I'm a Ph.D. candidate in EECS at UC Berkeley, advised by Rediet Abebe and Moritz Hardt. My research broadly explores how modern AI should be designed, deployed, and evaluated when embedded in human and societal contexts. 

I am a part-time collaborator with Zillow, working with Andy Krause to study the causal effect of Zestimate on market outcomes. Stay tuned for the paper!

In Fall 2024, I had the privilege of visiting the Harvard EconCS group, where I worked with Rediet Abebe and Ariel Procaccia.

In Summer 2024, I interned at AppLovin, focusing on improving and evaluating their bidder and recommender systems. Since then, AppLovin’s stock has grown fivefold!

To follow my scientific activities, subscribe to my personal newsletter, or check out the news section below. 

🚨 I expect to complete my PhD in Dec. 2025 and am currently open to opportunities in both industry and academia. Please feel free to reach out if you think I might be a good fit, and I’d be happy to share more!

• 🚨 Upcoming: Oct. 2025: I'll give a talk on participatory objective design at the INFORMS annual meeting, Oct. 26-29, 2025, Atlanta, GA. Please send me an email if you'd like to meet there!

• 🚨 Upcoming: Nov. 2025: I'll give two talks on direct alignment with heterogeneous preferences and the hidden cost of waiting for accurate predictions at EAAMO, Nov. 5-7, 2025, Pittsburgh, PA. Please send me an email if you'd like to meet there!

• July 2025: I presented three posters on direct alignment with heterogeneous preferences and the value of costly signaling in interactive alignment with inconsistent preferences at the EC workshop on Human-AI Collaboration, the EC workshop on Information Economics x Large Language Models, and the EC workshop on Swap Regret and Strategic Learning, July 10, 2025, Stanford University, CA!

• June 2025: I gave a talk on the hidden cost of waiting for accurate predictions at FORC, June 4-6, 2025, Stanford University, CA!

• May 2025: I gave a short talk on direct alignment with heterogeneous preferences at the LLM Agent Workshop, Berkeley, CA! (recording)

• April 2025: Jiduan Wu delivered an oral presentation of our work on the hidden cost of waiting for accurate predictions at ICLR, Apr. 24-28, Singapore!

• Feb. 2025: ⚡ Our work on the hidden cost of waiting for accurate predictions is selected for an oral presentation at ICLR 2025 (top 1.8% of submissions)!

• Dec. 2024: I have passed my Qualifying exam and advanced to candidacy!

• Fall 2024: I visited Harvard, working with Rediet Abebe and Ariel Procaccia! 

• To see my complete and up-to-date list of publications, visit my Google Scholar profile.

• To find the accompanying codes for my papers, check out my GitHub profile.

• To access recordings, slides, codes, and related materials, see the topic-sorted list of my publications below.

My work focuses on the challenges that emerge when designing and evaluating algorithms, especially those powered by machine learning, in interaction with humans and society. I think about these challenges through three lenses:

What to optimize? Designing objectives that capture diverse, inconsistent, or pluralistic human preferences.

1. Ali Shirali*, Arash Nasr-Esfahany*, Abdullah Alomar, Parsa Mirtaheri, Rediet Abebe**, and Ariel Procaccia**. "Direct alignment with heterogeneous preferences." Under review. (arXiv version, short talk at Berkeley LLM workshop)

🚨 Upcoming: I'll present the above work at EAAMO, Nov. 5-7, 2025, Pittsburgh, PA. Please send me an email if you'd like to meet there! 

2. Ali Shirali*, Jessie Finocchiaro*, and Rediet Abebe. "Participatory objective design via preference elicitation." In Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 2024. (FAccT'24 version, FAccT'24 talk (email me if not accessible))

🚨 Upcoming: I'll present the above work at the INFORMS annual meeting, Oct. 26-29, 2025, Atlanta, GA. Please send me an email if you'd like to meet there!

How to act / make decision?  Developing algorithms for decision-making under uncertainty, drawing on reinforcement learning and statistical decision-making principles.

1. Ali Shirali, Ariel Procaccia*, and Rediet Abebe*. "The hidden cost of waiting for accurate predictions." (oral ⚡) In The Thirteenth International Conference on Learning Representations (ICLR), 2025. (ICLR'25 version, arXiv version)

🚨 Upcoming: I'll present the above work at EAAMO, Nov. 5-7, 2025, Pittsburgh, PA. Please send me an email if you'd like to meet there! 

2. Ali Shirali, Rediet Abebe*, and Moritz Hardt*. "Allocation requires prediction only if inequality is low." (spotlight ⚡) In Forty-first International Conference on Machine Learning (ICML), 2024. (ICML'24 version, arXiv version)

3. Ali Shirali*, Alexander Schubert*, and Ahmed Alaa. "Pruning the way to reliable policies: a multi-objective deep Q-learning approach to critical care." In IEEE Journal of Biomedical and Health Informatics (JBHI), 2024. (IEEE version, arXiv version)

How to tell if it works?  Developing evaluation frameworks that reflect societal contexts, from causal and performative impacts to biases, feedback loops, and other distortions in evaluation.

1. Ali Shirali and Moritz Hardt. "What makes ImageNet look unlike LAION." In Transactions on Machine Learning Research (TMLR), 2025, (TMLR version, TMLR recording, arXiv version, code)

2. Ali Shirali, Rediet Abebe, and Moritz Hardt. "A theory of dynamic benchmarks." In the Eleventh International Conference on Learning Representations (ICLR), 2023. (ICLR'23 version, arXiv version)

3. Ali Shirali. "Sequential nature of recommender systems disrupts the evaluation process."  In Advances in Bias and Fairness in Information Retrieval (BIAS), 2022. (BIAS'22 version, arXiv version, code, BIAS'22 talk, BIAS'22 slides)

4. Rediet Abebe, Nicole Immorlica, Jon Kleinberg, Brendan Lucier, and Ali Shirali. "On the effects of triadic closure on network segregation." In Proceedings of the  ACM Conference on Economics and Computation (EC), 2022.  (EC'22 version, arXiv version, poster, short talk, EC'22 talk, TOC4Fairness talk)

5. Mahdiyar Shahbazi*, Ali Shirali*, Hamid Aghajan, and Hamed Nili. "Using distance on the Riemannian manifold to compare representations in brain and in models." In NeuroImage, 2021. (NeuroImage version, code)

Thesis:

1. (In Persian) Ali Shirali. "Prediction of customer churn from subscription services in response to recommendations." [Master's dissertation, Sharif University]. (Sharif University Library)

* and ** mean equal contribution