Wednesday, 19 December 2007

Natural language ambiguity

Following on from the pronoun ambiguity post, here's a reply I had:

Robert wrote:
Let's assume also that you entered in the conversation only at the end "John gave the man his hat"....many people do it, it's called eavesdropping, human nature to be curious. The question is, do you just walk away only knowing the resolution to a story without the climax? Do you pick up a book only to read the last page?

Just like humans, computers would need to ask questions to further understand situations of ambiguity.

This is true .. an AI in conversation could ask you questions to help resolve ambiguity.

But this seems to me an interesting difference between classical computer 'programs' and human thought. There are a lot of situations where you simply cannot ask these questions and resolve the ambiguity (or at least, not immediately).

A computer traditionally tends to be programmed with a definite set of instructions, where there is no ambiguity.

e.g. Instead of saying

'The man walked down the street'

a 'computer-type' sentence would be:

Man 249223 walked down street 394443.

The computer program would want to know precisely which man it was and which street. Was it the same man who did such and such a week earlier?

In natural language on the other hand, the whole system is geared around being able to deal with ambiguity .. delayed resolution.

An example would be a 'who done it' murder book. The whole book could be written, to be resolved in the final pages of the final chapter, where we find out who the killer was. And it is in this resolution we would fit together all the pieces of the story, like a jigsaw, and it all makes sense.

The same could be true of a scientist researching a problem. They investigate different aspects of the problem, and find little bits of truth and hypotheses that 'work', without necessarily understanding why. Then at a later point, there may come a time where a key fact is learned or reasoned, that suddenly resolves the previous ambiguities ... a 'eureka' moment if you will.

Pronoun ambiguity

I'm just rewriting the pronoun code for my brain. Natural languages are great for their ambiguity... consider this sentence in english:

"John gave the man his hat."

Now, whose hat was it? It could have been John's hat, OR the man's hat!!! Us humans don't know, so how is a computer supposed to know lol.

Now consider the sentence:

"Elizabeth gave the man his hat."

We now KNOW that it is the man's hat, because the gender resolves the ambiguity. This incidently is probably why languages have genders, to resolve ambiguity, yet keep things concise. In english we don't tend to use genders much, but in french they use genders for all objects, and that helps resolve ambiguity, at the cost of making the language a bit more complex. In latin, they have THREE genders, masculine, feminine and neuter!!

Which is better? Discuss lol!!

What is Egor?

No doubt you are wondering what the title of my blog means.

Well for the past year or so I have been dabbling in the field of building an artificial brain, somewhat akin to the human brain. I prefer to call it a brain than an AI, simply because that term is so overused, it's not cool. I prefer to be thought of as a mad scientist, and building a brain sounds much cooler than writing an AI.


Egor is a computer program I am working on which has an equivalent of a brain database, which can read and interpret natural language (english), answer questions, do reasoning and generally be quite a fun and helpful chap. I won't provide full information here about how Egor works, I'll save that for a later book or similar (if I am continually successful). I would rather fully test my ideas before revealing them, rather than showing an unfinished work.

Of course I am sure there are umpteen thousand different people working on their own little 'egors', but this one's mine, so that's why you are reading about it here lol. The fact that there are so many people working in this field, for several decades, without any (in my mind) really significant developments .. gives you an idea that the answer to the 'problem' of AI is not altogether obvious. Although it may be obvious once the answer has been found .. many things seem obvious in retrospect. Kind of like evolution .. so simple, yet it took ages for someone to work it out, and once it was worked out, it turned the world upside down.

The reason Egor is being designed to read english is simply because that is my own language. It has occurred to me at certain points that I could quite possibly build the whole system as multilingual. However, that is an addition that could be made at a later date, and the slight added complexity of this would probably exponentially add to the time of initial development. In short I'm quite happy to leave multilingual development to a later stage, or to other people. KISS. Keep it simple, stupid lol.

And why the name Egor? Actually I was very influenced by the movie 'Team America' and their advanced AI machine, called 'Intelligence'. The idea of a bumbling, slightly retarded computer is great and I feel is something to aim for.. I have even modelled Egor's voice on 'Intelligence'.


The name itself actually comes from the Frankenstein type movies. The mad professor's hunchbacked assistant is always called 'Egor', there's a master servant thing going on there. And so that is what I use, for Egor, the pitiful but obedient slightly cute servant. I wanted to be able to say 'BAD EGOR!!' when he gets things wrong, and 'good egor!' when he gets things right.

Introduction

Well as it nears the end of 2007, I find myself deciding to start a blog on my 'intellectual' type thoughts.

I have found myself in the past year writing posts quite often on forums, where I'm sure they are rapidly lost from the world, so I thought from now on it would be a good idea to keep them together in a series for anyone who is interested in the same areas.

It should be said at some point I would like to write a book on some of the ideas presented here, however, that will have to wait until I have investigated them further.

My principal 'intellectual' interests I have been pursuing recently, and which will probably form most of this blog, are:

AI
Evolution and biology
Programming

If all this sounds very nerdy, then I apologize. I assure you all that I do have many 'real life' interests outside of these, it's just that these are some of the most interesting to talk about, I'm sure real life dramas are much more of interest to my close friends than to the wider world.