Re: Why Software Is Bad and What We Can Do to Fix It
From: Traveler (traveler_at_nospam.com)
Date: 09/18/04
- Next message: Mark Nicholls: "Re: SubTyping versus SubClassing"
- Previous message: Jonathan Hendry: "Re: MVC - The right way"
- In reply to: Christopher Barber: "Re: Why Software Is Bad and What We Can Do to Fix It"
- Next in thread: Universe: "Re: Why Software Is Bad and What We Can Do to Fix It"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Sat, 18 Sep 2004 17:42:44 -0400
In article <pso656cbmlo.fsf@unicron.curl.com>, Christopher Barber
<cbarber@curl.com> wrote:
>Traveler <traveler@nospam.com> writes:
>
[cut]
>You claimed that "once timing is properly set ... it remains defect-free
>henceforth." What exactly does that mean? Isn't forgetting an indication
>that circuits are not entirely reliable?
Entirely? No one claimed that 100% reliability is possible, AFAIK.
Forgetting is a necessary component of cognition. Otherwise, you would
not be able to function because every little thing would remind you of
a zillion other things and you would need a lot more than 100 billion
neurons to contain it all. Sometimes, we forget things, not because
there is a bug in the system, but because we did not pay enough
attention to it. We don't pay enough attention to certain things
usually because we are distracted by something more important or we
just don't care. Often too, bad memory is a consequence of a physical
illness, old age, trauma or drugs and alcohol.
[cut]
>> >It is clearly asserted, but not clearly explained. I don't think the brain's
>> >elementary processes are really synchronous and had not seen any researchers
>> > claiming that it was.
>>
>> You are kidding me? Why does one need someone else (a researcher) to
>> prove something that is obvious?
>
>It may be obvious to you, but it is clearly not obvious to the rest of us.
>Yes, it is obvious that many neurons fire at the same time, but the rest
>of your claims about stability of timing could use some more explaining.
The reason that it matters is that the relative firing (signal
emission) order or timing of cells matters. If cells were not
synchronous, they would have to wait for sequential procedures
(algorithms) to finish before they can have a turn at doing their
stuff. Since algorithms have indeterminate durations, the timing would
be all messed up.
>> All neurons are synchronous, in the sense that they operate in parallel.
>> They wait for something to happen and they react by firing.
>
>Each neuron itself might be synchronous, although the fact that the
>"something" you refer to may actually be some arbitrary threshold triggered by
>thousands of inputs muddies the issue a bit.
Why? What is important is not the number of input connections but the
relative temporal order of the arrival of the afferent signals.
>However, two neurons do not
>always fire at the exact same time with respect to each other and are
>therefore asynchronous.
That is not the meaning of synchronous neurons. It means that the
neurons are all active concurrently. It has nothing to do with when
they fire. It means that, at every cycle (a discrete interval) every
neuron tests to see whether or not it needs to fire.
>> There are zillions of action potentials
>> occurring in the brain at any one time, i.e., synchronously. What
>> reference does one need to understand this simple fact?
>
>If you simply mean "synchronously" to mean simply happening at the
>same time without reference to the timing of the events, you should
>use a different word or else risk confusing computer folks who expect
>a more precise meaning.
Sorry, I was not aware of a different definition for the term
'synchronous.' AFAIK, 'synchronous' has always meant 'happening at the
same time', which is the definition given by the Merriam-Webster
dictionary.
>> >I think you really need to provide more explicit
>> >references to prior research to give your claims more weight.
>>
>> I don't think so. I cannot stand explanation by reference.
>
>I cannot stand when people cannot back up their assertions with
>hard evidence.
In that case, my site is probably not meant for you. Because this is
one of my pet peeves about peer-reviewed scientific papers.
>You should also consider that some readers may
>not already be well-read on matters of neuroscience and would like
>to read some supporting material relevant to your claims.
My site is not meant as a tutorial in neuroscience. The brain analogy
that I use is simple and should be common knowledge for most educated
people who are savvy enough to be interested in software construction
and operating systems. If someone is not educated enough to know that
the brain consists of neurons and that neurons are synchronous cells,
they have no business reading my site. I have to assume a certain
level of education. Otherwise, there would be no end to it. IOW, do
your own homework.
[cut]
>> >Any physical emulation of the brain would be massively
>> >less parallel than the brain using currently available technology and
>> >would thus be massively slower than the brain.
>>
>> Correct. But irrelevant to the issue of reliability.
>
>Perhaps, but extremely relevant to the suitability of this paradigm for
>real-world programming problems.
Why so?
[cut]
>> The brain is essentially a discrete signal processing mechanism. It
>> expects signals or spikes (hundreds of millions of them) to arrive at
>> their proper time. If they don't, the brain will fail.
>
>What exactly are you talking about? What exactly do you mean by "at their
>proper time?" Are you talking about exactly what point of time a signal
>reaches a synapse or the need for enough synapses to be stimulated at the
>same time for a neuron to fire?
The brain expects signals to arrive at a certain time relative to one
another. It learns to recognize things (visual objects, sounds,
speech, sensations, etc...) by the way in which they manifest
themselves in terms of sensory signals. There are only two types of
signals correlations: two signals are either concurrent of sequential.
If they are sequential, they are separated by a given interval which
is learned. At times, it is not the exact intervals that matter but
the relative order of the signals' arrivals. This is all crucial to
the brain's operation. The same is true of integrated circuits, as any
hardware IC engineer will tell you. And the same is true of software,
although you may not realize its importance.
>I still don't see what this has to do with reliability.
If the above does not help, fault me as a teacher. I've done my best.
But reliability depends on more than just accurate and consistent
timing. It also depends of the ability to easily identify
dependabilities, on compatibilities.
>I know you think I am just being a pain, but after hearing a lot of
>hand-waving brain analogies invoked by AI researchers and seeing it lead
>nowhere, I am naturally skeptical.
Skepticism is fine. I can understand people asking question because
they do not understand something. What I have little patience for, is
folks coming out of the woodwork and criticizing things that they do
not understand. My advice to critics is this: If you are going to
criticize, do your homework first. Only then will your criticism be
constructive.
Louis Savain
The Silver Bullet or How to Solve the Software Crisis:
http://users.adelphia.net/~lilavois/Cosas/Reliability.htm
- Next message: Mark Nicholls: "Re: SubTyping versus SubClassing"
- Previous message: Jonathan Hendry: "Re: MVC - The right way"
- In reply to: Christopher Barber: "Re: Why Software Is Bad and What We Can Do to Fix It"
- Next in thread: Universe: "Re: Why Software Is Bad and What We Can Do to Fix It"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
