The Washington Post reported below that Google has patented a way for self-driving cars to understand the hand signals it sees from cyclists.
It uses a combination of the sensors on the car, LIDAR, RADAR and the camera to get the data and then processes it to identify the cyclist, the cyclist’s arms and legs and then if the arms are signalling.
In itself, this is a very interesting algorithm and it will be interesting to see how accurate it is in a real-world scenario. I can see it having some difficulty in Amsterdam though.
There is though a wider issue here with respect to patenting an AI algorithm. Suppose the machine learns about recognising cyclists through some method of supervised learning. Having come up with the model of the world, is the patent for the learning machine or the mathematical model that it has learned?
Patenting a learning machine would seem to be rather broad, while the mathematical model is not patentable.
Nevertheless, it’s reasonable to expect to see a rash of patent applications with variations on the title “A method to do X with an Artificial Intelligence”.
In the meantime, let’s hope the Google Autonomous Car doesn’t knock down any cyclists.
On Machine Intelligence 