About Watson

Posted on January 27, 2017


I work for IBM in the mobile and emerging tech spaces.

I have been asked a lot about what Watson does and what cognitive means.

I was lucky enough to hear an over view from the Watson CTO yesterday,  and this is derived from my (rather sketchy) notes,  and hopefully is just enough to get you thinking,  but not so much as to bore the pants off you!
Watson is large collection of capabilities that are brought together to make a system,  so you could answer it as all things to all men,  and you’d not be far off,  so here is a go at what Watson is used for,  and from that we can think about a cognitive solution.
This is important because so many new solutions are being centered or improved with AI systems.

First lets separate analytics from cognitive.
With analytics I can ingest a whole load of data and from that I can infer patterns, correlate data and generally report back on what has been ingested.  Watson however is able to start to make future decisions and predict outcomes, and look forwards.  A big change.

So how does it do this.

[it does not employ a load of imps in small cages,  so let’s debunk that one right off the bat ;-) ]

Simply put Watson has it’s own URL which in this case stands for Understand, Reason and Learn,  lets unpack that a little.

Watson has a range of facilities to read/watch or listen to inputs and from that derive their meaning.  Eg not just transform speech to text,  but then to understand what the text means.  eg this is a letter from a daughter to her mother, and the subject is about an upcoming visit, or this picture is of two men playing tennis, with 1 racquet, grass, net and ball.
To do this Watson will collect the data and translate it into something it can understand,  and then will standardise what it sees.
Once this has happened it will correlate the data against previously ingested data, improving the translation and standardisation.

Next it will Reason,  and it has three ways of doing this
– Inductive
– Deductive
– Abductive
I’ll let you google those if you’re not sure what they mean (yes I had to too)  however this is a key part of the process and should not be underestimated in it’s power to make the system intelligent.  That said it also requires to have had sufficient training material to become fully powerful.
Once Watson has Understood and Reasoned it will act, that may be a conversation with the user, it may be simply posting a suggestion based on the reasoning, and many other things,  but natural language is key,  and the ability to make a prediction with reasons why it made that prediction,  and scoring on how certain it is,  are the most famous types of Action we see.

Last very much not least is Learn,  as Watson is taught is able to check it’s predictions / actions against reality and then will use this to improve it’s thinking.  Thus as you train it,  so it gets better and better.  This is true for the micro,  that is for a particular action,  but also in the macro,  for the system as a whole.  These feedback loops into the reasoning, and learning phases. And … unlike us humans it does not suffer from hubris and prejudices.

So there you have it,  simply put URL,  a very powerful way to build up a system that starts to apply dynamic, smart intelligence,  and this is why so many tech companies are driving hard down this path.




Posted in: Uncategorized