How does Cartica.AI change the world of Artificial Intelligence?

Traditional AI has significant challenges and limitations.

As an example let us look at computer vision for automotive safety or autonomy. These systems need to detect objects like a traffic sign, a car or a pedestrian. In order to learn about objects they have to be taught about how those look. One big problem is that there are many so called Edge Cases in which objects look slightly different from the hundreds / thousands which have been taught so far. So if traffic signs are partly covered, dirty or bended, trucks have a odd load, pedestrians are particularly small or tall or they wear different clothing, that is a problem for traditional neural networks.

As a solution more cases have to been taught as an attempt to cover mostly all possible edge cases. Traditional deep learning requires terabytes of data sampled from real life situations and literally thousands of people to tag objects that need to be identified. Entire companies have been created in India and other low cost locations to do just that.

Also in order to absorb that much teaching, neural networks need more processing capability and consume more energy. This means larger and more expensive chips which consume more energy (which is increasingly an issue for electric cars) and need additional cooling. The Tesla system for example is water cooled today.

Another problem for such systems is spoofing. With specifically designed patterns deep neural networks can be fooled and see something false (e.g. a traffic sign where there is none) or do not see something (e.g. oversee a pedestrian).

@Cartica uses a much different approach to deep learning. Which is much closer to how the human brain works. Instead of feeding every pixel into a deep neural network @Cartica identifies recurring patterns. We call them signatures.

Just like a new born looks into the world and does initially not understand anything around. He will start to create signatures of repeating patterns like common shapes, textures, colors, and co-concurrences of such. So when you show him one water bottle he can associate certain robust signatures with it and will be able to identify pretty much any bottle from then on.

We generate signatures, which are used to represent different objects in the world and do not require deep networks.

This is exactly what Cartica AI does. We call it self learning AI. We generate signatures and feed them into a much and simpler neural network. As a consequence we generate much better identification also in edge cases and need much less teaching because we replace it with ongoing self teaching.

Cartica's software can run on much smaller computer chips and can reduce the power consumption by a factor of 10. Consequently we reduce a lot of the cost of the system.

This is specifically important in automotive systems for the mass market such as NCAP systems for emergency braking and traffic sign recognition.

Spoofing is also much less of an issue since we do not work on pixel level but on signatures. There is no direct link between the pixels and the neural network any more.

All the above is the reason why I am very exited about Cartica and why I became an investor and board member.