13 February 2023

Context matters as much as data

In the novel Passion Play, author Sean Stewart has a rant about Sherlock Holmes. 

It’s one of Doyle’s famous scenes where Holmes says, without prompting, what Watson was thinking about the war he was in. I think it might have been this, the opening to “The Dancing Men”:

 “So, Watson,” said he, suddenly, “you do not propose to invest in South African securities?”
I gave a start of astonishment. Accustomed as I was to Holmes’s curious faculties, this sudden intrusion into my most intimate thoughts was utterly inexplicable.
“How on earth do you know that?” I asked.

 Holmes goes on to explain how he arrived at the conclusion. But while Holmes is famous for his “deduction,” what Holmes did was not deduction.

“Very likely not; but I can quickly show you a close connection. Here are the missing links of the very simple chain: 1. You had chalk between your left finger and thumb when you returned from the club last night. 2. You put chalk there when you play billiards, to steady the cue. 3. You never play billiards except with Thurston. 4. You told me, four weeks ago, that Thurston had an option on some South African property which would expire in a month, and which he desired you to share with him. 5. Your check book is locked in my drawer, and you have not asked for the key. 6. You do not propose to invest your money in this manner.”
“How absurdly simple!” I cried.

You couldn’t have come to the same conclusion as Holmes if you didn’t have the same detailed knowledge that Holmes had about Watson. You had to know practically everything about Watson. You had to know the context for all of these little details that Holmes observed.

It’s been reported that human brain size is shrinking. I think I’ve even said this in a Quora answer or two.

The Guardian mentions this claim (as well as the “sea squirt eats its own brain” myth) in a new article about language using artificial intelligence like ChatGPT.

Alain Gorely tracked the origin of this claim back to one particular paper. He found reasons to be skeptical of the “brains have shrunk in the last few thousand years.

Suzanna Herculano-Houzel looked at this sleuthing and wrote:

ALWAYS look at the data. Always. The data are one thing; the interpretation of the data quite another. Robust findings are the ones that already appear with basic statistics, not that require the complex analyses.

The funny thing is... the data are not at issue here. That is, the values that are used in the analysis are not in any way wrong or under dispute. It’s not, “Someone moved a decimal place, and that screws up the average.” 

What Gorely does is put the data in context. In particular, he asks, “How were the data collected? How consistent was the collection methods? Are the numbers in line with other numbers?”

“Show me the data” isn’t a definitive mic drop. And when people “look at the data,” they are usually doing something much more complicated.

No comments: