I talked to a customer recently who had studied AI at university, who said that he was envious of the way that here at Scientio we can just follow research where it takes us.

This isn't quite true - we still have to earn a living - but our ethos in general is to try to dicover first and sell second.

 

One of the great beauties of computer science is that - assuming you engineer your experiments properly - you can go very rapidly from new discovery to product.  It's this area that University research departments often fall down. They've got no lack of bright people, but frequently their solutions are great computer science and bad engineering.

 

One such new research idea that hit us is the possibilities inherent in the new questionnaire interface. This, to recap, since the product pages aren't online at the time of writing, is a new interface to our XMLMiner web service, that permits you to supply data to a rule set or collection of rule sets via a question and answer interface, rather than our existing web service.

The questionnaire interface is set up as a small web page designed to be hosted in a remote website in a frame, that talks directly to our XML Miner back end. You can pass parameters to select the processing map it uses, and the number of questions to display at once.  It's designed for situations where some complicated set of considerations, expressed as rules must be satisfied in order to get something. This might be entitlement - to receive benefits, screen for jobs etc. or to check compliance with laws or company policy.

 

This is what it looks like:

 

 

The Questionnaire interface takes advantage of the fact that Metarule inputs are closely typed to decide whether to display a dropdown or an edit box for each input.

 

Also, because in a typical application there may be hundreds of inputs, we've used some nifty AI techniques based on analysis of the rule sets to work out which are the most important questions - and thus inputs - to ask first, and as more data is entered to not ask for inputs that have become irrelevant.

 

Now i've said that we can control the number of questions asked at any one time. If we limit this to one, then we have, in effect a conversation.

 

This is much like the kind of conversation you might expect when asking any kind of professional for advice - a doctor. lawyer, trouble shooting technician, call centre employee, etc. 

 

Of course we already have a product that engages in conversations - ScientioBot.

 

Like all other ChatBots, ScientioBot does not record very much information as conversations progress, and cannot do very much at all in terms of offering advice. 

However combining these two ideas offers the chance of dramatically re-invigorating ChatBots.

Clearly a ChatBot that uses conventional technology to establish that a given customer is looking for information of a particular kind, and then calls up the approprite processing map to follow through the logic as a series of questions, is going to be very powerful, and very useful.

 

Also - and this is the key thing - it will be very simple to set up.  These technologies offer the chance to dramatically cut the cost and time required for chatbot customisation.  These costs and poor performance have been the barriers to chatbot takeup up to now.

 

I'd be very interested to talk to potential partners to take this further, We will develop this anyway, but it always speeds things up to have a partner we can try things out with.