Case Dossier — St George's Hospital Medical School Admissions Program (1980s)

Historical anchor for Session 2, paired with Ofqual 2020: the same failure, forty years apart.

What happened

In the late 1970s–early 1980s, St George's Hospital Medical School (London) automated the first-round screening of admissions applications with a computer program written to replicate the judgments of its human selection panel. It replicated them faithfully — including their biases.

The program systematically deducted points from women and from applicants with non-European names, reducing their chance of being called to interview. Because it encoded historical panel behaviour, the discrimination was consistent, invisible to applicants, and ran for years. Internal staff eventually raised concerns, and in 1988 the UK Commission for Racial Equality found St George's guilty of discrimination in its admissions practice. The school offered redress to some affected applicants. It remains one of the earliest documented cases of algorithmic discrimination in education.

The two lessons that survive to 2026

1. Automation launders bias into consistency. The program was arguably fairer than the panel in variance — it applied the same prejudice to everyone, every time. Consistency is not fairness; it is bias at scale with an audit-proof face.
2. "The computer decided" diffuses accountability. For years nobody owned the discrimination because nobody performed it. The machine did — and the machine was just doing what the institution had always done.

Discussion prompts

1. The program only mirrored the human panel. Is the program the problem, the panel, or the decision to freeze the panel's behaviour into code?
2. Who should have caught this, and at what stage — design, deployment, or review?
3. Map forward: a modern LLM is trained on historical human text the way this program was trained on historical panel decisions. What's the equivalent failure in AI-assisted marking, reference letters, or admissions today?

Facilitator notes

• This case predates "AI" discourse entirely — that is its power. It kills the "this is a new problem needing new ethics" framing: the principles violated (justice, explicability, accountability) were fully legible in 1988.
• Keep claims to the documented core: automated screening, penalties tied to gender and to non-European names, CRE finding in 1988. Avoid invented specifics (exact point values vary across retellings; don't cite numbers).
• Bridge to the bias demo (P03): "The 1980s version needed someone to code the bias. The 2020s version absorbs it from training data. Let's produce it live."