>> GOOGLE WORKING ON NEW AI ALGORITHMS - "THOUGHT VECTORS"

Sometimes it is nice to think out of the box - time for something different.

In a recent visit by my Melbourne colleagues and friends - the topic of artificial intelligence was hot; especially the movie Ex Machina; a science fiction film focusing on evaluating the human qualities of breath-taking female AI. On a slow news day; I came across an article where Google has announced it is working on a new type of algorithm called Thought Vectors - it was only natural to reflect how this could be of value for the Internet of Things.

I remember a statement that has been embedded in my mind since I got involved with technology:

Computers are stupid - they do not have a mind of their own; in fact; they do exactly what you tell them to do and are only as smart as the developer that programmed them.

Of course; when we throw in the fact that a computer can have a photographic memory and remember every small detail, or have the ability to evaluate every possible outcome before making a decision - it doesn't take long before the general public can be amazed and claim how smart and amazing technology can be. Which began the whole concept of Artificial Intelligence.

Unfortunately; AI is a deeply complex field of computer science where there is a tonne of effort to deal with reasoning, knowledge, planning, learning, natural language processing, perception and adaptation - something that is even a feat for the most powerful processing environments out on the market even today (one reason why Judgement Day hasn't happened yet).

The field of IoT is sub-standard compared to the processing power available on the desktop - mainly as power usage and energy release are factors that limit the longevity of such devices; so; it is obvious that complex algorithms have no place within the immediate vicinity of IoT. This is where the development of new algorithms must take place - potentially; trying to digitise everything in the world to a binary equivalent simply isn't feasible.

"Though Vectors" may provide some insight as a better way to approach complex algorithms or concepts with low powered devices - a good example is highlighted in the article where Professor Geoff Hinton makes the reference between Paris/France and Rome/Italy - the meaning changes by modifying one of the parameters.

“With machines at the moment, you get exactly what you wished for, the problem is we’re not very good at wishing for the right thing. When you look at humans, the recognition of individual words isn’t particularly impressive, the important bit is figuring out what the guy wants.”
source: Dr Hermann Hauser, article

It isn't the first time that there has been hints that current developers may be going about things the wrong way - I reported earlier the concept of learning from biology for security ; maybe it is time for a new thought approach to tackle the technology problems out there.