Berk Ustun
@berkustun.bsky.social
Assistant Prof at UCSD. I work on safety, interpretability, and fairness in machine learning. www.berkustun.com
Reposted by Berk Ustun
UK government project using AI to find benefit fraud resulted in:
- A 46% false fraud rate
- Anguish for families who were wrongly accused of fraud and had benefits stopped
- Months of additional work for government, setting up a hotline, correcting false fraud
www.theguardian.com/society/2025...
- A 46% false fraud rate
- Anguish for families who were wrongly accused of fraud and had benefits stopped
- Months of additional work for government, setting up a hotline, correcting false fraud
www.theguardian.com/society/2025...
HMRC trial of child benefit crackdown wrongly suspected fraud in 46% of cases
Exclusive: Almost half of families flagged as emigrants based on Home Office travel data were still living in UK
www.theguardian.com
November 9, 2025 at 3:11 PM
UK government project using AI to find benefit fraud resulted in:
- A 46% false fraud rate
- Anguish for families who were wrongly accused of fraud and had benefits stopped
- Months of additional work for government, setting up a hotline, correcting false fraud
www.theguardian.com/society/2025...
- A 46% false fraud rate
- Anguish for families who were wrongly accused of fraud and had benefits stopped
- Months of additional work for government, setting up a hotline, correcting false fraud
www.theguardian.com/society/2025...
Reposted by Berk Ustun
I’m giving an IDE seminar at @mitsloan.bsky.social tomorrow at 11am, on optimizing AI as decision support. Joint work w/ @ziyang.bsky.social @yifanwu.bsky.social @jasonhartline.bsky.social @berkustun.bsky.social
Come by if you’re around!
www.eventbrite.com/e/fall-2025-...
Come by if you’re around!
www.eventbrite.com/e/fall-2025-...
October 22, 2025 at 7:06 PM
I’m giving an IDE seminar at @mitsloan.bsky.social tomorrow at 11am, on optimizing AI as decision support. Joint work w/ @ziyang.bsky.social @yifanwu.bsky.social @jasonhartline.bsky.social @berkustun.bsky.social
Come by if you’re around!
www.eventbrite.com/e/fall-2025-...
Come by if you’re around!
www.eventbrite.com/e/fall-2025-...
Reposted by Berk Ustun
Who teaches an undergraduate principles of programming languages class? Looking for some inspiration to teach one at UCSD
September 22, 2025 at 11:48 PM
Who teaches an undergraduate principles of programming languages class? Looking for some inspiration to teach one at UCSD
Reposted by Berk Ustun
In a new paper, I try to resolve the counterintuitive evidence of Meehl’s “clinical vs statistical prediction” problems: Statistics only wins because the game is rigged.
The Actuary's Final Word on Algorithmic Decision Making
Paul Meehl's foundational work "Clinical versus Statistical Prediction," provided early theoretical justification and empirical evidence of the superiority of statistical methods over clinical judgmen...
arxiv.org
September 8, 2025 at 2:48 PM
In a new paper, I try to resolve the counterintuitive evidence of Meehl’s “clinical vs statistical prediction” problems: Statistics only wins because the game is rigged.
Reposted by Berk Ustun
Machine learning models can assign fixed predictions that preclude individuals from changing their outcome. Think credit applicants that can never get a loan approved, or young patients that can never get an organ transplant - no matter how sick they are!
July 14, 2025 at 4:11 PM
Machine learning models can assign fixed predictions that preclude individuals from changing their outcome. Think credit applicants that can never get a loan approved, or young patients that can never get an organ transplant - no matter how sick they are!
Reposted by Berk Ustun
Excited to be chatting about our new paper "Understanding Fixed Predictions via Confined Regions" (joint work with @berkustun.bsky.social, Lily Weng, and Madeleine Udell) at #ICML2025!
🕐 Wed 16 Jul 4:30 p.m. PDT — 7 p.m. PDT
📍East Exhibition Hall A-B #E-1104
🔗 arxiv.org/abs/2502.16380
🕐 Wed 16 Jul 4:30 p.m. PDT — 7 p.m. PDT
📍East Exhibition Hall A-B #E-1104
🔗 arxiv.org/abs/2502.16380
Understanding Fixed Predictions via Confined Regions
Machine learning models can assign fixed predictions that preclude individuals from changing their outcome. Existing approaches to audit fixed predictions do so on a pointwise basis, which requires ac...
arxiv.org
July 14, 2025 at 4:08 PM
Excited to be chatting about our new paper "Understanding Fixed Predictions via Confined Regions" (joint work with @berkustun.bsky.social, Lily Weng, and Madeleine Udell) at #ICML2025!
🕐 Wed 16 Jul 4:30 p.m. PDT — 7 p.m. PDT
📍East Exhibition Hall A-B #E-1104
🔗 arxiv.org/abs/2502.16380
🕐 Wed 16 Jul 4:30 p.m. PDT — 7 p.m. PDT
📍East Exhibition Hall A-B #E-1104
🔗 arxiv.org/abs/2502.16380
Reposted by Berk Ustun
ExplainableAI has long frustrated me by lacking a clear theory of what an explanation should do. Improve use of a model for what? How? Given a task what's max effect explanation could have? It's complicated bc most methods are functions of features & prediction but not true state being predicted 1/
July 2, 2025 at 4:53 PM
ExplainableAI has long frustrated me by lacking a clear theory of what an explanation should do. Improve use of a model for what? How? Given a task what's max effect explanation could have? It's complicated bc most methods are functions of features & prediction but not true state being predicted 1/
Reposted by Berk Ustun
Having a lot of FOMO not being able to be in person at #FAccT2025 but enjoying the virtual transmission 💻. Tomorrow Jakob will be presenting our paper "Perils of Label Indeterminacy: A Case Study on Prediction of Neurological Recovery After Cardiac Arrest".
June 25, 2025 at 9:30 PM
Having a lot of FOMO not being able to be in person at #FAccT2025 but enjoying the virtual transmission 💻. Tomorrow Jakob will be presenting our paper "Perils of Label Indeterminacy: A Case Study on Prediction of Neurological Recovery After Cardiac Arrest".
Explanations don't help us detect algorithmic discrimination. Even when users are trained. Even when we control their beliefs. Even under ideal conditions... 👇
Right to explanation laws assume explanations help people detect algorithmic discrimination.
But is there any evidence for that?
In our latest work w/ David Danks @berkustun, we show explanations fail to help people, even under optimal conditions.
PDF shorturl.at/yaRua
But is there any evidence for that?
In our latest work w/ David Danks @berkustun, we show explanations fail to help people, even under optimal conditions.
PDF shorturl.at/yaRua
June 24, 2025 at 7:16 PM
Explanations don't help us detect algorithmic discrimination. Even when users are trained. Even when we control their beliefs. Even under ideal conditions... 👇
Reposted by Berk Ustun
“Science is a smart, low cost investment. The costs of not investing in it are higher than the risk of doing so… talk to people about science.” - @kevinochsner.bsky.social makes his case to the field #sans2025
April 26, 2025 at 9:03 PM
“Science is a smart, low cost investment. The costs of not investing in it are higher than the risk of doing so… talk to people about science.” - @kevinochsner.bsky.social makes his case to the field #sans2025
Reposted by Berk Ustun
Hey AI folks - stop using SHAP! It won't help you debug [1], won't catch discrimination [2], and makes no sense for feature importance [3].
Plus - as we show - it also won't give recourse.
In a paper at #ICLR we introduce feature responsiveness scores... 1/
arxiv.org/pdf/2410.22598
Plus - as we show - it also won't give recourse.
In a paper at #ICLR we introduce feature responsiveness scores... 1/
arxiv.org/pdf/2410.22598
April 24, 2025 at 4:37 PM
Hey AI folks - stop using SHAP! It won't help you debug [1], won't catch discrimination [2], and makes no sense for feature importance [3].
Plus - as we show - it also won't give recourse.
In a paper at #ICLR we introduce feature responsiveness scores... 1/
arxiv.org/pdf/2410.22598
Plus - as we show - it also won't give recourse.
In a paper at #ICLR we introduce feature responsiveness scores... 1/
arxiv.org/pdf/2410.22598
Reposted by Berk Ustun
When RAG systems hallucinate, is the LLM misusing available information or is the retrieved context insufficient? In our #ICLR2025 paper, we introduce "sufficient context" to disentangle these failure modes. Work w Jianyi Zhang, Chun-Sung Ferng, Da-Cheng Juan, Ankur Taly, @cyroid.bsky.social
April 24, 2025 at 6:18 PM
When RAG systems hallucinate, is the LLM misusing available information or is the retrieved context insufficient? In our #ICLR2025 paper, we introduce "sufficient context" to disentangle these failure modes. Work w Jianyi Zhang, Chun-Sung Ferng, Da-Cheng Juan, Ankur Taly, @cyroid.bsky.social
Reposted by Berk Ustun
Denied a loan, an interview, or an insurance claim by machine learning models? You may be entitled to a list of reasons.
In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse
In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse
April 24, 2025 at 6:19 AM
Denied a loan, an interview, or an insurance claim by machine learning models? You may be entitled to a list of reasons.
In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse
In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse
Reposted by Berk Ustun
Absolute banger.
April 19, 2025 at 7:43 PM
Absolute banger.
Reposted by Berk Ustun
Many ML models predict labels that don’t reflect what we care about, e.g.:
– Diagnoses from unreliable tests
– Outcomes from noisy electronic health records
In a new paper w/@berkustun, we study how this subjects individuals to a lottery of mistakes.
Paper: bit.ly/3Y673uZ
🧵👇
– Diagnoses from unreliable tests
– Outcomes from noisy electronic health records
In a new paper w/@berkustun, we study how this subjects individuals to a lottery of mistakes.
Paper: bit.ly/3Y673uZ
🧵👇
April 19, 2025 at 11:04 PM
Many ML models predict labels that don’t reflect what we care about, e.g.:
– Diagnoses from unreliable tests
– Outcomes from noisy electronic health records
In a new paper w/@berkustun, we study how this subjects individuals to a lottery of mistakes.
Paper: bit.ly/3Y673uZ
🧵👇
– Diagnoses from unreliable tests
– Outcomes from noisy electronic health records
In a new paper w/@berkustun, we study how this subjects individuals to a lottery of mistakes.
Paper: bit.ly/3Y673uZ
🧵👇
Reposted by Berk Ustun
🚨 Excited to announce a new paper accepted at #ICLR2025 in Singapore!
“Learning Under Temporal Label Noise”
We tackle a new challenge in time series ML: label noise that changes over time 🧵👇
arxiv.org/abs/2402.04398
“Learning Under Temporal Label Noise”
We tackle a new challenge in time series ML: label noise that changes over time 🧵👇
arxiv.org/abs/2402.04398
Learning under Temporal Label Noise
Many time series classification tasks, where labels vary over time, are affected by label noise that also varies over time. Such noise can cause label quality to improve, worsen, or periodically chang...
arxiv.org
April 13, 2025 at 5:40 PM
🚨 Excited to announce a new paper accepted at #ICLR2025 in Singapore!
“Learning Under Temporal Label Noise”
We tackle a new challenge in time series ML: label noise that changes over time 🧵👇
arxiv.org/abs/2402.04398
“Learning Under Temporal Label Noise”
We tackle a new challenge in time series ML: label noise that changes over time 🧵👇
arxiv.org/abs/2402.04398
Reposted by Berk Ustun
The CHI Human-Centered Explainable AI Workshop is back!
Paper submissions: Feb 20
hcxai.jimdosite.com
Paper submissions: Feb 20
hcxai.jimdosite.com
Home | HCXAI
ACM CHI 2025 Workshop on Human-Centered Explainable AI (HCXAI). May 2025 (Yokohama, Japan & hybrid). Submit your Paper (EasyChair)
hcxai.jimdosite.com
February 5, 2025 at 11:34 PM
The CHI Human-Centered Explainable AI Workshop is back!
Paper submissions: Feb 20
hcxai.jimdosite.com
Paper submissions: Feb 20
hcxai.jimdosite.com
Reposted by Berk Ustun
🧵on the CFPB and less discriminatory algorithms.
last week, in its supervisory highlights, the Bureau offered a range of impressive new details on how financial institutions should be searching for less discriminatory algorithms.
last week, in its supervisory highlights, the Bureau offered a range of impressive new details on how financial institutions should be searching for less discriminatory algorithms.
January 21, 2025 at 6:44 PM
🧵on the CFPB and less discriminatory algorithms.
last week, in its supervisory highlights, the Bureau offered a range of impressive new details on how financial institutions should be searching for less discriminatory algorithms.
last week, in its supervisory highlights, the Bureau offered a range of impressive new details on how financial institutions should be searching for less discriminatory algorithms.
Reposted by Berk Ustun
Engaging discussions on the future of #AI in #healthcare at this week's ICHPS, hosted by @amstatnews.bsky.social.
JCHI's @kdpsingh.bsky.social shared insights on the safety & equity of #MachineLearning algorithms and examined bias in large language models.
January 8, 2025 at 9:59 PM
Engaging discussions on the future of #AI in #healthcare at this week's ICHPS, hosted by @amstatnews.bsky.social.
JCHI's @kdpsingh.bsky.social shared insights on the safety & equity of #MachineLearning algorithms and examined bias in large language models.
Reposted by Berk Ustun
📣 CANAIRI: the Collaboration for Translational AI Trials! Co lead @xiaoliu.bsky.social @naturemedicine.bsky.social
Perhaps most important to AI translation is the local silent trial. Ethically, and from an evidentiary perspective, this is essential!
url.au.m.mimecastprotect.com/s/pQSsClx14m...
Perhaps most important to AI translation is the local silent trial. Ethically, and from an evidentiary perspective, this is essential!
url.au.m.mimecastprotect.com/s/pQSsClx14m...
url.au.m.mimecastprotect.com
January 6, 2025 at 10:36 PM
📣 CANAIRI: the Collaboration for Translational AI Trials! Co lead @xiaoliu.bsky.social @naturemedicine.bsky.social
Perhaps most important to AI translation is the local silent trial. Ethically, and from an evidentiary perspective, this is essential!
url.au.m.mimecastprotect.com/s/pQSsClx14m...
Perhaps most important to AI translation is the local silent trial. Ethically, and from an evidentiary perspective, this is essential!
url.au.m.mimecastprotect.com/s/pQSsClx14m...
Reposted by Berk Ustun
🪩New paper🪩 (WIP) appearing at @neuripsconf.bsky.social Regulatable ML and Algorithmic Fairness AFME workshop (oral spotlight).
In collaboration with @s010n.bsky.social and Manish Raghavan, we explore strategies and fundamental limits in searching for less discriminatory algorithms.
In collaboration with @s010n.bsky.social and Manish Raghavan, we explore strategies and fundamental limits in searching for less discriminatory algorithms.
December 13, 2024 at 1:34 PM
🪩New paper🪩 (WIP) appearing at @neuripsconf.bsky.social Regulatable ML and Algorithmic Fairness AFME workshop (oral spotlight).
In collaboration with @s010n.bsky.social and Manish Raghavan, we explore strategies and fundamental limits in searching for less discriminatory algorithms.
In collaboration with @s010n.bsky.social and Manish Raghavan, we explore strategies and fundamental limits in searching for less discriminatory algorithms.
Reposted by Berk Ustun
I'm seeking a postdoc to work with me and @kenholstein.bsky.social on evaluating AI/ML decision support for human experts:
statmodeling.stat.columbia.edu/2024/12/10/p...
P.S. I'll be at NeurIPS Thurs-Mon. Happy to talk about this position or related mutual interests!
Please repost 🙏
statmodeling.stat.columbia.edu/2024/12/10/p...
P.S. I'll be at NeurIPS Thurs-Mon. Happy to talk about this position or related mutual interests!
Please repost 🙏
Postdoc position at Northwestern on evaluating AI/ML decision support | Statistical Modeling, Causal Inference, and Social Science
statmodeling.stat.columbia.edu
December 10, 2024 at 6:18 PM
I'm seeking a postdoc to work with me and @kenholstein.bsky.social on evaluating AI/ML decision support for human experts:
statmodeling.stat.columbia.edu/2024/12/10/p...
P.S. I'll be at NeurIPS Thurs-Mon. Happy to talk about this position or related mutual interests!
Please repost 🙏
statmodeling.stat.columbia.edu/2024/12/10/p...
P.S. I'll be at NeurIPS Thurs-Mon. Happy to talk about this position or related mutual interests!
Please repost 🙏
Reposted by Berk Ustun
There is just about a month left before the abstract deadline for @facct.bsky.social 2025🥳😳🙈. Really looking forward to seeing all the submissions. If you are submitting, don’t forget to check out the Author Guide (and the reviewer and AC guide as well). facctconference.org/2025/aguide
ACM FAccT - 2025 Home
facctconference.org
December 11, 2024 at 9:28 AM
There is just about a month left before the abstract deadline for @facct.bsky.social 2025🥳😳🙈. Really looking forward to seeing all the submissions. If you are submitting, don’t forget to check out the Author Guide (and the reviewer and AC guide as well). facctconference.org/2025/aguide