arXiv subjects

Andrew Koh

Publications and source records attributed to Andrew Koh.

17 recordsLinked to original sources

arXiv2026

Mechanism Design for Alignment and Control

We develop a framework for mechanism design with AI agents whose alignment (preferences) and capabilities (feasible actions and information) are unknown. We want such agents to act on our behalf so mechanisms must incentivize both honesty and obedience. A one-sided imitation structure---capabilities can be concealed but not counterfeited---yields a revelation principle, a characterization of implementable policies via nested cyclical monotonicity, and conditions under which eliciting higher-order beliefs can discipline multiple agents. We apply our framework to stylized examples of (i) sandbagging in which a more capable agent pretends to be less capable; (ii) an alignment--interpretability trade-off, where the two are substitutes in the instrument but complements in value; (iii) discipline via peer scoring; (iv) coupling rewards to induce competition among multiple agents; and (v) scalable oversight and reward shaping.

econ.TH↗

arXiv2026

Racing to Ruin

We study R&D competition in the shadow of disaster: advancing the technology frontier raises the risk of permanently ending all firms' payoffs. Under perfect monitoring and common knowledge of rationality, the equilibrium frontier is bounded below by the optimal stopping time of a monopolist, and above by that of a representative firm that persistently but mistakenly believes its rival is about to stop. We then analyze how the frontier is shaped by transparency (speed of monitoring) and trust (belief in the rationality of rival firms).

econ.TH↗

arXiv2026

Data-Driven Automation

We build a dynamic model of data-driven automation in which data (i) is heterogeneous and task-specific; (ii) accumulates endogenously as a byproduct of economic activity; and (iii) exhibits spillovers such that data generated by one task can augment the productivity of another. Along the transition path of automation, data plays a dual role in simultaneously augmenting the productivity of already-automated tasks and expanding the automation frontier. We derive tight conditions for the economy to be partially versus fully automated in the long-run. In the latter case, automation exhibits rich short-run dynamics that depend on the pattern of data spillovers but is always slow in the long-run: the share of tasks produced by labor decays asymptotically as a power law in time. We show that the economy is generically inefficient and analyze how a planner optimally tilts the direction of data accumulation. With endogenous capital accumulation, data-driven automation generates explosive growth but stagnant long-run wages.

econ.TH↗

arXiv2026

Technology Speed Limits

We study optimal technology regulation when private learning occurs both through doing (scaling up the technology) and through waiting (as time passes). We show that an adaptive speed limit -- a cap on the rate at which the technology can increase per-unit time -- delivers optimal worst-case guarantees over all learning processes and/or preferences, and is the only time-consistent mechanism that does so.

econ.TH↗

arXiv2025

Full-Stack Alignment: Co-Aligning AI and Institutions with Thick Models of Value

Beneficial societal outcomes cannot be guaranteed by aligning individual AI systems with the intentions of their operators or users. Even an AI system that is perfectly aligned to the intentions of its operating organization can lead to bad outcomes if the goals of that organization are misaligned with those of other institutions and individuals. For this reason, we need full-stack alignment, the concurrent alignment of AI systems and the institutions that shape them with what people value. This can be done without imposing a particular vision of individual or collective flourishing. We argue that current approaches for representing values, such as utility functions, preference orderings, or unstructured text, struggle to address these and other issues effectively. They struggle to distinguish values from other signals, to support principled normative reasoning, and to model collective goods. We propose thick models of value will be needed. These structure the way values and norms are represented, enabling systems to distinguish enduring values from fleeting preferences, to model the social embedding of individual choices, and to reason normatively, applying values in new domains. We demonstrate this approach in five areas: AI value stewardship, normatively competent agents, win-win negotiation systems, meaning-preserving economic mechanisms, and democratic regulatory institutions.

cs.LG↗

arXiv2025

An Economy of AI Agents

In the coming decade, artificially intelligent agents with the ability to plan and execute complex tasks over long time horizons with little direct oversight from humans may be deployed across the economy. This chapter surveys recent developments and highlights open questions for economists around how AI agents might interact with humans and with each other, shape markets and organizations, and what institutions might be required for well-functioning markets.

econ.GN↗

arXiv2024

Informational Puts

We analyze how dynamic information should be provided to uniquely implement the largest equilibrium in binary-action coordination games. The designer offers an informational put: she stays silent if players choose her preferred action, but injects asymmetric and inconclusive public information if they lose faith. There is (i) no multiplicity gap: the largest (partially) implementable equilibrium can be implemented uniquely; and (ii) no commitment gap: the policy is sequentially optimal. Our results have sharp implications for the design of policy in coordination environments.

econ.TH↗

arXiv2024

Flexible Demand Manipulation

We develop a simple framework to analyze how targeted persuasive advertising shapes market power and welfare. A designer flexibly manipulates the demand curve by influencing individual valuations at a cost. A monopolist prices against this manipulated demand curve. We fully characterize the form of optimal advertising plans under ex-ante and ex-post welfare measures. Flexibility per se is powerful, and can substantially harm or benefit consumers vis-a-vis uniform advertising. We discuss implications for regulation, intermediation, and the joint design of manipulation and information.

econ.TH↗

arXiv2024

Inertial Coordination Games

We analyze inertial coordination games: dynamic coordination games with an endogenously changing state that depends on (i) a persistent fundamental players privately learn about over time; and (ii) past play. The speed of learning determines long-run equilibrium dynamics: the risk-dominant action is played in the limit if and only if learning is slow such that posterior precisions grow sub-quadratically. This generalizes results from static global games and endows them with a learning foundation. Conversely, when learning is fast such that posterior precisions grow super-quadratically, shocks can propagate and generate self-fulfilling spirals.

econ.TH↗

arXiv2024

Robust Technology Regulation

We analyze how uncertain technologies should be robustly regulated and how regulation should evolve with new information. An adaptive sandbox comprising a zero marginal tax up to an evolving quantity limit is (i) robust: it delivers optimal payoff guarantees when the agent's learning process and/or preferences are chosen adversarially; (ii) dominant: it outperforms other robust and regular mechanisms across all agent learning processes and preferences; (iii) time-consistent: it is the only robust mechanism that can be implemented without commitment. Robustness is important: absent robust regulation, worst-case payoffs can be arbitrarily poor and are induced by weak but growing optimism that encourages excessive risk-taking. Our results offer optimality foundations for existing policy and speak directly to current debates around managing emerging technologies.

econ.TH↗

arXiv2024

Persuasion and Optimal Stopping

We develop a duality-based first-order approach to dynamic persuasion in optimal stopping problems with general action-, state-, and time-dependent preferences. A direct-communication reduction recasts the design problem as a semi-static program over joint distributions of stopping beliefs and times; strong duality and a near-necessary first-order condition then reduce it to a one-dimensional differential equation in a multiplier that prices the agent's continuation incentive, characterizing the optimum as a \emph{concavification} of a multiplier-augmented payoff. We demonstrate the method in three applications: dynamic binary persuasion, where optimal policies combine \emph{suspense generation} with \emph{action-targeting}; a structural result by which the principal's time-risk preferences alone determine whether suspense is optimal; and dynamic linear persuasion, where the optimum is \emph{dynamic tail-censorship}.

econ.TH↗

arXiv2023

Leveraging Audio-Tagging Assisted Sound Event Detection using Weakified Strong Labels and Frequency Dynamic Convolutions

Jointly learning from a small labeled set and a larger unlabeled set is an active research topic under semi-supervised learning (SSL). In this paper, we propose a novel SSL method based on a two-stage framework for leveraging a large unlabeled in-domain set. Stage-1 of our proposed framework focuses on audio-tagging (AT), which assists the sound event detection (SED) system in Stage-2. The AT system is trained utilizing a strongly labeled set converted into weak predictions referred to as weakified set, a weakly labeled set, and an unlabeled set. This AT system then infers on the unlabeled set to generate reliable pseudo-weak labels, which are used with the strongly and weakly labeled set to train a frequency dynamic convolutional recurrent neural network-based SED system at Stage-2 in a supervised manner. Our system outperforms the baseline by 45.5% in terms of polyphonic sound detection score on the DESED real validation set.

eess.AS↗

arXiv2022

Attention Capture

We develop a unified analysis of how information captures attention. A decision maker (DM) faces a dynamic information structure and decides when to stop paying attention. We characterize the convex$\unicode{x2013}$order frontier and extreme points of feasible stopping times, as well as dynamic information structures which implement them. This delivers the form of optimal attentional capture as a function of the designer and DM's relative time preferences. Intertemporal commitment is unnecessary: sequentially optimal information structures always exist by inducing stochastic interim beliefs. We further analyze optimal attention capture under non instrumental value of information. Our results speak directly to the attention economy.

econ.TH↗

arXiv2022

Language-Based Audio Retrieval with Converging Tied Layers and Contrastive Loss

In this paper, we tackle the new Language-Based Audio Retrieval task proposed in DCASE 2022. Firstly, we introduce a simple, scalable architecture which ties both the audio and text encoder together. Secondly, we show that using this architecture along with contrastive loss allows the model to significantly beat the performance of the baseline model. Finally, in addition to having an extremely low training memory requirement, we are able to use pretrained models as it is without needing to finetune them. We test our methods and show that using a combination of our methods beats the baseline scores significantly.

cs.SD↗

arXiv2022

Automated Audio Captioning with Epochal Difficult Captions for Curriculum Learning

In this paper, we propose an algorithm, Epochal Difficult Captions, to supplement the training of any model for the Automated Audio Captioning task. Epochal Difficult Captions is an elegant evolution to the keyword estimation task that previous work have used to train the encoder of the AAC model. Epochal Difficult Captions modifies the target captions based on a curriculum and a difficulty level determined as a function of current epoch. Epochal Difficult Captions can be used with any model architecture and is a lightweight function that does not increase training time. We test our results on three systems and show that using Epochal Difficult Captions consistently improves performance

cs.CL↗

arXiv2021

Automated Audio Captioning using Transfer Learning and Reconstruction Latent Space Similarity Regularization

In this paper, we examine the use of Transfer Learning using Pretrained Audio Neural Networks (PANNs), and propose an architecture that is able to better leverage the acoustic features provided by PANNs for the Automated Audio Captioning Task. We also introduce a novel self-supervised objective, Reconstruction Latent Space Similarity Regularization (RLSSR). The RLSSR module supplements the training of the model by minimizing the similarity between the encoder and decoder embedding. The combination of both methods allows us to surpass state of the art results by a significant margin on the Clotho dataset across several metrics and benchmarks.

cs.CL↗

arXiv2020

Adapting BERT for Word Sense Disambiguation with Gloss Selection Objective and Example Sentences

Domain adaptation or transfer learning using pre-trained language models such as BERT has proven to be an effective approach for many natural language processing tasks. In this work, we propose to formulate word sense disambiguation as a relevance ranking task, and fine-tune BERT on sequence-pair ranking task to select the most probable sense definition given a context sentence and a list of candidate sense definitions. We also introduce a data augmentation technique for WSD using existing example sentences from WordNet. Using the proposed training objective and data augmentation technique, our models are able to achieve state-of-the-art results on the English all-words benchmark datasets.

cs.CL↗