Bridging Prediction and Intervention Problems in Social Systems (Simons Institute, 2026)
2026-01-12..2026-01-18
Berkeley, United States
who it's for
causal inference researchers, applied machine learning researchers, human-AI interaction researchers, policy and public-sector analysts, education-technology researchers, clinical informatics researchers, evaluation designers
topics
counterfactual risk assessment; LLM-guided reinforcement learning; dynamic thresholding designs; human-AI agreement calibration; clinical algorithm fairness; operational dosage in RCTs; personalization value estimation; probabilistic decision making by LLMs; scheduling with time-evolving uncertainty; algorithmic UDAP and consumer protection; revealed-preference utility elicitation from LLMs
| title | author | length |
|---|---|---|
| Triage Score: A Counterfactual Risk Assessment Instrument | Kosuke Imai | 40m |
Proposes a triage score that replaces a risk-only prediction with expected counterfactual utilities and shows point identification under an additive… | ||
| LLM-Guided Reinforcement Learning for Mastery Learning: Large-Scale Field Evidence | Hamsa Bastani | 35m |
A city-scale RCT enrolls ~1,000 Taipei high schoolers where everyone uses a GPT‑4 LLM-powered AI tutor with pedagogical guardrails and the randomization… | ||
| Non-parametric Causal Inference in Dynamic Thresholding Designs | Stefan Wager | 30m |
Argues that threshold rules in time-series settings identify a dynamic marginal policy effect (long-run benefit per extra treated person) and gives a concrete… | ||
| Agreement and Alignment for Human-AI Collaborative Decision Making | Aaron Roth | 45m |
Defines a frequentist online model where each day a human and an AI exchange short probability or action messages, observe the outcome, and algorithms enforce… | ||
| Human Values in Clinical Algorithms, Old and New | Arjun Kumar Manrai | 39m |
Lab builds an AI CPC discussant called Dr. CabOT by combining GPT-4/O3-style models, embeddings of all NEJM case records, a literature search RAG layer and… | ||
| When Prediction Becomes Policy: What Do Risk Models Actually Learn? | Shion Guha | 34m |
Predictive risk models trained on administrative and narrative case notes in child welfare learn patterns of agency contact, surveillance intensity, and… | ||
| Evaluating probabilistic decision making by LLMs: a decision-theoretic framework and application... | Bryan Wilder | 32m |
Proposes a revealed-preference elicitation that asks an LLM for both verbalized probabilities and its chosen action, fits a discrete-choice utility model… | ||
| Algorithmic UDAP | Talia Gillis | 32m |
Compares disparate impact and UDAAP by formalizing both in a simulated lending setting using Adult Income, German Credit and HMDA data, training many… | ||
| Opening Remarks | Lydia Liu, Angela Zhou, Inioluwa Deborah Raji | 25m |
Claims that predictive accuracy alone fails to predict downstream impact and that an integrated viewpoint is necessary, with a formal split between predictive… | ||
| Operational Dosage: The Impact of Capacity Constraints on RCT Evaluation | Hannah Li | 35m |
RCTs of human-delivered services show that finite staffing creates an operational dosage effect: as patient count rises per fixed number of callers, queueing… | ||
| Estimating the Value of Personalization | Emma Brunskill | 35m |
Introduces a new estimator that, from historical data, tests whether a personalized policy yields higher expected outcomes than the best one-size-fits-all… | ||
| The Role of Standardized Tests in Admissions: Predictions vs. Decisions | Sharad C. Goel | 34m |
Analysis uses a held-out predictive model on 16,000 applications to first-year grades; adding GREs raises predictive fit from R² from .19 to .31 but simulated… | ||
| Individual Gaps in Prediction Generalization | Marzyeh Ghassemi | 31m |
Shows that DenseNet and ERM-trained image models encode sex, age, race and that greater encoding correlates with larger group performance gaps (over 20% gap… | ||
| Beyond Prediction Performance: How Modeling Decisions Shape Fairness Outcomes in Statistical... | Frauke Kreuter | 28m |
German administrative labor market records (2% sample) predict long‑term unemployment across about 700,000 episodes, with regression and machine‑learning… | ||
| The Feedback Loop of Statistical Discrimination | Jackie Baek | 32m |
Model of repeated hiring where the firm's evaluation noise for a group improves only when that group supplied hires last period, so the initial noise level… | ||
| Decision theoretic foundations for human-AI collaboration | Jessica Hullman | 31m |
Introduces decision-theoretic benchmarks that quantify when pairing humans and models is actually valuable, defining a value of complementation and a… | ||
| Evaluation Design | Ashia Wilson | 29m |
Evaluation scores are statistical claims about value and should state the estimand and inference assumptions; common metrics hide cost and prevalence… | ||
| Recommendation in equilibrium for the NYC High School Match | Nikhil Garg | 33m |
Collaboration with NYC Public Schools designs and deploys an email-based informational intervention that recommends nearby high-performing high schools with… | ||
| Automating the US Welfare State: Historical Experiments in Technological Federalism | Marc Aidinoff | 36m |
Historical research shows U.S. welfare computerization aims less at abstract prediction than at solving political coordination: microsimulation tools like… | ||
| On Machine Learning for Prediction and Prioritization in the Allocation of Scarce Societal Resources | Sanmay Das | 35m |
Shows that ML is most useful as measurement not as final allocator: BART models estimate household re‑entry probabilities from St. Louis intake data… | ||
| Machine Learning Who to Nudge: Causal vs Predictive Targeting in a Field Experiment on Student... | Jann Spiess | 34m |
Field experiment on 25,000 CUNY students uses randomized texts to raise FAFSA renewals and finds a 6.4 percentage-point average boost from 37% baseline. Causal… | ||
| The Challenge of Valid Evaluations | Amanda Coston | 33m |
Argues standard train/test evaluation misleads when decisions affect outcomes and measurements are proxies, and shows concrete methods to fix that: derive… | ||
| The Inadequacy of Offline LLM Evaluations: A Need to Account for Personalization in Model Behavior | Angelina Wang | 32m |
Claim: offline one-shot API evaluations miss how models behave when personalized in chats. Empirical setup: 800 real users of ChatGPT and Google Gemini… | ||
| Scheduling with Time-Evolving Uncertainty for Content Review Prioritization in Social Media | Thodoris Lykouris | 49m |
Introduces a new queueing model where post view counts resolve over time and develops an asymptotically optimal scheduling algorithm (an opportunity‑adjusted… | ||
| nothing matches | ||