Learning When Human Decisions Matter: Executor-Relative Consultation Value for LLM Agents
Organizations: AWS Agentic AI
Abstract
Large language models increasingly automate complex tasks, but human decision capacity does not scale with their capabilities. In many applications, humans remain essential for providing private context, authority, or judgments unavailable to the agent. We formulate human decision capacity allocation: given multiple consultation opportunities and a limited budget, at which decisions should an LLM agent consult a human? The central challenge is estimating consultation value, which is executor-dependent because the same human input may benefit different downstream agents differently. We introduce Consultation Value Distillation (CVD), which estimates consultation values offline using Monte Carlo executor rollouts and distills these values into lightweight predictors for budgeted allocation, requiring no rollout-based value estimation at deployment. Across five batch and three sequential settings with two frontier executors, CVD improves allocation through joint and continuation-aware value learning, while independent consultation values remain competitive in several settings. CVD also broadly outperforms direct prompting and Value of Information where applicable, remains robust with substantially less rollout supervision, and enables nearly lossless approximate allocation at low search cost.
Figures & tables
| Benchmark | Regime | Train | Test |
| Animals | Batch, Seq. | 245 | 75 |
| NaturalPlan Trip Planning | Batch | 521 | 79 |
| NaturalPlan Meeting Planning | Batch | 80 | 20 |
| Flight Recommendation | Batch | 113 | 30 |
| ThinkingBox-D (TB-D) | Batch, Seq. | 60 | 60 |
| ThinkingBox-H (TB-H) | Seq. | 114 | 30 |
| Sonnet 5 | GPT-6.1 Sol | ||||||||||
| Ours | Ours | ||||||||||
| Setting | Benchmark | Prompt | VoI | Local Value | CVD | Prompt | VoI | Local Value | CVD | ||
| Batch | Animals | 6 | 3 | 0.16 | 0.05 | 0.25 | 0.58 ∗ | 0.19 | 0.17 | 0.09 | 0.40 ∗ |
| Trip Planning | 3 | 2 | 0.09 | — | 0.79 | 0.83 | 0.31 | — | 0.75 | 0.80 | |
| Meeting Planning | 10 | 4 | 0.00 | — | 0.91 | 1.00 | 0.22 | — | 0.88 | 0.82 | |
| Flight | 5 | 3 | 0.24 | 0.18 | 0.00 | 0.33 | 0.29 | 0.14 | 0.09 | 0.00 | |
| Target: Sol | Target: Sonnet | ||||
| Benchmark | Own | Sonnet Sol | Own | Sol Sonnet | |
| Animals | 0.93 | 0.40 | 0.15 ∗ | 0.58 | 0.14 ∗ |
| Trip Planning | 0.22 | 0.80 | 0.85 | 0.83 | 0.84 |
| Meeting | 0.50 | 0.82 | 0.91 ∗ | 1.00 | 0.89 ∗ |
| Flight Rec. | 0.87 | 0.00 | 0.25 ∗ | 0.33 | 0.08 |
| TB-D | 0.13 | 0.97 | 0.99 | 1.00 | 1.00 |
Appendix figures & tables17 assets
Supplementary material from the paper’s appendix.
Appendix
| Benchmark | Predictor information |
| Animals | Question partition statistics and pairwise redundancy. |
| Trip | Stay/flight feasibility, hidden-date uncertainty, and features from a precomputed no-consultation draft. |
| Meeting | Availability-window and travel-feasibility statistics. |
| Flight | Query-flight variation, support-choice cues, and features from a precomputed no-consultation recommendation. |
| TB-D | Decision type, weight, option count, workflow position, and public-note cues. |
| TB-H | Workflow state, decision metadata, consultation history, and remaining budget. |
| Task | Executor | Adaptive fixed | Ratio | |
| Animals | Sonnet | 2 | [ , ] | |
| 3 | [ , ] | |||
| Sol | 2 | [ , ] | ||
| 3 | [ , ] | |||
| Flight | Sonnet | 2 | [ , ] | |
| 3 | [ , ] |
| Sets per task | Samples per label | ||||||||
| Benchmark | Feasible | Labeled | Sonnet 5 | GPT-6.1 Sol | Sonnet 5 | GPT-6.1 Sol | |||
| Animals | 0 | 1 | 1 | 8 | 8 | 0.88 | 0.99 | 0.84 | 1.00 |
| 1 | 6 | 6 | 8 | 8 | 0.91 | 0.97 | 0.93 | 0.97 | |
| 2 | 15 | 15 | 3 | 3 | 0.87 | 0.94 | 0.85 | 0.94 | |
| 3 | 20 | 20 | 2 | 2 | 0.85 | 0.92 | 0.81 | 0.92 | |
| 4 | 15 | 15 | 1 | 1 | 0.76 | 0.88 | 0.70 | 0.89 | |
| Batch | Sequential | ||
| Hyperparameter | CVD | Local Value | Both |
| Hidden width | 64 | 64 | 64 |
| Optimizer | AdamW | AdamW | AdamW |
| Learning rate | |||
| Weight decay | |||
| Epochs | 400 | 400 | 300 |
| Benchmark | Setting | Input dim. | Train. examples | CVD | Local Value |
| Animals | Batch | 15,680 | 9,281 | 897 | |
| Trip | Batch | 4,168 | 9,153 | 769 | |
| Meeting | Batch | 3,440 | 9,537 | 1,153 | |
| Flight | Batch | 2,938 | 10,049 | 1,665 | |
| TB-D | Batch | 360 | 9,921 | 1,537 | |
| Animals | Seq. | 28 | 251,034 | 6,081 | 6,081 |
| Executor | CVD ( ) | CVD ( ) | |
| Sonnet 5 | 72% | 0.33 | 0.39 |
| GPT-6.1 Sol | 67% | -0.28 | 0.31 |
| Benchmark | None | Random | Prompt | VoI | Local Value (ours) | CVD (ours) | Oracle | |||
| Sonnet 5 | Animals | 1 | 75 | 0.9800 | 0.9556 | 0.9537 0.0008 | 0.9565 0.0002 | 0.9459 0.0000 | 0.9476 0.0001 | 0.9414 |
| 2 | 0.9800 | 0.9198 | 0.9217 0.0012 | 0.9223 0.0004 | 0.9030 0.0005 | 0.9037 0.0006 | 0.8921 | |||
| 3 | 0.9800 | 0.8716 | 0.8655 0.0005 | 0.8697 0.0013 | 0.8622 0.0008 | 0.8499 0.0006 | 0.8343 | |||
| Trip | 1 | 79 | 0.027 | 0.135 | 0.129 0.002 | — | 0.174 0.001 | 0.173 0.001 | 0.176 | |
| 2 | 0.027 | 0.363 | 0.370 0.004 | — | 0.426 0.006 | 0.429 0.001 | 0.443 | |||
| Meeting | 1 | 20 | 2.170 | 2.284 | 2.361 0.026 | — | 2.595 0.003 | 2.619 0.012 | 2.634 |
| Benchmark | None | Random | Prompt | VoI | Local Value (ours) | CVD (ours) | Oracle | |||
| Sonnet 5 | Animals | 1 | 75 | 1.72 [ 2.13, 1.41] | 0.00 | 0.13 [0.02,0.26] | 0.07 [ 0.17,0.06] | 0.68 [0.51,0.82] | 0.56 [0.39,0.75] | 1.00 |
| 2 | 2.17 [ 2.64, 1.82] | 0.00 | 0.07 [ 0.17,0.03] | 0.09 [ 0.18,0.01] | 0.61 [0.44,0.76] | 0.58 [0.42,0.72] | 1.00 | |||
| 3 | 2.90 [ 3.43, 2.53] | 0.00 | 0.16 [0.01,0.32] | 0.05 [ 0.04,0.14] | 0.25 [0.10,0.40] | 0.58 [0.42,0.81] | 1.00 | |||
| Trip | 1 | 79 | 2.64 [ 4.48, 1.90] | 0.00 | 0.16 [ 0.56,0.27] | — | 0.95 [0.73,1.17] | 0.92 [0.69,1.15] | 1.00 | |
| 2 | 4.18 [ 5.88, 3.21] | 0.00 | 0.09 [ 0.24,0.41] | — | 0.79 [0.54,1.06] | 0.83 [0.59,1.08] | 1.00 | |||
| Meeting | 1 | 20 | 0.33 [ 0.47, 0.24] | 0.00 | 0.22 [ 0.09,0.54] | — | 0.89 [0.67,1.08] | 0.96 [0.73,1.18] | 1.00 |
| Benchmark | None | Random | Prompt | VoI | Local Value (ours) | CVD (ours) | Batch set | Best fixed set | Oracle | |||
| Sonnet 5 | Animals | 2 | 75 | 0.9800 | 0.9198 | 0.9185 0.0003 | 0.9191 0.0023 | 0.8928 0.0001 | 0.8958 0.0020 | 0.9037 | 0.8921 | 0.8760 |
| 3 | 0.9800 | 0.8716 | 0.8376 0.0014 | 0.8626 0.0032 | 0.8286 0.0005 | 0.8307 0.0029 | 0.8499 | 0.8337 | 0.7698 | |||
| TB-D | 1 | 60 | 0.716 | — | 2.571 0.049 | — | 2.923 0.026 | 3.101 0.004 | — | 3.150 | 3.250 | |
| 2 | 0.716 | — | 3.643 0.014 | — | 4.099 0.029 | 4.412 0.010 | — | 4.370 | 4.607 | |||
| TB-H | 2 | 30 | 1.483 | — | 1.681 0.075 | — | 3.467 0.041 | 4.169 0.020 | — | 4.075 | 4.742 | |
| GPT-6.1 Sol | Animals | 2 | 75 | 0.9790 | 0.8859 | 0.8793 0.0040 | 0.8808 0.0020 | 0.8576 0.0007 | 0.8600 0.0016 | 0.8736 | 0.8433 | 0.8255 |
| Benchmark | None | Random | Prompt | VoI | Local Value (ours) | CVD (ours) | Batch set | Best fixed set | Oracle | |||
| Sonnet 5 | Animals | 2 | 75 | 1.37 [ 1.55, 1.22] | 0.00 | 0.03 [ 0.01,0.07] | 0.02 [ 0.05,0.08] | 0.62 [0.52,0.71] | 0.55 [0.45,0.64] | 0.37 [0.25,0.47] | 0.63 [0.56,0.70] | 1.00 |
| 3 | 1.06 [ 1.14, 0.99] | 0.00 | 0.33 [0.29,0.38] | 0.09 [0.05,0.13] | 0.42 [0.36,0.48] | 0.40 [0.35,0.46] | 0.21 [0.16,0.28] | 0.37 [0.33,0.42] | 1.00 | |||
| TB-D | 1 | 60 | 0.00 | — | 0.73 [0.66,0.80] | — | 0.87 [0.81,0.92] | 0.94 [0.90,0.97] | — | 0.96 [0.93,0.99] | 1.00 | |
| 2 | 0.00 | — | 0.75 [0.70,0.80] | — | 0.87 [0.83,0.91] | 0.95 [0.92,0.97] | — | 0.94 [0.92,0.96] | 1.00 | |||
| TB-H | 2 | 30 | 0.00 | — | 0.51 [0.42,0.59] | — | 0.80 [0.73,0.86] | 0.91 [0.88,0.94] | — | 0.89 [0.85,0.93] | 1.00 | |
| GPT-6.1 Sol | Animals | 2 | 75 | 1.54 [ 1.70, 1.40] | 0.00 | 0.11 [0.04,0.19] | 0.08 [0.03,0.14] | 0.47 [0.37,0.56] | 0.43 [0.35,0.51] | 0.20 [0.11,0.29] | 0.71 [0.65,0.77] | 1.00 |
| Sonnet 5 | |||||||||
| Explicit decomposition | Local-value auxiliaries | ||||||||
| Bench. | R | PR | S2 | Joint add. | Joint v+M | D+v | Aux. loss | Wider | |
| Animals | 1 | +0.09 [ 0.04, +0.27] | +0.12 [ 0.00, +0.30] | +0.12 [ 0.00, +0.30] | +0.05 [ 0.03, +0.16] | +0.05 [ 0.03, +0.16] | +0.00 [+0.00, +0.00] | — | — |
| Animals | 2 | +0.02 [ 0.04, +0.09] | 0.05 ∗ [ 0.09, 0.01] | +0.03 [ 0.11, +0.16] | +0.03 [ 0.05, +0.13] | 0.08 [ 0.21, +0.04] | 0.07 ∗ [ 0.15, 0.01] | — | — |
| Animals | 3 | +0.08 [ 0.03, +0.19] | +0.07 [ 0.01, +0.17] | +0.04 [ 0.09, +0.18] | 0.14 ∗ [ 0.26, 0.02] | +0.05 [ 0.05, +0.16] | +0.04 [ 0.02, +0.10] | — | — |
| Trip | 1 | +0.07 [ 0.12, +0.30] | +0.02 [ 0.20, +0.25] | +0.02 [ 0.20, +0.25] | +0.04 [ 0.14, +0.25] | 0.04 [ 0.27, +0.20] | +0.07 ∗ [+0.00, +0.21] | — | — |
| Sonnet 5 | ||||||
| Within-episode / relative supervision | Simpler allocators | |||||
| Bench. | Within-ep. | Contrast-D | Hybrid | Local Value | Single model | |
| Animals | 1 | +0.10 [ 0.02, +0.28] | +0.04 [ 0.04, +0.15] | +0.06 [ 0.03, +0.18] | +0.12 [ 0.00, +0.30] | +0.00 [ 0.05, +0.06] |
| Animals | 2 | +0.01 [ 0.12, +0.14] | +0.11 ∗ [+0.01, +0.23] | 0.04 [ 0.12, +0.04] | +0.03 [ 0.13, +0.17] | 0.06 ∗ [ 0.11, 0.01] |
| Animals | 3 | 0.37 ∗ [ 0.61, 0.19] | +0.17 [ 0.00, +0.33] | +0.06 [ 0.07, +0.18] | 0.33 ∗ [ 0.57, 0.15] | 0.03 [ 0.08, +0.01] |
| Trip | 1 | +0.08 [ 0.12, +0.36] | +0.06 [ 0.15, +0.34] | 0.05 [ 0.22, +0.07] | +0.02 [ 0.20, +0.25] | 0.11 ∗ [ 0.19, 0.03] |
| Bench. | (1) Set vs. singleton labels | (2) Remove additivity | (3) Raw vs. standardized inputs | (4) All sets, uniform weights | |
| Animals | 1 | +0.01 [ 0.05, +0.07] | +0.01 [ 0.05, +0.08] | 0.03 [ 0.10, +0.04] | +0.04 [ 0.16, +0.25] |
| Animals | 2 | +0.14 [ 0.08, +0.35] | +0.06 [ 0.02, +0.16] | 0.05 [ 0.14, +0.04] | 0.06 [ 0.15, +0.01] |
| Animals | 3 | +0.13 [ 0.04, +0.34] | +0.25 ∗ [+0.06, +0.45] | 0.22 ∗ [ 0.39, 0.05] | +0.10 ∗ [+0.02, +0.19] |
| Trip | 1 | +0.10 [ 0.08, +0.36] | 0.08 [ 0.34, +0.11] | +0.01 [ 0.20, +0.25] | 0.03 [ 0.30, +0.20] |
| Trip | 2 | 0.01 [ 0.28, +0.20] | 0.15 [ 0.38, +0.05] | +0.21 [ 0.03, +0.48] | 0.04 [ 0.23, +0.12] |
| Flight | 1 | +0.12 [ 0.22, +0.45] | +0.02 [ 0.32, +0.35] | +0.22 [ 0.23, +0.66] | +0.04 [ 0.13, +0.30] |
| Bench. | Condition | by | Ensemble result | vs. default [95% CI] | Retention | |
| Animals | 1 | R1 | (1, 1, 1, 1) | 0.549 | -0.005 [-0.056,+0.042] | 0.99 |
| Animals | 1 | R2 | (2, 2, 2, 1) | 0.554 | +0.000 [+0.000,+0.000] | 1.00 |
| Animals | 1 | R4 | (4, 3, 2, 1) | 0.650 | +0.096 [+0.000,+0.289] | 1.17 |
| Animals | 1 | default | (8, 3, 2, 1) | 0.554 | +0.000 [+0.000,+0.000] | 1.00 |
| Animals | 2 | R1 | (1, 1, 1, 1) | 0.606 | +0.022 [-0.026,+0.093] | 1.04 |
| Animals | 2 | R2 | (2, 2, 2, 1) | 0.558 | -0.026 [-0.064,+0.000] | 0.96 |
| Bench. | Condition | Labeled sets by | % of sets | Ensemble result | SD (seeds) | vs. default [95% CI] | Retention | |
| Animals | 1 | M7 | (1,6,0,0,0,0,0) | 10.9% | 0.709 | 0.000 | +0.155 [-0.019,+0.421] | 1.28 |
| Animals | 1 | M16 | (1,6,3,3,2,1,0) | 25.0% | 0.615 | 0.057 | +0.061 [-0.025,+0.212] | 1.11 |
| Animals | 1 | M32 | (1,6,6,9,6,3,1) | 50.0% | 0.628 | 0.043 | +0.074 [-0.006,+0.224] | 1.13 |
| Animals | 1 | D42 | (1,6,15,20,0,0,0) | 65.6% | 0.549 | 0.000 | -0.005 [-0.056,+0.042] | 0.99 |
| Animals | 1 | default | all | 100.0% | 0.554 | 0.000 | +0.000 [+0.000,+0.000] | 1.00 |
| Animals | 2 | M7 | (1,6,0,0,0,0,0) | 10.9% | 0.510 | 0.000 | -0.074 [-0.256,+0.099] | 0.87 |
| Executor | Bench. | Exact | Greedy | Beam(3) | Greedy+swap | |
| Sonnet 5 | Animals | 1 | 1.00 / 0.55 / 7 | 1.00 / 0.55 / 7 | 1.00 / 0.55 / 7 | 1.00 / 0.55 / 7 |
| Sonnet 5 | Animals | 2 | 1.00 / 0.58 / 22 | 1.00 / 0.58 / 12 | 1.00 / 0.58 / 19 | 1.00 / 0.58 / 16 |
| Sonnet 5 | Animals | 3 | 1.00 / 0.60 / 42 | 0.89 / 0.54 / 16 | 1.00 / 0.60 / 28 | 1.00 / 0.60 / 24 |
| Sonnet 5 | Trip | 1 | 1.00 / 0.89 / 4 | 1.00 / 0.89 / 4 | 1.00 / 0.89 / 4 | 1.00 / 0.89 / 4 |
| Sonnet 5 | Trip | 2 | 1.00 / 0.81 / 7 | 0.87 / 0.77 / 6 | 1.00 / 0.81 / 7 | 1.00 / 0.81 / 7 |
| Sonnet 5 | Flight | 1 | 1.00 / 0.39 / 6 | 1.00 / 0.39 / 6 | 1.00 / 0.39 / 6 | 1.00 / 0.39 / 6 |
| Value correlation | Target: GPT-6.1 Sol | Target: Sonnet 5 | ||||||||
| Benchmark | raw | disatt. | Own | Sonnet | Prompt | Own | Sol | Prompt | ||
| Mixed 20Q–Animals | 3 | 0.43 | 0.45 | 0.93 | 0.40 | 0.15 ∗ | 0.19 | 0.58 | 0.14 ∗ | 0.16 |
| NaturalPlan Trip Planning | 2 | 0.93 | 0.97 | 0.22 | 0.80 | 0.85 | 0.31 | 0.83 | 0.84 | 0.09 |
| NaturalPlan Meeting Planning | 4 | 0.77 | 0.95 | 0.50 | 0.82 | 0.91 ∗ | 0.22 | 1.00 | 0.89 ∗ | 0.00 |
| Flight Recommendation | 3 | 0.43 | 0.46 | 0.87 | 0.00 | 0.25 ∗ | 0.29 | 0.33 | 0.08 | 0.24 |
| ThinkingBox-D | 1 | 0.97 | 1.00 | 0.13 | 0.97 | 0.99 | 0.54 | 1.00 | 1.00 | 0.58 |
Human-AI Decision Making
Human-AI decision making studies how people and AI systems share, support, or delegate decisions across tasks and institutions.
See all 199 papers