The Presidents, the Geneticists, and the AI Glitch
I saw someone on Bluesky testing ChatGPT's capabilities by asking it to draw a diagram of all U.S. presidents since Herbert Hoover. The results? Hilariously wrong — including "John F. Katwady" (1991–1968) and "Bill Christen" (1999–2001) — paired with equally strange headshots. See picture.
That surprised me. For over a year, I've used ChatGPT for literature searches, grant proposal drafts, and idea feedback, and it's been far more reliable.
So I decided to run my own test
I asked ChatGPT for the top 20 geneticists since Mendel, with a two-word summary of their main contribution and the years they were born and died. The written list it gave me was solid, with a link to where it got the info:
Twenty Top Geneticists Since Mendel
1. Hugo de Vries (1848–1935) — Gene rediscovery (WIRED, Wikipedia)
2. Carl Correns (1864–1933) — Cytoplasmic inheritance (Wikipedia)
3. Erich von Tschermak (1871–1962) — Mendel confirmation (Wikipedia)
4. William Bateson (1861–1926) — Genetics champion (juliantrubin.com)
5. Thomas Hunt Morgan (1866–1945) — Chromosome heredity (Wikipedia)
6. Alfred Sturtevant (1891–1970) — Genetic mapping (Wikipedia)
7. Barbara McClintock (1902–1992) — Jumping genes (Wikipedia)
8. George Beadle (1903–1989) — One gene enzyme (Wikipedia)
9. Oswald Avery (1877–1955) — DNA identity (News-Medical)
10. Erwin Chargaff (1905–2002) — Base composition (News-Medical)
11. Rosalind Franklin (1920–1958) — DNA structure (News-Medical)
12. James Watson (1928 ) — Double helix (Biology Wise)
13. Francis Crick (1916–2004) — Double helix (News-Medical, Biology Wise)
14. Herbert Boyer (1936 ) — Recombinant DNA (Biology Wise)
15. Paul Berg (1926–2023) — DNA recombination (Biology Wise)
16. Werner Arber (1929 ) — Restriction enzymes (Biology Wise)
17. Hamilton O. Smith (1931 ) — DNA cutting (Biology Wise)
18. Craig C. Mello (1960 ) — RNA interference (Wikipedia)
19. James V. Neel (1915–2000) — Genetic epidemiology (Wikipedia)
20. Wen Hsiung Li (1942 ) — Molecular clock (Wikipedia)
Then I asked it to make a diagram of the same list (see right). That's when it all went wrong: misspelled names, bad dates, and only 16 scientists instead of 20.
The takeaway
The problem wasn't the information — it was turning that information into an image. ChatGPT's written responses are mostly reliable, and of course, all outputs should be thoroughly reviewed for accuracy and validity (and sometimes it goes completely off the rails). And while it still makes mistakes, it's already a remarkable tool that will keep improving.
Therefore, before dismissing and laughing at its shortcomings, it is essential to recognize that AI tools are rapidly evolving and may soon assume responsibilities currently performed in our professional roles. The real question isn’t whether AI gets names right today in a picture — it’s how ready we are when it gets everything right tomorrow.
José Miguel
P.S. And yes, ChatGPT wrote my first draft of this based on a thread I wrote in Bluesky, and Grammarly and ChatGPT reviewed my final draft.
