lukestein’s avatarlukestein’s Twitter Archive—№ 16,557

  1. Fresh WP: “Measuring and Mitigating Racial Bias in LLM Mortgage Underwriting” with Don Bowen, McKay Price, and Ke Yang Our audit study asks AI to assess simple mortgage applications (real HMDA data w/ randomized race and credit scores) 🦶🦶🏼🦶🏿 Download lukeste.in/llmbias 1/
    oh my god twitter doesn’t include alt text from images in their API
    1. …in reply to @lukestein
      ChatGPT (we also assess other models) • Claims it’s unbiased… • …but it 𝗶𝘀 biased against Black applicants. Particularly at low credit scores. (We also have this for other risk measures.) • We can partly close the racial gap just by asking for unbiased responses. 2/
      oh my god twitter doesn’t include alt text from images in their API
      1. …in reply to @lukestein
        LLMs recommend denying more loans and charging higher interest rates to Black applicants They would, on average, need credit scores ~120 points higher than white applicants to receive the same approval rate; ~30 higher to get same interest rate 3/
        oh my god twitter doesn’t include alt text from images in their API
        1. …in reply to @lukestein
          LLM’s mortgage racial bias biggest for applications with lower credit scores [or high DTI, high LTV]: • The Black–white approval rate gap is ~56% greater for low-score applicants than at average (13.3pp vs. 8.5) • Interest rate gap ~32% greater (47bp vs. 35) 4/
          1. …in reply to @lukestein
            We find anti-Black bias in mortgage underwriting recommendations from a number of LLMs Black borrowers with low credit scores suffer the most 5/
            oh my god twitter doesn’t include alt text from images in their APIoh my god twitter doesn’t include alt text from images in their API
            1. …in reply to @lukestein
              (Training data incl. long history of racial disparities in mortgages, plus prompts may trigger bias learned elsewhere) We give access to explicit race information, which should make it easy to 𝘢𝘷𝘰𝘪𝘥 discrimination if LLMs can ignore it, as they know they should… 6/
              1. …in reply to @lukestein
                Somehow, just asking LLM to be unbiased • Eliminates approval recommendation gap (on average and across different credit scores) • Reduces average racial interest rate gap by about 60% (from 35bp to 14), with even larger effects for lower-credit-score Black applicants 7/
                oh my god twitter doesn’t include alt text from images in their API
                1. …in reply to @lukestein
                  There’s more in the (first) draft lukeste.in/llmbias Obviously no one should (/would?😢) make critical financial decisions using prompts so simple. But genAI is getting deployed and importance of audits and intentional design are even more important in more complex systems!
                  1. …in reply to @lukestein
                    Two random extras: • My coauthors only use burners (smart!) which is why I didn’t tag them (won't even tell me their @.s). An honor to work with this Lehigh dream team! • Matt's tweet apropos for anyone who didn't know this and wants to read our tables x.com/matt_blackwell/status/1796596054948970594
                    1. …in reply to @lukestein
                      Update: @TedRalphs apparently convinced @Ke_Yang_ to set up a non-anon account 🤝 @TedRalphs/1798085056168345724