We are living in the world where mentioning AI scares most people right away – decades of primitive and silly fiction, depicting horrors of “robots taking over humanity” built yet another solid foundation for moral panic, whenever machine smartness is mentioned.
However, should we really be worried? And why?
Think about it this way – AI is evolving, and insanely complex (while at the same time insanely inefficient and fundamentally primitive) monsters of neural networks are becoming a thing of the past. And while it takes a while to get from perceptron to Google’s Deep Dream – we are definitely moving in the right direction.
Which direction? Opening the black box of AI.
Traditional classification and correlation engines were pretty amazing in finding similarities within large datasets, and organizing them by statistical relevance. But – all that magic was happening within the same, ridiculously complex, structure of countless “layers” of neural network, feeding each other.
Even building that kind of “learning” machine was quite a feat, think what would it take to somehow extract a snapshot of its virtual “mind” and make it somehow meaningful and understandable to humans.
Well, there is a classic, but perfectly working approach where, instead of stacking all these layers in one monolith, they may be separated, and interaction between them organized around very specific contracts – allowing to clearly and uniformly expose the interim output of these components of the “thinking machine”.
Resulting snapshots of the machine learning states would provide insights in classification, correlation, and other complex processes occurring within AI. They may be formatted as images, or videos, or text, or anything else useful for interaction with human senses. They may be sequenced and organized to explain each conclusion to which machine came, when “learning”.
It is almost too simple, but very functional, nevertheless!
Brain Mechanic experiments with patent-pending XAI (eXplainable AI) technologies, building functional prototypes since mid-2018, and so far results were very encouraging. To the extent of mapping AI “explanations” into our business process flows, and making output of that “talking AI” a native product of the systems we create.
Want to learn more? Join our team! We are blessed with many talents, but always open to meet new ones. Also, the way we are approaching it, you should be able to benefit from using our technologies to your own purposes, and under your own efforts recognition – while making all of us benefit from your genius.