BILL: Lance, I have a question and a meta question:
a) List all the presidents and vice presidents who were colorblind.
b) Do you know this sort of thing since you are colorblind, and hence
might have looked this up in the past?
LANCE: (b) No. And if Marco Rubio runs for president in 2028 he will need to earn my vote. He won't get it just for being colorblind. As for (a) I asked Claude.
Only one case is reasonably well documented, and it isn't a vice president. Bill Clinton is red-green colorblind: during the 1996 presidential debates, both Clinton and Bob Dole needed a modified cue-light system, since individual colored lights would have been hard for either of them to read, so a three-light arrangement was used instead — an anecdote that traces back to the Encyclopedia of Genetic Disorders and Birth Defects. Dole, of course, was Ford's running mate in 1976 and lost, so he never made it to the vice presidency.
Eisenhower is the other plausible case. His grandson David wrote that Ike confused browns and greens, sometimes asking for a brown suit when he meant green, and couldn't tell what color his golf balls were — someone had to write "DDE" on them. That's a family recollection rather than a diagnosis, but it's the kind of detail people don't invent.
Everything else you'll find online — Madison, Chester Arthur, Washington failing to distinguish crimson from scarlet, Lincoln calling lavender blue — comes from content-farm pages that cite nothing checkable, and I'd treat it as noise. Lincoln's documented eye problem was strabismus (eyes do not line up), not color vision.
As for vice presidents: I can't find a single one with any credible claim. That's a bit surprising statistically — roughly 8% of men of Northern European descent are red-green deficient, and with fifty-odd vice presidents you'd expect several. The explanation is almost certainly that nobody was looking. Color blindness is invisible unless someone asks, the Ishihara test only dates to 1917, and a nineteenth-century politician had no occasion to disclose it. Clinton's case surfaced only because television lighting forced the issue.
BILL: They left out Hubert Humphrey whose colorblindness was well known and well documented. It was why he was unable to serve in WW II. His not serving was used against him early in his career.
LANCE: I asked Claude about that and here is what it said:
Good catch — that's the vice-presidential case I missed, and it's better documented than anything on the presidential side. Humphrey tried to enlist after Pearl Harbor and was turned down three times, twice by the Navy and once by the Army, on account of color blindness along with other physical problems. Biographical accounts add that he also had a double hernia and lung scarring, possibly from unpasteurized milk.
This one holds up because it left a paper trail: military induction physicals in the 1940s included Ishihara-style screening, so his condition was formally tested and recorded rather than inferred from someone's recollection about suits or golf balls. It also mattered politically — the failed enlistment dogged him, since he was a young man of draft age who spent the war teaching and working for the WPA, and opponents used it against him in his early Minneapolis campaigns.
So the honest tally is one vice president (Humphrey), one president (Clinton), plus Eisenhower as a plausible unconfirmed case. Which, incidentally, undercuts the "nobody was looking" explanation I gave a moment ago — the real filter was that the condition surfaces only when an institution forces a test, and the military did that for exactly the cohort of politicians who came of age around the world wars.
BILL: SO, why did it get it wrong? This is not something obscure which requires guesswork like speculating on whether Washington was colorblind (I can't tell scarlet from Crimson either).
LANCE: Google AI did even worse.
BILL-AI can solve Erdős problems, but it can't track down that Hubert Humphrey was colorblind. Why is that?
LANCE- Claude was just too lazy to check over the military records of every vice-president. AI are people too!

wait this is interesting. by the time i google AI-ed this phrase, i get the following response: The most notable U.S. Vice President known to be color blind was Hubert Humphrey, who served as the 38th Vice President under President Lyndon B. Johnson from 1965 to 1969.
ReplyDeletefollowing this, I get also The Great Billie Clinton on the list, and Marcopolous Rubio along with Bobbie (Banana) Dole.
So what is happening here? Google AI is stalking complexity blog? and improving in real-time? I doubt this is done automatically but a team is manually overlooking these short comings.
Google does frequently and automatically index pages, and this blog is indexed at least once a day, the domain owners can check these by going to the Google search admin page.
DeleteFor me (in Germany) Google AI mentions Humphrey and specifically links this blog post as a source.
DeleteThere are three different things here:
ReplyDelete1. the model: Google seems to have switched its Search AI mode from more capable more expensive Pro models to less capable much cheaper Flash and Flash Lite models. This has made its quality drop significantly. You can think of it as a bait and switch, or as Cory Doctorow calls it Enshitification.
2. Model itself doesn't know this fact. Even if it appeared in its training data, it is low frequency low value, so it hasn't memorized it.
3. AI Search performs a RAG using Google's Knowledge Graph and Search Index. That does not read all the pages, it just runs a semantic query on both and gets a handful of resulting pages. Knowledge Graph likely didn't have the information about colorblindness linked to the VPs so didn't return anything and Search One's the semantic embedding was not close enough to any source with the relevant information.
This all means that you should trust what you read on AI Mode of Google much less.
If you want to do an indepth retrieval of information, you should use Pro/Thinking with Deep Research mode.
It is partly an issue of economics. For a demonstrative math case they can spend massive amount of compute and time, but for a typical search query it is too expensive to even use their Pro model.
Reports state that they are facing computing power shortage, and trying to save capacity for their enterprise customers and for renting to Anthropic.
The recent Google Deepmind leadership exodus is reported to be partly because Google different want to spend resources on science projects anymore.
@10:19: thanks. you seem to have done your homework on this.
ReplyDeleteI also know how Google will excuse this, they will tell u something like unit cost with latency requirements, using Google Pro users will need to wait 5-10 seconds for a response but with Flash, they instantly get it in flash and that's why we switched to Flash, and of course you are right it happens to be cheaper.
regarding the RAG process, Google will politely tell you something like: the writer completely misunderstands and undersells Google's retrieval infrastructure. Google does not exclusively rely on dense vector embeddings (semantic search) to retrieve documents. It uses a highly complex hybrid of lexical search (keyword matching), graph-based ranking (PageRank), and neural/semantic search.
Regarding compute shortage: Google will tell you, Google Search and Google Cloud operate distinct infrastructure pools.
Regarding leadership change at Google Deepmind, they will tell you they sponsored essentially a project that won the Nobel Prize, blah blah blah, but I agree with anon, Google deteriorated over the last 4 years to a point of no return. It's been a clusterfk dealing with the bureaucracy and their unintelligible salesfolks .... Given that even Jeff Dean left, tells you the rest of the story. Ironically, he is the guy that flew over to validate the Deepmind deal, so he opened the door for Hassabis to enter