The “Explore” algorithm and the Pictures of my 2023
A small rant about Flickr’s Explore and its pretension to rank pictures by interestingness, à propos my pictures of 2023
The image above is my most interesting picture of 2023 according to Flickr. I personally prefer a few others, but this is the one that ranked the highest according to Explore, Flickr’s proprietary algorithm that pre-dates the current times when the word algorithm is common parlance. By now it seems that folks are well aware that most of the things that we see online are picked by ethereal contraptions called algorithms. And so this one picture I took at the Jebel Jais mountains is the one that “Explore” ranked as most interesting instead of this macro shot of a bee on a yellow flower in Hamburg’s Planten un Blomen, for example, which I personally find a bit obvious but way more compelling.
And this brings me to my main quarrel with Flickr’s Explore. In principle, the idea that interestingness can be assessed algorithmically is equal parts fascinating and quixotic. My first question would be: What are the features of such a pretentious algorithm? In the case of Explore, the “features” are shrouded in some mystery, but Flickr is ready to explain that the internal algorithm considers factors like:
Interestingness: Signals of engagement such as comments, faves, and views.
Activity: Participation in the Flickr community, such as groups and trending tags.
And so by looking at Flickr’s own explanation, we go from -interestingness- to the thing that has become the main currency for internet content: Engagement. In other words, we have an algorithm that uses the popularity of a certain picture as a proxy for “interestingness”. That is the issue right there: As we learned from Iñarritu’s Birdman, popularity is not much more than the slutty little cousin of prestige. And prestige, I would add as a corollary, is not much more than the slutty little sister of interestingness. Prestige is pompous and self indulgent, but I can think of no better compliment than to call anything or anyone interesting.
This one of the Burj Al Arab with seagulls in the foreground is probably one of my favorites from 2023
With the important caveat above: The distinction between popularity and interestingness, we can leave all snobbery aside and take a kinder look at Explore. For starters, it seems to be a much more complex and feature-rich algorithm than its main competitor: 500px’s Pulse. In fact, 500px explains that their counterpart to Explore consists of “a score out of 100 points” that measures how -popular- a photo is “based on Views & Likes on your photo from the community”. In practice, what I have observed as an assiduous uploader to both communities is that Pulse tends to favour people with the most followers: The more followers you have, the likelier it is that you get more views and likes. This, in my experience, tends to result in the same highly “followed” photographers featured over and over on 500px’s “Popular” page. Needless to say, I don’t have the data to back this up but I have made many similar observations across many years, so take it as anecdotal evidence from a very good source.
On the other hand, the complexity of Flickr’s Explore tends to translate into what may be perceived as randomness or the very failure of the algorithm to distinguish between a good picture and an ostensibly bad one. It is not uncommon to see rather artless snapshots on the Explore daily selection of 500 most interesting pictures. That seems to be the price to pay for the same randomness that keeps the selection diverse in terms of content. While the photographic quality of the selection is questionable, it is hard to argue that Flickr’s Explore is constantly dominated by the same photographers or the same type of photography. Just don’t be surprised if your dullest pictures rank higher.
This one of EMG Champ in Venice Beach is probably one of my favourite portraits of all time and it was overlooked by both Flickr’s Explore and 500PX’s “Popular” page.
With the help of the Dopiaza set manager, I can create Flickr albums that rank all the pictures I took in a year by “Interestingness”. It’s a fun experiment, because the order of the albums in question is also affected by the same algorithmic randomness I mentioned above. Some pictures will rank higher depending on when I uploaded them, how many people interact with them and the metadata I attach to them. I like to try to influence the ranking on a regular basis to favour the pictures I like the most, but I don’t always succeed. The result is that some albums create a selection of hundreds of pictures that overlook my actual favorites. In fact, a weird hobby of mine is to observe the randomness in which the algorithm chooses pictures for a given album, inspecting how it moves the pictures up and down as I upload new ones. I can never fully predict which ones will be on top. This rant, I suppose, is me trying to rescue from total oblivion those that were overlooked and celebrating those that the algorithm got right.
And this is one of my favorite quotidian pictures of life around the Alster, in Hamburg. Explore and Pulse largely overlooked it as well.
The beautiful lines of the Westin Grand, in Berlin
For the curious, you can find the full album with the selection of the 200 most interesting pictures of 2023 (ranked by Flickr’s notion of interestingness) here.