A.I. experts may be unable to create an A.I. alone.  But they may get significant help from other sciences.  A key idea behind the Singularity is that different sciences will combine synergistically.  I hope that advances in psychology and neurology will lead to advances in A.I.

The futurologist Raymond Kurzweil thinks that an A.I. will only come when we know more about the human brain, and how to model it in a computer.

I think that the biggest contribution to A.I. will be made by psychologists.  If they can tell us how children learn, and how they can acquire language, this will be very helpful.

A.I. has not advanced.

Computer hardware has advanced rapidly.  It has advanced step by step, advancing a little each year.  Computer hardware is far better than it was a few years ago.  But this step_wise advance has not happened with A.I. software.

We have not been walking step by step towards a really intelligent computer.  We are barely closer to an A.I. than we were 20 years ago.  Hardware has advanced, but creating A.I. is mostly a software problem.

The field of A.I. is fragmented, divided into different sub_fields.  There have been some successes in sub_fields of A.I., such as chess playing, reading printed text (OCR), and Expert Systems.  Despite these successes, there has been no overall progress.

Why has A.I. not advanced?

Programmers working in different sub_fields have written A.I. programs of quite different kinds.  Programs from one sub_field are unrelated to programs in another sub_field.  Therefore anything learned in writing one type of program is not used when writing another.

What can be done about it?

In order for a new A.I. program to build upon a previous program, these programs would need a single way of representing data.  This way programs which work differently could co_operate, each doing some task with the same set of data.  I foresee a time when English is used as a common way to store data.  Facts learned by one program could be passed to another program in the form of plain English sentences.

The range of grammatical forms that humans use every day is huge.  It would be a Herculean task to write a program that could use such a wide range of grammatical forms, if each form has to be programmed in by hand.  For this reason, the first program that can use everyday English would have to learn English, just as a child does.

In addition to learning grammar, we use complex logic to understand everyday English.  A program that could make full use of English would have to acquire a lot of human_type logic, not only grammar.

But we don't yet know in detail how children learn language.  It's not exactly clear how children learn.  It seems to be statistically initially, but then at a certain point you will see it is not just statistical.  They are reasoning.  It's remarkable.

To make a computer program that can learn in the same way as children learn is a very complex task.  To achieve this will require progress in different fields, as listed below.
I see a need for computer programs that imitate children in all of the following behaviors :

• learning,
• reasoning,
• acquiring language, and
• actually using language.

Achieving all this will be extremely difficult!
First it will take a lot of work by psychologists to discover how children do these things.  This is not a job that computer programmers can do.  

Programs to do each of these things can be written separately.  Eventually, techniques used in writing all of these programs will be used in creating one single program which can do all of these tasks.  At that point we will have a program which can learn new words and new grammatical structures, until its grasp of English is as great as its creators.  This would be an A.I.   At this point a superhumanly_intelligent A.I. will probably be just a few steps further ahead.

--- End. ---