I recently took a look at the original patent application and realized that Google has referenced it in ten of their patents. The application has also been referenced by Microsoft, Apple, IBM, Ebay, Yahoo!, Kodak, BT, Fujitsu, Nokia, Myspace, and many more. I'm not sure how much this indicates adoption of some of these ideas, but it got me thinking about Affinax in the context of the state of recommender systems and AI today. I must admit that even after some years of working on other things I would still be very interested in collaborating with those in the field who might want to, or are currently trying to, develop and implement these ideas as revolutionary and disruptive products. So here's my attempt to summarize Affinax and some of its potential benefits and implications in hopes of starting some new conversations about it.
A good way to understand Affinax is in terms of the theory of mind (ToM). ToM is central to human social behavior and may be the key factor underlying our distinctive cognitive abilities. For example, ToM may have provided the foundation for the emergence of music and language in human evolution. Essentially, it is the capacity to explain and predict people’s observable actions in terms of unobservable mental states such as beliefs and desires. Our brains may be constructing internal representations of other people, running simulations and making predictions. I've referred to this concept in the past as a good friend who knows you well, and as your identity proxy. Affinax mirrors ToM except that it considers all things in the world (people, groups, products, other objects, ideas, etc.) to be social agents because all things in the world form relationships with humans. The mental attitudes of (or for) these agents can be reduced to mental representations (i.e. mental models) in the form of human ontology (traits, demographic data, interests, beliefs, affiliations, history, desires, genetics, etc.).
In Affinax, the representation of an agent is constructed by the agent's relative affinities for other agents whose representations' ontologies the first agent's representation assumes in proportion to the level of affinity. That first agent's representation then contributes its ontology to other agents' representations according to their affinity for it, and so on and so on. A virtuous cycle is established within the increasingly dense (and accurate) matrix of representations whose emergent property can be thought of as a universal social engine. This engine, similar to ToM in humans, is able to make immediate predictions of interest, behavior, risk, etc. that would disrupt multiple industries including marketing, recreation, career, social, medicine, insurance, etc.
The patent application describes a very simple system and example whose implications may be missed at first glance. But the expanded system evolves (develops iteratively) into a dense matrix of ontology and affinity that would represent everything in human context. All things become predictive social behavioral representations, as per ToM, which enables the development of products that benefit from precise targeting and behavioral prediction.
Other implications of this technology are reflected in the intentionally sensational title of this post. For your consideration:
- Just as consciousness is an emergent property of the human brain, artificial consciousness and intelligence may result from similar emergent properties of a complex, socially informed, non-biological system.
- In humans, the machinery that computes information about other people’s awareness may be the same machinery that computes information about our own awareness. ToM, which Affinax mirrors, may have been a key factor in the development of self-consciousness.
- The way in which a thing relates to humanity may be considered the ultimate meaning of that thing, and therefore constitutes a reasonable building block for consciousness or sentience in a system (see below) processing the meanings of many such things, such as Affinax.
- Affinax is a learning prediction engine, and prediction is a computational correlate of consciousness.
- The subjective internal experience that we call consciousness may simply be the flood of ontological data associated with whatever is the focus of awareness or attention at any particular moment. Your ontological data for the color blue, based in part on the history of your personal experiences with blue, i.e.the associations, significance, emotion or "meaning" you attach to it (see above), are different from mine and that's why it is said that we see them differently. Affinax would be able to provide similar meaning to any non-biological ToM system focusing on a particular object or idea and would therefore provide it with the same internal experience of consciousness that we humans feel.
No comments:
Post a Comment