Audrey - AI

A. L. I. C. E.

Releasing resources to the front line
Sir Peter Gershon

The Public DataWeb A.I. Vision

Currently we use a Verbot talking head, this is an Avatar, She is called Kirsty.

Kirsty can provide prompts, she can read out web pages and she can provide online help. She has many novel features but is far from any real 'intelligence'

Our research shows that such avatars on public access systems must be fully under the control of the user - for example the user on our systems can select Kirsty's help and switch her of at any point, she is under their control.

This is a very important aspect of usability in public services, other avatar kiosk systems put the character as the main facility, we believe this is a step to far and consequently puts of a very high percentage of users. 

Audry is a more programmable avatar and can be utilized into more complex services

Computer Artificial Intelligence will not be like human intelligence. We believe it will evolve from the approach we have so far been involved with, from the vantage point of the front interface, from usability, and putting all other computational activities 'underneath' a personalised 'conversation' with such an interface. 

This first stage of A. I. will be the computer interface(s) of the near future.

Public DataWeb build kiosk solutions, this means understanding the user needs; the public are diverse in their abilities to react to and use computers so our strategy is re-designing data access and re-purposing systems (websites) for ANY user.

We believe that kiosks in public places are only a part of the usability equation, and due to the regime and understandings of building 'intelligent' public access kiosks, we have gained unique insights and as a result we are in a position to move these methods onto other platforms.

We believe that Mobiles and Digital Interactive TV will be the methods to interact with all the people, and this will be achieved, by a conversation with the computer interface. 

Without experience in public access kiosk operation, the insights we have gained would be very hard to comprehend, and public kiosks of some type will probably always be required, even when most people carry a personal agent around with them in their mobile phone.

The beginnings of such a conversation are emerging here in our work; through interfaces that are touched, or clicked upon, then responding by performing tasks, like linking to a website. In public access systems these choices are predefined by the system builders. With open-ended web searches the results are more often than not inconclusive. In public access services like kiosks they are conclusive as a defined result is provided. Each operation results in reaching the 'correct' result. With a web search via Google, the results are not always conclusive, and may not match the aim of the user.

Intelligence in this case is understanding the aim of the user.

This can be achieved by fore knowledge of certain (known) needs and by a question and answer routine.

We can cheat by predefining this aim and produce systems that have the outward appearance of intelligence. Due to the fact that we can build structures around a known subject, that has a border around it so that it does not go on forever, it has known ends.

An expert system on any given subject is the start; like education, on one particular part of a subject, or in council services like filling out a benefit claim.
The Verbot Audry can speak, and ask questions, and can prompt options from the person seeking to make a benefits claim, or from a student studying a particular education subject. Such a dialogue looks impressive but is it A.I.?

To be honest NO.

But it will suffice in making a beginning, and will soon produce operations that greatly assist, council claimants and students, and people seeking all manner of services that can be assisted by known outcomes.

Obviously this type of approach needs at least one human operator with the knowledge required to answer all the possible options (an expert) but once this knowledge is 'scripted', or linked to back office processes (or certain websites) then it only needs to be located into the front end programme and out of the 'magicians hat' comes something new, and with high added value.

But is it A.I.
No
Will it ever be?
Maybe?
Are you a chatbot
YES!

This dialogue begins to take on the shape of a dialogue well known in computing, begun in the very early days of computing and today is progressing with A. L. I. C. E. to some extent, but without, real information its just a series of open ended questions and answers that eventually lead to nowhere.

Public DataWeb methods contain front end operations learned from operating public access kiosk systems.
We can link information by hypertext links onward from 'conversations embedded in text' then to jump to explanations or to more choices explaining about or giving more background about the subject in question, but is it A.I.

Well know.
Irritating isn't it?

But it does move things along, say for Council frontline services, reducing overheads, obtaining public inputs that save time and generate efficiencies. (i.e. Gershon type efficiencies for council services.)

These methods can be applied to public services and probably have a role in teaching as teaching assistants; but the human knowledge of the subject must be injected into the methods that produce the Verbot that reacts to the users.

Human babies most probably learn the meaning of things before learning the language and words that then describe the meaning; whilst computers can readily show (on screens) and even read out the words in any language, they have no 'knowledge' or understanding of the meanings, of these words.

This is the pivotal point in the A.I. universe.
Computers may have knowledge…. but understanding; that's entirely another issue.

However it is not a barrier to producing intelligent services from the systems that do, now operate.

In the book "The Age of Spiritual Machines" by Ray Kurzwiel 
in the chapter on Context and Knowledge he addresses this same problem. He also identifies the pivotal problem. He explains that for a piece of any single subject of knowledge there are major problems. His example is about birds. So the programmer has to tell the computer by some means that there is a thing called a bird. That birds have certain abilities, they fly, that is unless they are Penguins and Ostriches. Oh, yes, or if they are dead. You can see the problem. And also you will need to explain to the computer programme what dead means! 
Such iterative descriptions of meaning have a habit of going on, on, on, and on, interconnecting iteratively forever. Ray says that the process of building these structures is enormous a massive bottleneck to the development of A.I. then goes on to neatly side step it, ignoring it and presuming someone somewhere, out there will solve it by making the silicon, or the programme or some other invention to be able to solve it. You might as well believe in genies bringing three magical wishes.

There is no substitute.

This work must be done! 

We can only start in small ways.

Understanding a need in a given area or subject by an expert who can harvest the validated links to the subject, perhaps link too and intelligent web portal and fix it all together, and it will need software to check it for updates etc, but every so often a human editor would be required. A.I. theorists are waiting for a program to have innate capacity to do all this, without any human assistance. Web Services and update methods for Google point the way, and they have many staff doing similar work! The evolution of these processes will result in clusters of validated knowledge. This points the way.

So if we build knowledge structures of sufficient size and value, with interconnected meta data and with semantic linkages, and lots of descriptive key words and with explanatory short descriptive scripts, and if we put up a talking head, a verbot, that will intermittently react to the users inputs and converses with the computers user(s) whilst opening an internal hidden dialogue with the 'knowledge' then we can fool you, into the belief in our A.I.

Allan Turing believed that if all knowledge was put into a machine, and if the outputs were handled in some correct manner then a certain type of A.I. could be achieved.

But will it know of itself?

Will it understand what it gives out as knowledge?

Our answer is NO.