TDTinglong Dai

Tinglong Dai

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Tinglong Dai has a total of 240 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Democratizing Optimization with Generative AI

    by David Simchi-Levi, Tinglong Dai, Ishai Menache, Michelle Xiao Wu

    Recent breakthroughs in generative artificial intelligence (GenAI) have captured the public imagination, whereas mathematical optimization remains little recognized beyond specialists. In this article, we argue that GenAI can bridge the persistent gap between optimization's technical capabilities and its limited real-world uptake. We present the 4I framework (Insight, Interpretability, Interactivity, Improvisation) as a set of design principles for combining GenAI with mathematical optimization. Insight establishes a trusted, up-to-date view of current operations; Interpretability explains model logic and trade-offs; Interactivity enables conversational what-if analysis; Improvisation supports model updates and event-driven re-optimization. By making optimization tools more intuitive, explainable, interactive, and adaptable, we envision a future in which frontline decision-makers can base their choices on rigorous, iterative analysis. We discuss how GenAI complements, rather than replaces, optimization: GenAI lowers barriers to modeling and interpretation; mathematical optimization reliably enforces the business goals, rules, and hard constraints encoded in the model. We also address emerging concerns, from hallucinations to the risk of over-reliance, and outline research directions for integrating GenAI and optimization responsibly. The GenAI boom gives the optimization community an opportunity to expand its impact by making optimization more accessible and trustworthy to a wider audience.


  • Open Access

    Wrestling with Environmental and Social Risks in Large-scale Infrastructure Projects

    by Sun Kim, Sriram Narayanan, Tinglong Dai

    The World Bank requires the infrastructure projects it finances to identify their environmental and social (E&S) risks before approval and to plan how to address them. We compare the E&S risks that 271 projects in 56 countries documented at appraisal and at completion. An E&S dictionary and a large language model identify the sentences that describe project-induced E&S risks in the 542 reports; the resulting measure agrees with human coding and the Bank's E&S categories. By completion, documented E&S risk portfolios concentrate, often on Land acquisition and resettlement. Among the 207 projects with enough documented risk at both stages, a one-standard-deviation increase in concentration is associated with 4.0 more months of delay and a cost change 11.1 percentage points larger, although the cost association is more sensitive to specification. Of the projects that documented land risks at completion, 93 percent had already documented them at appraisal. When the completion reports explain why a land issue was hard to settle, they mention project-affected people more often, and contractors less often, than when they explain other difficulties. The pattern is consistent with limited agency, in which a project team can foresee a risk and still depend on others to settle it. Crossing agency with the stakeholder's relationship to the project gives four ways to settle an issue: direct, mitigate, coordinate, and wrestle. For each identified risk, a project team should establish whose decision will settle it and, before committing to a schedule, plan for the decisions it cannot take itself.

  • The Sound We Haven’t Heard Before: A Commentary on “Fighting Fire with Fire: Infusing AI into Peer Review to Sustain Quality Scholarship”

    by Tinglong Dai

    This commentary responds to “Fighting Fire with Fire: Infusing AI into Peer Review to Sustain Quality Scholarship” [Bhargava HK, Bana S, Zhang Z, Brandimarte L, Choudary V, Li F, Loupos P, Zantedeschi D (2026) Fighting fire with fire: Infusing AI into peer review to sustain quality scholarship. Management Sci. Forthcoming.]. It shares the paper’s core position, that journals should govern artificial intelligence openly rather than leave it to the shadows, but it questions the premise that rising submission volume poses an existential threat. The safeguard is human judgment. Our shared fatigue with derivative, repetitive work is a sensor more discerning than any AI detector, because a model trained on what we have already rewarded cannot produce the novelty that our community prizes. The threat to a journal like Management Science is thus more contained than the original alarm implies, provided AI remains a non-final input and humans retain ownership of novelty and contribution. One test should govern any AI workflow: does it ask more of human judgment, or less? This commentary was accepted by Christoph Loch.


  • Open Access

    FDA drug quality assurance inspections: frequency, outcomes, and time to reinspection, 2008-2025

    by Yunxiang Sun, Jeromie Ballreich, Joy Acha, Riya Modi, Mohammad Ali Alamdar Yazdi, Maqbool Dada, Tinglong Dai, Gerard Anderson, Mariana P Socal

    Background: The U.S. Food and Drug Administration (FDA) inspections are central to oversight of the pharmaceutical manufacturing base, but long-term inspection and reinspection patterns remain incompletely characterized. Methods: We analyzed 25 629 FDA Drug Quality Assurance inspections conducted from October 2008 through December 2025. We described annual, regional, and outcome patterns. Linking inspections to facilities, we examined time to reinspection according to the outcome of preceding inspections. Results: Annual Drug Quality Assurance inspections declined from 1594 in 2019 to 461 in 2020 and increased to 1218 in 2025. The United States remained the largest single region of inspection activity, but its contribution declined over time. The proportion of favorable inspection outcomes (no action indicated—NAI) decreased from 40.0% before the pandemic to 19.3% during and 23.7% after. Unfavorable outcomes (voluntary action indicated—VAI—and official action indicated—OAI) increased accordingly. The mean reinspection interval was 3.18 years after NAI, 2.96 after VAI, and 2.12 after OAI (P < .001). Median times were 4.08, 3.51, and 3.02 years, respectively. Conclusion: FDA inspection activity had not returned to prepandemic levels by 2025. Unfavorable inspection outcomes were associated with earlier reinspection, consistent with risk-responsive oversight. However, the available data cannot establish whether follow-up was timely or adequate.

  • Open Access

    Managing the Human Fallback: Skill Investment Under Improving AI and Worker Mobility

    by Simrita Singh, Naireet Ghosh, Tinglong Dai

    When firms deploy autonomous AI, they must decide how much work to leave to the system and how much to keep workers engaged. This decision affects current output and future human capital. We develop a parsimonious two-period model in which AI may outperform the worker when it functions, but may fail with positive probability. A firm chooses worker engagement; engagement lowers current output for below-benchmark workers, but changes future skill through learning and erosion. We distinguish two dimensions of AI progress: capability, the system's output when it works, and reliability, the probability that it works. In a single-firm benchmark, engagement is valuable only as fallback investment. The firm engages the least-skilled workers most, because they have the largest skill gaps and are least costly to bring toward a useful fallback level. With worker mobility, engagement also affects labor-market sorting: workers prefer jobs that build more valuable skill trajectories. This sorting motive targets higher-skill workers near the AI frontier, where skill gains are more valuable and engagement is less costly. Mobility can therefore reverse the engagement pattern, shifting investment from the least-skilled toward the most-skilled workers below the AI benchmark. Mobility also reshapes how AI progress affects engagement: greater capability raises engagement by increasing the value of the skill trajectory a firm offers, whereas greater reliability can raise or lower it because it reduces fallback need while also changing learning opportunities. Under worker mobility, human–AI work design becomes a problem of human-capital investment, in which allocating work today shapes future skill.

  • Open Access

    Would You Trust a Doctor Who Uses ChatGPT?

    by Tinglong Dai

    In this chapter, we examine what generative AI is doing to medicine. Drawing on a body of recent evidence, we argue that technical progress is necessary but not sufficient: even as models improve, the institutional scaffolding that absorbs them will decide whether that progress translates into trustworthy care. Visible AI use can lower peer and patient ratings of physicians. Liability and reimbursement pressures distort when and how AI is used. Sustained reliance can corrode the skills clinicians still need. Today's generative AI systems produce what is plausible, not what is feasible under hard constraints, and the agent frameworks built around them do not close that gap. The enthusiasm for AI as a substitute for clinical judgment is, at best, premature and, at worst, a category mistake. Better models will help, but only alongside the procurement, workflow design, local validation, and auditable governance that turn a technical artifact into a clinical instrument. In medicine, trust has always been institutional. The question is not whether to use AI, but whether the surrounding clinical system has made that use worthy of trust.

  • Incentive Design and Pricing Under Limited Inventory

    by Ruiting Zuo, Tinglong Dai, Jussi Keppo

    Problem definition: A firm faces random demand for a service it delivers on a given future date. To boost demand, the firm hires a sales agent who exerts unobservable effort continuously over time. The firm is concerned not only with increasing current demand, but also with smoothing demand over time to avoid losing goodwill if realized demand exceeds available inventory. Methodology/results: We study the firm’s incentive design problem using a novel continuous-time principal-agent framework, in which demand drifts over time in response to the agent’s unobserved effort, as well as the price the firm charges. To induce the agent’s sales effort, the firm chooses an incentive scheme that depends on the remaining inventory and the time to the service (e.g., time to departure in the case of airlines). We characterize the firm’s optimal incentive scheme under both static and dynamic pricing policies. Using parameter values calibrated from the airline industry, we numerically show that under dynamic pricing, the use of a static incentive scheme helps the firm reap nearly all the benefits of the corresponding dynamic incentive scheme. In contrast, the use of a fully static strategy results in a significant loss of efficiency. Managerial implications: Comparing partially dynamic strategies, we find that dynamic contracting outperforms dynamic pricing when inventory is abundant. However, under limited inventory, the relative advantage depends nonmonotonically on demand elasticity: dynamic pricing dominates over the empirically relevant range of moderate to high elasticity, whereas dynamic contracting becomes more effective when elasticity is very low or in theoretical limits of extreme elasticity. History: This paper was selected as part of the 1RR initiative between the M&SOM journal and the MSOM Society. This paper was part of the 2024 MSOM Interface of Finance, Operations and Risk Management (iFORM) SIG Conference. Funding: R. Zuo acknowledges funding support from the Guangzhou Municipal Education Bureau University Research Project [Grant 2024312314] and the Guangzhou-HKUST(GZ) Joint Funding Program [Grant 2024A03J0630]. T. Dai was partially supported by the Bernard T. Ferrari Professorship at Johns Hopkins University. J. Keppo acknowledges funding support from the National Research Foundation, Singapore, and A*STAR under its RIE2020 Industry Alignment Fund–Industry Collaboration Projects grant call [IAF-ICP Grant No. I2001E0059, SIA-NUS Digital Aviation Corp Lab]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2026.0118 .

  • The Spillover Effect of Suspending Nonessential Surgery: Evidence from Kidney Transplantation

    by Minmin Zhang, Guihua Wang, Tinglong Dai

    Kidney transplantation is a life-saving procedure for patients with end-stage kidney disease. A delay in kidney transplantation has serious, even life-threatening, consequences. This study examines the impact of suspending nonessential surgery on deceased-donor kidney transplantation. Between March and April 2020, amid the COVID-19 pandemic, multiple states in the United States temporarily suspended nonessential surgery. Although these suspensions were not intended for deceased-donor kidney transplantation, because it is an essential surgery, the literature on service operations implies such suspensions may have either a positive or a negative spillover effect, depending on whether hospitals maintain or reduce resources. Motivated by these divergent implications, we estimate the potential spillover effect of suspending nonessential surgery on deceased-donor kidney transplantation. Analyzing a data set of all U.S. kidney transplant procedures, we observe a steep decline in transplant volume during the early months of the pandemic, with states that suspended nonessential surgery experiencing steeper declines. Using a difference-in-differences approach, we estimate a 13% reduction in transplant volume due to state-level suspension of nonessential surgery. This negative spillover effect is particularly pronounced in low-efficiency transplant centers with long cold ischemia times (CITs) but less so in high-efficiency centers, suggesting CIT is a key indicator of operational resilience against disruptions. Our mediation analysis reveals more than 40% of the spillover effect is attributable to changes in healthcare employment. This study indicates that in future public health crises, policymakers should adopt nuanced strategies to secure the healthcare workforce essential for supporting critical services, especially in centers with longer CITs. This paper was accepted by Jayashankar Swaminathan, operations management. Funding: This work was supported by the Johns Hopkins Bloomberg School of Public Health [Hopkins Business of Health Initiative seed grant, “COVID-19 and the Business of Health”]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.03624 .

  • Open Access

    Assured autonomy: How operations research powers and orchestrates generative AI systems

    by Tinglong Dai, David Simchi-Levi, Michelle Xiao Wu, Yao Xie

    Generative artificial intelligence (GenAI) is shifting from conversational assistants toward agentic systems—autonomous decision-making systems that sense, decide, and act within operational workflows. This shift creates an autonomy paradox: as GenAI systems are granted greater operational autonomy, they should, by design, embody more formal structure, more explicit constraints, and stronger tail-risk discipline. We argue that stochastic generative models can be fragile in operational domains unless paired with mechanisms that provide verifiable feasibility, robustness to distribution shift, and stress testing under high-consequence scenarios. To address this challenge, we develop a conceptual framework for assured autonomy grounded in operations research (OR), built on two complementary approaches. First, flow-based generative models frame generation as deterministic transport characterized by an ordinary differential equation, enabling auditability, constraint-aware generation, and connections to optimal transport, robust optimization, and sequential decision control. Second, operational safety is formulated through an adversarial robustness lens: decision rules are evaluated against worst-case perturbations within uncertainty or ambiguity sets, making unmodeled risks part of the design. This framework clarifies how increasing autonomy shifts OR’s role from solver to guardrail to system architect, with responsibility for control logic, incentive protocols, monitoring regimes, and safety boundaries. These elements define a research agenda for assured autonomy in safety-critical, reliability-sensitive operational domains.


  • Open Access

    Financial incentives increase uptake and perceived effectiveness of autonomous medical AI, yet patients still seek human reconfirmation

    by Haiyang Yang, Tinglong Dai, Risa M. Wolf

    Artificial intelligence (AI) is increasingly used for diagnostic screening. In diabetic retinopathy screening, autonomous AI systems can identify disease without a specialist. However, patient trust in AI remains a challenge, and it remains unknown whether and how financial incentives-commonly used to promote uptake of medical services-can increase patient acceptance of AI-based care. We examined, in a randomized vignette experiment, whether eliminating a copay would increase patients' willingness to choose AI over an eye care professional's (ECP's) exam, and how patients would respond to an AI diagnosis. We conducted a randomized vignette experiment with 248 U.S. adults with type 1 diabetes, recruited online between August and December 2024. Participants were presented with a primary-care vignette in which they were due for annual diabetic eye screening. An autonomous AI tool was available as an alternative to ECP referral. Participants were randomized in a 2×2 design: (1) no copay vs. $50 copay for AI screening; (2) copay waiver sponsored by insurer vs. AI developer. Participants chose between AI or ECP screening, and rated AI effectiveness. Those choosing AI or ECP indicated their willingness to seek a second opinion (from ECP or AI, respectively) after receiving a normal vs. abnormal result. Waiving the copay significantly increased AI selection (81% vs. 43%, p < 0.001); the source of the incentive did not affect choice. No copay also improved perceived AI effectiveness (mean score 3.67 vs. 3.24, p = 0.02). Patients choosing AI (vs. ECP) were far more likely to seek reconfirmation after abnormal (mean 6.69 vs. 2.41) or even normal (3.43 vs. 2.47) results (both p < 0.001). Eliminating cost barriers increases participants' willingness to opt for AI-based screening, but they still seek human reconfirmation. Incentives can promote AI adoption, but integration strategies must address persistent preferences for ECP reconfirmation-even when AI results are normal.