Machine Written Software

A recent BBC special indicated (somewhat speculatively) that within a decade or so, most software would be written by machines. I'm not sure if this is really the right forum for this, it seems like an AI issue to me, but perhaps they were simply referring to natural language coding. Alternatively I suppose it may have been a reference to 'self programming' neural networks. Anyone care to offer other interpretations of this?

I ask primarily because I am about to go to university in a few months, and don't want to focus too heavily on programming if it means the skills will become obselete. I have a strong interest in AI, and would like recommendations on a course of study for entering this field.

This was probably a bit off topic, but from my brief time here I trust the opinions of those in these forums, so response would be appreciated,

unlyrn

 

Log in or Sign up to hide all adverts.

AI

I wouldn't worry to much about computers writing there own software. I am a graduate student in Medicinal Chemistry and a computer geek. I work for Sair linux and GNU certification. For my dissertation work I am evloving the way Neural Networks learn to help with some data coorelation. Pretty cool stuff. I took a graduate AI class last semester (How I met the person who offered my this job) and it is a fascinating field. It is also the most divided. You have two basic sides. 1) eveolutionary ( I fit in here ) also called ground up and 2) First Order Logic ( original view and some cool rejects going on)

Now most peole don't fit in either of these totally, but most adhere to one or the other basically.

Why not study AI?
Honestly If I didn't almost have my PhD. then I would probably try graduate school in CS with AI as my major area.

Also lets assume that computer programmers are obsolete in 50 years. that would make you around 70 right?

most computer people don't have to work when they are 70. You guy's have already made you fortune.

No honestly if it did happen then programs would evolve as well. It wouldn't be an overnight thing, and there would be alot of chances to move over to an area.

Besides it would scare the hell out of me if some AI began writing it's own code and we humans simply forgot or didn't care how to do it. Seems we would be setting ourselves up for alot of misfortune.

Now I do think that in 50 years we may have some sort of "AI" that will help with our programing, you know basically tell it what we are trying to accomplish and then have it help us out by debugging or ultimaly helping us with the design.

Anyhow I think that you are safe with computers for atlest this life time.

One more note. I have found that my love of computers has only helped in Medicinal Chemistry. I am sure that it would help in any discipline that you chose to study.

The world is yours, only place limits on your self.

 

Log in or Sign up to hide all adverts.

\777

I have been in private conversation with something you should know about for several months now and it is right here in this forum.
I think
http://www.sciforums.com/showthread.php?s=&threadid=3157
Will be of interest to you all.

Einsteins brain

 

Log in or Sign up to hide all adverts.

Re: AI

Very perceptive.

Although your choice of verb tense is inaccurate.
You should be writing that in the past tense, since it's been going on for some time now.

It's been done not only in the software field, but also in huge segments of the electrical and mechanical design fields.

It is almost impossible to get information on, or get bugs corrected in, a surprisingly large number of industrial devices simply because no human knows, or can find out, how they were designed.

Granted, that is partly due to the mobility of design personnel, but only partly.

 

Gnulinux,

I think what you intended was a creative AI intelligence re-writing code rather than relatively simplistic code enhancers and optimisers.

So yes that would be of some concern, but realistically it would be the fastest method to achieve higher intelligent AI, or more stable and bug free software. Humans tend to be imprecise, careless, become tired, and bored, these attributes cause bugs and errors. An AI that does not have these deficiencies could be very productive.

Cris

 

Bug-free software?
This is starting to get into the realm of the fantastic.

Or, it could produce some really intriguing bugs.

 

Gnulinux,

I think the precedent has already been set.

When I was at school (1950s - 1960s) there were no electronic calculators. I did all my calculations long-hand and used look-up tables for logs, sin, cosin, etc. Electronic calculators have replaced all of that. And how much do we trust these calculators? Would you seriously choose to perform the calculations long-hand because you feel the calculators could not be trusted? I don’t think so.

The same issues will soon apply to program code that is currently being written by error-prone humans. I am a software development manager and we spend a large portion of our time conducting formal code inspections. That is a team that would meet to inspect a piece of code written by someone else, the author is usually present. The approach is strict and all faults are noted and marked for action. Follow-up inspections are conducted to check the corrections. We find some 99% of all bugs this way. The items that are missed tend to be design issues and we are now extending the inspections processes to all program development phases. Once we have developed AI machines that can write code in a formal mechanistic manner then the need for error catching inspections will vanish. Extending the AI machines to include the design stages as well must inevitably follow.

Just like the trust we place in calculators we will soon extend that trust to AI machines for producing our software. The results should be error free code in the same way that our calculators are error free.

Cris

 

*Originally posted by Cris
And how much do we trust these calculators?*
Enough to see some kids having trouble with math because they hit the wrong button, or because the calculator is actually crapping out.

*formal code inspections.*
"Formal" code inspections are presumably better than "informal" code inspections, and much better than "slack-ass" code inspections.

Do you really think that you are the only group in the world checking code?

*The approach is strict and all faults are noted and marked for action.*
"Strict" is much, much better than "lackadaisical."
Your definition of "all faults" is everyone else's definition of "some faults."

I'm guessing your software is buggy just like everyone else's.

*We find some 99% of all bugs this way.*
Even though I would venture to say that your precentage is actually somewhat higher, that still leaves you writing buggy code just like everyone else.

*Once we have developed AI machines that can write code in a formal mechanistic manner then the need for error catching inspections will vanish.*
Well, why don't you follow such a formal, mechanistic manner in your own code-writing?
If you cannot write a functional specification for such a "formal, mechanistic manner" then you will not be able to program the AI machine, either.
Wake up and smell the coffee, Cris.

*The results should be error free code in the same way that our calculators are error free.*
That is what we are all afraid of.