What happens when social media becomes part of the evidence
There was an interesting detail in the reporting from France this week that caught our attention at Thetafan.
French police investigating the looting and unrest are using videos posted on TikTok, Instagram, Discord and other social networks as a starting point for investigations. From there, officers can compare what they see online with footage from shops, public transport and a network of around 80,000 security cameras across the Paris region.
France 24 describes one investigation in which officers followed four suspects across different camera feeds as they moved through Paris, almost all the way home.
The French government has also said facial recognition can be used to compare images circulating on social media with police databases.
There are clearly important questions around privacy, proportionality and how technology like this should be governed. But there is another aspect of the story that we found particularly interesting.
It shows what starts to become possible when information that used to sit in completely separate places can be considered together.
One video rarely tells the whole story
A social media video on its own may tell an investigator very little. Add the time it was posted, where it was filmed, CCTV from the same area and footage showing what happened before and afterwards, and suddenly the picture becomes much clearer.
That principle applies well beyond policing.
We increasingly leave signals across almost everything we do online. We watch, comment, share, return, ignore, follow, talk to AI and come back to the same communities over time. Yet much of this behaviour is still measured as a collection of individual events.
A view is counted as a view. A comment is counted as a comment. A conversation becomes another session.
What is much harder to measure is the relationship forming underneath all of those interactions.
And that is the part that interests us.
From separate signals to a changing relationship
Thetafan is being developed to measure the relationships people form with content, communities and AI, looking at characteristics such as trust, influence, dependency, loyalty, manipulation and safety.
Our patent-pending continuation-in-part work takes this further by considering how information from multiple sources can contribute to those measurements over time.
That distinction matters because human relationships are not static.
Someone might interact with an AI system once and show very little attachment to it. They might then return repeatedly, begin trusting its recommendations, rely on it for particular decisions or gradually change the way they behave around it.
Looking at any one of those interactions in isolation could miss what is happening.
Looking across them begins to reveal the relationship.
And if those signals can be analysed as they arrive, the measurements themselves can start to change in real time, rather than simply describing what happened weeks or months earlier.
Identification and understanding are different problems
This is also where the French example and Thetafan separate.
The police are ultimately trying to answer who was involved and what happened.
Thetafan is interested in another question:
What kind of relationship is developing, and how is it changing?
That could matter increasingly as people spend more time with AI systems, creators, social platforms and digital communities.
We already have extraordinary amounts of behavioural data. The interesting challenge now may not be collecting even more of it, but understanding what all those separate signals tell us when they are viewed together.
The events in France offer one very visible example of that shift. We suspect it will not be the last.
The events in France offer one very visible example of that shift. We suspect it will not be the last.