No AI	1.0	0.5	0.0	1.0
Basic AI ( $2\times$ )	0.5	0.5	0.1	2.0
Strong AI ( $5\times$ )	0.2	0.5	0.1	5.0
Frontier AI ( $10\times$ )	0.1	0.5	0.1	10.0

Table 4: Organizational scaling configuration parameters.

Configuration	Throughput $\tau$	$n^{*}$	$T^{*}$
No AI	$1\times$	100.0	100.0
Basic AI	$2\times$	59.8	83.7
Strong AI	$5\times$	31.6	63.2
Frontier AI	$10\times$	18.3	54.8

Table 5: Optimal team size and minimum wall-clock time for

E=5{,}000

Figure 5: Organizational scaling results. Top left : Wall-clock time vs. team size shows U-shaped curves with optimal

n^{*}

(dots) that shifts left as agents improve. Top center : Total human effort grows superlinearly past

n^{*}

. Top right : Optimal team size grows as

O(\sqrt{E})

but shrinks with agent capability. Bottom left : Better agents systematically reduce optimal team size. Bottom center : Minimum achievable wall-clock time scales as

O(\sqrt{E})

for all configurations. Bottom right : Work efficiency (fraction of effort spent on useful work vs. coordination) decays rapidly with team size, faster for more capable agents.

Explore similar work

May 12, 2026

Human-AI Productivity Paradoxes: Modeling the Interplay of Skill, Effort, and AI Assistance

Generative Artificial Intelligence (AI) tools are rapidly adopted in the workplace and in education, yet the empirical evidence on AI's impact remains mixed. We propose a model of human-AI interaction to better understand and analyze several mechanisms by which AI affects productivity. In our setup, human agents with varying skill levels exert utility-maximizing effort to produce certain task outcomes with AI assistance. We find that incorporating either endogeneity in skill development or in AI unreliability can induce a productivity paradox: increased levels of AI assistance may degrade productivity, leading to potentially significant shortfalls. Moreover, we examine the long-term distributional effect of AI on skill, and demonstrate that skill polarization can emerge in steady state when accounting for heterogeneity in AI literacy -- the agent's capability to identify and adapt to inaccurate AI outputs. Our results elucidate several mechanisms that may explain the emergence of human-AI productivity paradoxes and skill polarization, and identify simple measures that characterize when they arise.

Ali Aouad, Thodoris Lykouris, Huiying Zhong

Massachusetts Institute of Technology

AI Literacy Human-AI Interaction

Ask PDF

Apr 27, 2026

Leverage Laws: A Per-Task Framework for Human-Agent Collaboration

We propose a per-task leverage ratio for human-agent collaboration: human work displaced by an agent, divided by the human time required to specify the task, resolve mid-run interrupts, and review the result. The denominator decomposes into three channels through which a conserved per-task information requirement must flow, each with its own time-cost scalar. We show that information density itself is directional and bounded by separate ceilings on human-to-agent and agent-to-human flow, and that the asymptotic behavior of leverage decomposes into two scaling axes (capability and memory) with a non-zero floor on the planning term set by irreducible task novelty bounded by human throughput. We extend this per-task analysis to a windowed leverage measure that accommodates recurring tasks, spawned subtasks, and amortized system-design investment. The per-task ceiling does not bind the windowed measure, though both remain bounded:

L_{\text{task}}

by per-task novelty,

L_{\text{window}}

by the stock of accumulated planning investment that pays out within the window. The framework operationalizes aspects of earlier qualitative work on supervisory control (Sheridan, 1992), common ground (Clark & Brennan, 1991), and mixed-initiative interaction (Horvitz, 1999) within a single normative ratio, and produces a list of testable empirical questions that we leave as open problems.

Stan Loosmore

Human-AI Interaction Human-AI Collaboration AI and Labor

Ask PDF

Jun 16, 2026

Searching for Synergy in Shared Workspace Human-AI Collaboration

Automated AI agents are increasingly capable, yet many scientific and professional tasks require human judgment and contextual expertise. We use simulated shared-workspace human-AI teams as a controlled testbed for studying how collaboration structure shapes team behavior. Using the Collaborative Gym environment with tasks from DiscoveryBench, we vary team compositions and collaboration structures across 1,482 sessions. We find that adding additional collaborators can lower performance when coordination structure is absent. We then evaluate collaboration scaffolding that combines shared group memory with simulated human-in-the-loop (HITL) gates, where selected actions require approval from a designated simulated participant. This scaffolding improves performance, most clearly in three-person teams, with clearer responsibility signals and stronger routing of expertise to team actions. Overall, our results suggest that coordination structure is central to whether available capability improves team outcomes.

Nachiket Kotalwar, Rohini Das, Carolyn Rose

1Carnegie Mellon University.

Multi-Agent Coordination Human-in-the-Loop AI Human-AI Collaboration

Ask PDF