Casual mistakes from the world of AI

WRITTEN

TOPICS

TAGS

“AI”, such as it is, has been around in some form for a while now. The only thing truly new about AI is how it has been widely touted as the next big thing – people who are otherwise not technically savvy have been spoonfed the catchphrase “AI” without realizing that anything that comes from a computer and involves logic that learns from iterative, repeated usage, is “AI”. A good example of a AI feature set that was unceremoniously integrated into a lot of our daily technology digest is the feature in Google Photos that attempts to categorize the subjects of your photos.

This feature is pretty amazing, if you consider how much effort it might take an average human to categorize and catalogue thousands of photos of various locations, objects, and different kinds of cameras. By and large, the logic gets it right, but occasionally it fails. These near-misses are kind of hilarious at times, but it shows how computers “think” – just knowing what something doesn’t know is more powerful piece of knowledge than just getting it right. Here are some examples of what Google Photos’ AI algorythm has determined about my pictures. Some of them are way off, some of them are very close, but still not correct.

Here are some categorizations that were not exactly right:

This is a photo of Interstate 84 in Portland, Oregon, but Google thinks this is a “race track”. It makes sense – the staircase on the right, and the angle of the road does make it superficially resemble a race track.

This photo of me was taken in 2006 in Lower Manhattan, but Google put this in the category of “Sailboats”. The large vessels docked below on the piers are boats, but certainly not sailboats, but I can see why the AI thought they might be. It didn’t bother to associate this with “Skylines” or “Bridges”, even though those features are much more prominent.

It categorized this screenshot of Beverly Crusher and Deanna Troi working out on the Holodeck as “Dancing” which is not entirely inaccurate, but it wanted to associate Beverly Crusher as Cloris Leachman (see left).

It also thinks this album cover is “Dancing”, but they’re just riding bicycles and waving. Luv was a pretty silly, throwaway disco pop trio from the Netherlands. The bicycles are on point, but this photo didn’t get associated with “Bicycles” grouping in Google Photos.

This picture of me at age 7 is a “Factory”.

This still shot of the movie “Straight Talk” with Dolly Parton was filed under “Cooking”, but that’s just a sign that Google Lens can usually automatically glean text from a number of types of images. But she’s not actually cooking… or cookin’.

By and large, their algorythm and software logic gets most of the images correct. The ones that they don’t, however, are full of context and nuance, which is a sign that you still have to have human involvement at the end of the day.

More to read

I guess he’d rather be in Colorado

gxnknkr

Sledgehammer: A self-help motivator in the form of a pop song

sledgehammer

Memory lane rabbit hole: photos from my childhood

matt rutledge in victoria bc

Google Sheets has massively improved

Number of Rutledges in US States

One does not need photographic evidence

Paying the man 89 cents

You know what rings my bell? That sound on Anita Ward’s disco classic!

ring my bell