Fido AI is a framework for automatic knowledge acquisition from any text data. Instead of working separately on narrow tasks, Fido AI is comprehensive and modular in terms of natural language understanding. It constantly learns new aspects of using human language that effectively form libraries of language models. Every language model is a high-level reusable realization related to a certain aspect of human communication, e.g. recognizing requests, recommendations or root causes. Every new language model is free to draw on the existing models. The more it knows, the easier it acquires the new knowledge. This way Fido AI constantly enriches its knowledge acquisition capabilities.

Fido AI represents a hybrid approach (the third wave of AI). It combines strong learning capabilities of statistical methods with strong reasoning capabilities of symbolic methods. The principles are very simple -- Fido AI uses a certain method when it works better than the others, without being dogmatic in any manner. If Fido AI needs to recognize abusive phrases, the statistical methods are usually better. If Fido AI needs to understand the context around these phrases to decide if the whole message represents hate speech or not, then the symbolic methods are used.




According to DARPA, there are three waves of AI. In the past, we have had symbolic AIs with handcrafted knowledge. Nowadays, we can see the flowering of the second wave -- statistical learning. We have already seen its great successes, and recently we become more and more aware of its limitations. Fido AI is a pioneer of the third wave of AI -- contextual adaptation.








There are many attempts to train machines how to mimic us humans by showing them a number of labeled examples. At Fido, we believe this is a dead end. Instead, we need to teach machines how to reason like we do.