Reproducibility: Academics and Smart Software Share a Quirk

January 15, 2023

I can understand why a human fakes data in a journal article or a grant document. Tenure and government money perhaps. I think I understand why smart software exhibits this same flaw. Humans put their thumbs (intentionally or inadvertently) put their thumbs on the button setting thresholds and computational sequences.

The key point is, “Which flaw producer is cheaper and faster: Human or code?” My hunch is that smart software wins because in the long run it cannot sue for discrimination, take vacations, and play table tennis at work. The downstream consequence may be that some humans get sicker or die. Let’s ask a hypothetical smart software engineer this question, “Do you care if your model and system causes harm?” I theorize that at least one of the software engineer wizards I know would say, “Not my problem.” The other would say, “Call 1-8-0-0-Y-O-U-W-I-S-H and file a complaint.”

Wowza.

“The Reproducibility Issues That Haunt Health-Care AI” states:

a data scientist at Harvard Medical School in Boston, Massachusetts, acquired the ten best-performing algorithms and challenged them on a subset of the data used in the original competition. On these data, the algorithms topped out at 60–70% accuracy, Yu says. In some cases, they were effectively coin tosses¹. “Almost all of these award-winning models failed miserably,” he [Kun-Hsing Yu, Harvard] says. “That was kind of surprising to us.”

Wowza wowza.

Will smart software get better? Sure. More data. More better. Think of the start ups. Think of the upsides. Think positively.

I want to point out that smart software may raise an interesting issue: Are flaws inherent because of the humans who created the models and selected the data? Or, are the flaws inherent in the algorithmic procedures buried deep in the smart software?

A palpable desire exists and hopes to find and implement a technology that creates jobs, rejuices some venture activities, and allows the questionable idea that technology to solve problems and does not create new ones.

What’s the quirk humans and smart software share? Being wrong.

Stephen E Arnold, January 15, 2023

Written by Stephen E. Arnold · Filed Under AI, News, Statistics

Comments

Comments are closed.

Search the site
Subscribe to Beyond Search
Feature archive
News archive

Stephen E. Arnold monitors search, content processing, text mining and related topics from his high-tech nerve center in rural Kentucky. He tries to winnow the goose feathers from the giblets. He works with colleagues worldwide to make this Web log useful to those who want to go "beyond search". Contact him at sa [at] arnoldit.com. His Web site with additional information about search is arnoldit.com.