|Wesley R. Elsberry
Joined: May 2002
|Quote (GaryGaulin @ Sep. 28 2017,21:50)|
|Quote (Wesley R. Elsberry @ Sep. 28 2017,21:19)|
|Quote (GaryGaulin @ Sep. 28 2017,20:50)|
|Quote (Wesley R. Elsberry @ Sep. 28 2017,20:37)|
|I've quoted Heiserman on Gamma and how he describes it as making "educated guesses", and I've pointed out where Heiserman's flowcharts and explication for Beta lay out random choice and do not say "guess".|
How David Heiserman describes Gamma as making "educated guesses" does not make it wrong to call a mechanism that generates a random response (when it needs to take a guess) a "GUESS" in a circuit that I had to draw, based upon ALL applicable models that parallel Heiserman's
You are going in circles in order to make it appear that you have a valid argument. Mudslinger......
Reiterating the point that Gary (1) relies on Heiserman but (2) gets Heiserman *wrong* is not a circle.
Then you should have no problem proving that the David Heiserman Beta and Gamma is not at the most fundamental level: a sensory addressed RAM where 2 bit motor controls and confidence levels are stored in Data by a random response/choice guess mechanism for taking a "guess" when its confidence level in a given response is zero.
Trivially, Gary is wrong even in his attempted misdirection. In his haste to reduce things, Gary has overlooked Heiserman's initialization of stored actions to "stop code", which also induces random choice of an alternative. I've even quoted the relevant bit recently:
The next step is to determine whether or not the motion code fetched from memory is valid. For the purposes of the present demonstration, the only invalid motion response is a stop code -- the code entered into all sections of the Beta memory during the INITIALIZE BETA SYSTEM operation.
If it turns out that the motion code picked up from Beta memory is invalid (a stop code), the system resorts to a bit of Alpha-like behavior and picks up a randomly generated motion code. That new motion code is then tested for validity. If it, too, is invalid, the system picks up another random motion code.
Sooner or later, the system finds a valid motion-code response to the current contact situation. The more first-hand experience the Beta has undergone in its environment, the greater the likelihood the response picked up from Beta memory will be valid.
I've already noted various other places where Gary's statements about Heiserman have simply been wrong. Gary doesn't dispute those. They still stand. Like this one:
Having a "Gamma" subroutine altering the contents of memory for systems that must have no changes made at all would just cause conflicts that crash the system.
It should come as no surprise to anyone that Gary's statement above is directly contradicted by Heiserman on pp.21 and 281 of his "Robot Intelligence" book.
Those Gamma-generated responses are bits and pieces of knowledge gained at various stages of its earlier life. The conjectured response is immediately tried whenever the occasion arises. If this response happens to work, the creature's confidence grows and it has additional information for firming up its notions about dealing with other future events. But in the event a conjectured response is partly or, in some instances, altogether wrong, the creature resorts to Alpha reflex activity.
The memory of a Gamma-class robot is in a state of continuous fluctuation and change, at least as long as it is interacting with a rich and dynamic environment. Put the machine into a sterile environment and you will find very little Gamma activity taking place. As a result, the creature will be largely unprepared to deal with unforeseen circumstances.
So portions of the creature's memory that carry confidence levels of 0 or 1 are loaded with suggested components of motion codes compiled in an earlier part of the subroutine. Responses carrying confidence levels of 2 or greater are not affected at all by the GAMMA FUNCTION subroutine.
In other words, a Gamma-class robot will never perform at a level lower than Beta-class, as it uses exactly the same fallback for handling incorrect responses. Heiserman also never envisaged Gamma processes as overturning good experience in his robots, contrary to Gary's claim. Gary has railed against people for dissing Heiserman in the past, but here we see Gary himself engage in clearly unwarranted critique of Heiserman. Plus, the "crash the system" part is ludicrous; the worst thing that ever happens in one of these systems is that an incorrect move is made, and that is handled without any such thing as a system crash. Gary will not be able to show that any such thing is justifiable on the basis of Heiserman's books.
So much for staying true to Heiserman; Gary doesn't even understand what Heiserman was saying.
Gary has "guess (not mutate)" as part of his discussion. His latest bit above shows definitively that he cannot base that on Heiserman, as all he now claims is that his connotation of "guess" is simply random choice. It has taken a while, but here we have it from Gary that he has no other basis for his "guess (not mutate)" phrasing than himself. He cannot claim Heiserman as his authority for "guess (not mutate)" because he has specifically repudiated using the only part of Heiserman's work that might possibly differ from random mutation. Thanks, Gary!
And, of course, Gary still has done nothing toward showing that he does not himself rely on "old junk".
"You can't teach an old dogma new tricks." - Dorothy Parker