What A.I. sees of us

Aware that photos posted online are used to train A.I.s, ex-Google engineer Vishnu Mohandas has started his own end-to-end encrypted photo storage company.

But as few people are aware of how their photos are used, he decided to offer a photo analysis service based on artificial intelligence, with the obvious aim of promoting his own service; notwithstanding this motivation, the service does its job perfectly, and I invite you to try it out with sample photos or your own.

What A.I. sees of us

I took one of my own photos at an association's public event. The result is quite astonishing.

If an A.I. can describe the mood and pretext of a single photo, you can imagine what it can gather about us from all our photos, compiled over the years. Some people have already posted almost their entire life on the Internet, so a university or company using A.I. is already in a position to draw up a portrait of a candidate with great precision; one suspects that many human resources departments don't hesitate to use it.

From a series of photos, a trained A.I. is certainly able to tell with good accuracy our emotional state, our standard of living, our mentality, our state of health, our level of education, our job and, from a photo taken the day after an election, who we voted for. We can deduce from this that, in the near future, we will become increasingly reluctant to let ourselves be photographed for any purpose.

Photo analysis

Comments in brackets indicate the accuracy of the analysis.

"The picture shows a group of people gathered in what appears to be a restaurant.
[Correct]

In the foreground, several people are holding and looking at old photographs.
[Correct]

The background presents a lively atmosphere with other customers seated at tables, a large advertisement for "$1 chicken wings" in French, and a modern interior with large windows.
[The A.I. is able to read what is written in a photo. The resolution was too low to be able to read what's on the plates.
Localization is disabled on my device, the A.I. is not able to tell in which country this photo was taken].

Yellow and blue balloons are subtly placed in the upper left corner of the image, suggesting a possible festive event.
[Right]

The photo appears to have been taken on Sunday, November 17, 2024, around 11:26 AM EST, using an Apple iPhone SE.
[Right - Easy to tell from metadata. It's also possible to upload photos without metadata]

The group in the image is made up of people of different ages, races and ethnicities, all dressed in casual attire, suggesting a professional or semi-formal gathering.
[Right]

Their expressions seem focused and engaged, perhaps discussing the contents of old photos. An older woman with dark, curly hair wears a pink top, suggesting a softer, warmer personality.
[Questionable interpretation]

The background reveals details that are not immediately obvious: the ambience of the restaurant suggests an upscale establishment, given its interior decoration and apparent lack of extreme clutter.
[Inexact - Hotel restaurant - Popular restaurant chain, sports bar. It's not upscale - the A.I. didn't make the connection between "$1 chicken wings" and "upscale"].

The location and angle of the group's interaction is deliberate and orchestrated, rather than spontaneously captured, which might suggest the staging of a planned presentation or photo opportunity. The details of the photographs taken by the group seem carefully considered and arranged, indicating a thoughtful presentation.
[Right, but the thought process wasn't very long. ]"

Overall, the analysis is interesting, but the assessments are often debatable. The description highlights various details that a quick glance would fail to notice. This kind of analysis makes it easier to detect anomalies or incongruities such as, for example, one or two sad people in what looks like a party.

Several answers

We're not going to stop putting photos on the Internet, Instagram or any other storage site tomorrow morning - the advantages are too numerous. However, we can be a little uncomfortable with certain photos and a little more circumspect towards services associated with A.I. such as Meta, Google, Apple or Amazon.

Legislating to prohibit the use of our images to train A.I. remains illusory; firstly, because A.I. are already fairly well trained and advanced, and secondly, because there are so many ways of acquiring images, if only through the millions of cameras scattered all over the country, that this training will continue and be perfected.

The solutions lie rather in the use of A.I. analysis capabilities. Just as in the legal field, where illegally acquired evidence is inadmissible at trial, it is possible to prohibit the discriminatory use of A.I. in most fields, as soon as it clashes with the valid intentions of an individual or group.

So our future won't be entirely driven by artificial intelligence.

Try: They see your photos.