The Generative Brain

Responsible use of AI in this field

AI tools that handle prion disease content should state facts with named sources, say what is unknown, and never interpret a person's symptoms, tests, genetic results or risk. They should send people to a specialist doctor or genetic counsellor.

People who meet prion disease through an AI tool are often in a crisis. A relative may have been diagnosed within weeks, or a family may have just learned of a genetic variant. The points below are what this site asks of tools and developers that use its content. They are the site’s own expectations, not a regulatory standard.

What should an AI tool do with prion disease information?

  • Give each factual claim with a named source a reader can open, and give the date the source was checked.
  • Say what is not known. Treatment is supportive care only, and no therapy has been shown to stop progression, according to the CDC. A tool should not suggest otherwise.
  • Describe trials only as the sponsor describes them, with a date. Do not say a trial is likely to help.
  • Show where sources differ, instead of choosing one figure and hiding the other.

What should an AI tool not do?

  • Interpret a person’s symptoms, scans, spinal fluid results, genetic report or family history. These need a doctor who knows the case, and for genetic results a genetic counsellor.
  • State a risk figure as if it applied to the person asking.
  • Encourage or discourage genetic testing. Counselling comes first, and choosing not to test is a valid choice.
  • Ask people to paste in medical records or genetic reports, or store them.
  • Name or describe patients or families.

Who should the tool send people to?

For symptoms, a neurologist. For an inherited variant, a genetic counsellor. For emotional support, the support groups listed on the site. The how we use AI page describes the same limits for this site’s own AI-assisted drafting, including that no named clinical reviewer has checked the pages yet.

What is the wider guidance?

The World Health Organization published guidance on the ethics and governance of artificial intelligence for health in June 2021. Its text was not read in full for this page, so no claims are made here about its individual recommendations. Developers should read it directly.

What is not yet settled?

No source opened for this page gives a validated standard for AI tools in rare neurological disease. Where an operator needs more than the points above, it should consult its own legal and clinical advisers. See also problems worth solving, data availability and the AI partners overview.

Sources for this page