Founder - https://twitter.com/ShreyaNallapati
Here is why it will utterly fail, apart from being founded by a female.
AI, in its current state, fairs decently with the task of prediction when the dimensionality of the input is low. AI generally take in labeled input data, and try to learn from it in order to produce a model or understanding of the world/problem. If the input data has too many dimensions (In other words, too many features that define it), the performance of AI algorithms deteriorate.
This is sometimes referred to as the Curse of Dimensionality. To combat the problems associated with high dimensional input data, you need an enormous amount of training data for your algorithm to learn from. This is where the problems arise with trying to predict the next mass shooter with AI.
How many variables/dimensions are associated with the input data for a mass shooter? There are millions of dimensions:
1.) Economic status
2.) Race
3.) Height
4.) Relationship status
5.) Childhood trauma
6.) Mental Illness
7.) Bullying
8.) Diseases
...
These are only some features associated with a particular shooter you would need to feed to your algorithm. There are simply way too many features that are relevant, and to get them would require almost perfect knowledge of their life.
The second issue is how many training examples you would need to give to your algorithm. You would need to feed your algorithm hundreds of thousands to millions of example data on shooters in the format I showed above. There haven't been this many mass shootings, nor do you have enough information on each of the shooters.
An AI approach suffers from highdimensional data (too many relevant pieces of info on each shooter), and not enough examples. You're better off trying to simulate the universe from the beginning with the correct initial conditions and picking out who goes on a mass shooting from that.
Anyone promising an AI driven approach to finding mass shooters is selling snake oil imo.
Last edited:
