ReverseMatrix

Reverse Matrix is a synthetic data offering, mainly aimed at law enforcement, and people building tools for their use.

The police are starting to use AI/ML (Artificial Intelligence and Machine Learning) tools to both improve their efficiency (getting AI to replace non-core parts of their work, so officers can do the bits that need brains, emotional intelligence, and physical presence), and also to manage large data analysis tasks to ‘get ahead of’ crime and terrorist actions.

If you want to develop and refine data analytics and ML tools for use across a population the size of a region or city, you need lots of known data to train on.

There are currently two main ways of getting this, each with its own problems:

Collect real data on a real city’s population. This is both impossible to get hold of from a privacy/legal perspective, and also isn’t really ‘known’ data, as it doesn’t include the inner motivations of people that lead to crimes being committed. Current crime data is based on reported crime, which missing a possibly large amount of unreported crime, which is still crime an ML tool should be considering.
Generate synthetic data based on the statistics of the real data. This might solve the proportionality issue, if the synthetic data isn’t too closely following the real data, but it lacks the complex relationship-based behaviour that real people exhibit: your social networking list may include relatives, recent friends, friends you went to school with, work colleagues, etc.. These patterns behind the data are what real investigations use, to refine searches and prioritise lines of inquiry.

Reverse Matrix is taking a third way: building a city-sized model, populated by autonomous software agents with histories, jobs, families, and realistic social networks.

It’s entirely fictional. We’ve made it all up from scratch, and it doesn’t use real people or places as templates, so there’s no privacy or Personal Data issues.

It has complex structure, so you can pull investigative threads (phone records, electoral roll, driving licence database, car registrations) to work back from a crime, to suspect, to an arrest. When a road has to be closed for maintenance/repair, modelled people have to find a new way to work, or to get their children to school.

It has known crime and known crime motivations, so it’s possible to train and test policing tools such as ‘predictive policing’ and check the fairness and effectiveness of various profiling approaches. These can optimise for actual crime in the model, rather than just reported incidents, before being transferred out to the real world.

Sounds great! Why isn’t everyone doing this?

because it’s really hard work to do well. It’s like writing a 100 novels, and designing a complete city, and coding a modern computer game.

How can I find out more?

email John with a summary of who you are and what you’d like to do.

Why the name?

because John loves The Matrix story. In that film, the artificial reality is populated by enslaved human minds and controlled by the ‘machines’. In this model, the humans are in charge, and the people are code, although they don’t know that. Get it? Whatever — it was a nice short domain name… :-)