UT Austin has stopped using a ML-based system called GRADE to evaluate applicants for its Ph.D. in CS following backlash, seven years after first introducing it
Lilah Burke / Inside Higher Ed : Tweets: @susanschorn , @microbiomdigest , and @hypervisible Tweets: @susanschorn : Holy fucking shit, this was in use for SEVEN YEARS???? U of Texas will stop using controversial algorithm to evaluate Ph.D. applicants https://www.insidehighered.com/ ... Elisabeth Bik / @microbiomdigest : U of Texas at Austin has stopped using a machine-learning system to evaluate applicants for its Ph.D. in computer science. Critics say the system exacerbates existing inequality in the field. https://www.insidehighered.com/ ... @hypervisible : “The UT researchers who made GRADE trained it on a database of past admissions decisions. The system uses patterns from those decisions to calculate its scores for candidates.” 🤯🤬 https://twitter.com/...
Context & Ripple Effects
GRADE ran for seven years inside UT Austin's CS Ph.D. admissions before researchers on Twitter flagged it, making this less a launch story than an exposure story: the system was retired only after outsiders noticed it. That places it squarely in the arc of algorithmic gatekeeping coverage stretching back to reporting that algorithms sifting job applications penalize the poor, through the UK statistics regulator's decision to review Ofqual's exam-grade algorithm the same summer.
The pattern across that coverage is consistent: automated screening enters quietly, runs without disclosure, and gets pulled only when someone outside the institution surfaces it — as happened again when a medical student reverse-engineered the AI tools used by medical colleges on suspicion they were filtering his applications.
First-order effects
- Applicants to UT Austin's computer science Ph.D. program will be evaluated by human reviewers rather than GRADE, whose critics argued it amplified existing inequality in the field.
- The UT researchers who built GRADE face public attribution of the system's flaws, since the backlash named them directly.
Second-order effects
- Any other department or university running comparable ML screeners now faces a disclosure question: the cheapest defense is auditing and explaining the model before critics do it for them.
- Admissions-technology vendors selling automated screening inherit the burden of proof, since the GRADE episode gives faculty senates and graduate programs a concrete precedent for demanding human fallbacks.
Third-order effects
- If the pattern holds, algorithmic gatekeeping in admissions and hiring gets policed primarily by adversarial publicity — reverse-engineering and social-media callouts — rather than by pre-deployment audits, pushing institutions toward formal review regimes like the one applied to Ofqual's grading algorithm.
- Seven years of undetected use suggests universities lack inventory of their own models, pointing toward governance requirements that treat admissions algorithms as auditable records rather than internal tools.
The trend: Algorithmic screening in education and hiring is entering an era where deployment runs long and quiet, and retreat happens only when external scrutiny forces disclosure.