Not sure whether tools to explain #MachineLearning and #AI models are ready for prime time in consumer credit? My new research with @FinRegLab and @jannspiess suggests cautious optimism. Join us virtually tomorrow, read the report, or get the tl;dr below

RT @FinRegLab: Signed up yet for tomorrow’s virtual @NIST workshop? @prstark13 & our research partners Drs. @BlattnerLaura & @jannspiess….

RT @BrookingsEcon: Financial services is an important case study for #AI governance, writes @melissakoide (@FinRegLab), both because of its….

RT @PJakiela: Open letter to AEA Executive Committee regarding the locations of the next ASSA meetings (scheduled for New Orleans & San Ant….

Great time to read this one again to understand why #Tether isn't immune to #TerraUSD challenges. "A few colleagues and I canvassed Wall Street traders to see if any had seen Tether buying anything. No one had."

RT @FinRegLab: Missed our virtual conference, “Machine Learning Explained” in April? It's now live on our website. The webinar focused on n….

“All of this screams light on climate, heavy on blockchain,” said Eli Mitchell-Larson, a researcher at @OxfordNetZero. The complexity of the systems “may obfuscate the key question: did you unequivocally remove or reduce a tonne of carbon?”.

How to interpret the output of explainability tools is key. In a complex world, there is often more than one correct response to the question why a model behaved a certain way. Don't expect sparse explanations to capture all relevant nuances of model behavior.

Choosing the right tools matters when using ML/AI models. Not all tools in our study performed well. Tools need to be matched to their task. Explaining a loan rejection for an individual is a very different challenge than explaining overall model fairness/disparities.

What we found: Some tools rose to the challenge - they identified important drivers of an adverse credit decision and were able to figure out aspects of the model that were driving disparities for minority applicants. But two key lessons for lenders:.

What did we do? Partnered with 7 leading AI companies (thank you @itsArthurAI, @Zest_AI, @StratyfyInc, @RelationalAI, @h2oai, @solas_ai, @fiddlerlabs) and got them to explain why our credit models rejected some loan applicants and exhibited disparities for minority applicants.

RT @StanfordHAI: “You can’t expect the same tool that tells why an applicant was rejected to also be the right tool to explain why a model….

RT @melissakoide: I’m so darn proud of the FinRegLab team and their awesome commitment and hard work over the past two years to get us to t….

RT @FinRegLab: Two days away! Have you signed up to watch yet? @StanfordHAI @NIST @CommerceGov .

Words of wisdom. Thanks @nealemahoney.

RT @FinRegLab: We’re Hiring! FinRegLab is seeking highly motivated individuals with backgrounds in financial services academic research, in….

RT @FinRegLab: Every week, ML underwriting models decide thousands of credit applications. @FinRegLab’s first major ML report describes cur….

RT @SoCalFirePhoto: We just entered Greenville on Hwy 89 This video was taken from Bidwell & Ann St. I’m so sorry for the town of Greenvill….

RT @CAFireScanner: #DixieFire (Plumas-Butte Co) - All of #Chester now under mandatory evacs & people should evacuate east to the shelter in….

"The concern is that when big tech firms enter the payments market, their access to user data from associated digital business lines may allow them to achieve a dominant position, leading to fees that are even higher than those charged [. ] currently.".