Samurai Labs’ novel approach represents the third wave of AI (DARPA classification) that combines deep neural networks with reasoning based on language and experts’ knowledge. The recent scientific benchmark of five products available on the market shows that Samurai detects the most violence (84% recall), and -- what’s even more important -- raises false alarms four times less often than other solutions. This level of a precision (over 80%) is paradigm shift, allowing Samurai to be the first AI system for real-time online violence prevention. It means that Samurai prevents harm to those most vulnerable -- especially children -- rather than simply seek to ameliorate it afterwards.
The Third Wave of AI Explained
The Defense Advanced Research Projects Agency, or DARPA, has classified AI technology into three waves.
The first relied on hand-crafted knowledge. This AI could reason over narrowly defined problems, but it suffered from an almost total lack of learning capability.
The second wave involves machine and deep learning to learn from labeled examples. However, the second wave cannot reason and utilize experts’ knowledge.
Both waves have failed to solve the online violence detection problem. The first wave could not learn, whereas the second wave could not reason, passing through violence (low recall) and raising far too many false alarms (low precision).
Having foreseen the limitations of these technologies, we decided to take a different path.
Samurai represents the third wave. Combining deep learning with symbolic reasoning based on language and experts’ knowledge allowed us to push both precision and recall far beyond current limitations. Our AI consists of a number of precise contextual models that can learn based on the given knowledge, and abstract new knowledge to detect even the violence cases that were never seen before.