Re: An AI program that teaches itself

2017-10-21 Thread Russell Standish 
On Sat, Oct 21, 2017 at 09:58:20AM -0400, John Clark wrote:
> On Sat, Oct 21, 2017 at 12:33 AM, Brent Meeker  wrote:
> 
> ​> ​
> > The problem is that, like most real problems, improving computer code has
> > no simple one-dimensional measure of "better".  Go games are won or lost.
> >
> 
> A computer program ​that does the same thing as another but is smaller and
> executes faster is objectively better
> ​; and although there is no guarantee small fast programs usually have
> fewer bugs than large slow programs, and the bugs they do have are easier
> to find and fix.
> 
> And if you complain that speed size and robustness are 3 dimensions not one
> then try making the most money. That's the great thing about the Free
> Market, one dimension rules them all. ​
> 

And in any case, there is such a thing as multiobjective optimisation,
by using the so-called Pareto frontal method.

I agree with John, that machine learning could be used to optimise
attributes such as performance, code quality (which is related to
size), a process known in the industry as refactoring, but in order to
do that one also needs formal verification techniques that proves that
the resultant code satisfies the original design requirements. Not
much existing code is written in such a way to make that possible.

That suggests another possible use for AI - take an existing code, and
get it to write unit tests that improves code coverage (stopping when 100%
coverage is reached). However, you cannot remove a human completely
from the loop, as when I do this process manually, bugs are found and
removed, which is one of a number of positive side-effects of this
procedure.

The trouble is there is still a gap between 100% code coverage, and
correctly implementing the requirements...


-- 


Dr Russell StandishPhone 0425 253119 (mobile)
Principal, High Performance Coders
Visiting Senior Research Fellowhpco...@hpcoders.com.au
Economics, Kingston University http://www.hpcoders.com.au


-- 
You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.

Re: An AI program that teaches itself

2017-10-21 Thread spudboy100 via Everything List 

Well, the news this week had a dimension all its own, with IBM's simulation of 
a 56 qubit processor that beats what Google currently has at 49 qubits, and 
would run at 1 billion times faster than the schematic simulation itself does. 
For me, I am not totally craving an intelligent Guy in a Box, which was what AI 
was all about, but rather, a machine that scans all science papers and by ultra 
fast combination and sorting, whips up new inventions and medicines, all on its 
own. 


-Original Message-
From: John Clark 
To: everything-list 
Sent: Sat, Oct 21, 2017 9:58 am
Subject: Re: An AI program that teaches itself



On Sat, Oct 21, 2017 at 12:33 AM, Brent Meeker  wrote:




  

​> ​
The problem is that, like most real problems, improving computercode has no 
simple one-dimensional measure of "better".  Go gamesare won or lost.




A computer program ​that does the same thing as another but is smaller and 
executes faster is objectively better
​; and although there is no guarantee small fast programs usually have fewer 
bugs than large slow programs, and the bugs they do have are easier to find and 
fix. 




And if you complain that speed size and robustness are 3 dimensions not one 
then try making the most money. That's the great thing about the Free Market, 
one dimension rules them all. ​






 John K Clark









 



-- 
You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.


-- 
You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.

Re: An AI program that teaches itself

2017-10-21 Thread John Clark 
On Sat, Oct 21, 2017 at 12:33 AM, Brent Meeker  wrote:

​> ​
> The problem is that, like most real problems, improving computer code has
> no simple one-dimensional measure of "better".  Go games are won or lost.
>

A computer program ​that does the same thing as another but is smaller and
executes faster is objectively better
​; and although there is no guarantee small fast programs usually have
fewer bugs than large slow programs, and the bugs they do have are easier
to find and fix.

And if you complain that speed size and robustness are 3 dimensions not one
then try making the most money. That's the great thing about the Free
Market, one dimension rules them all. ​


 John K Clark

-- 
You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.