Wesley R. Elsberry
Posts: 4991
Joined: May 2002
|
| Quote (GaryGaulin @ Feb. 28 2015,00:57) | | Quote (Wesley R. Elsberry @ Feb. 27 2015,23:13) | | Gary has explicitly disclaimed any need for biological plausibility before, |
You are a liar. |
Really? Why is it that I can find this link, then?
Gary disclaims the need to implement actual models
My commentary following that:
| Quote |
Now, let's take a moment to consider Gary's blithe discussion of removing biologically-relevant modeling from his PSC code. Trehub spends considerable time in his book establishing the conditions for making models that can be supported as biologically-relevant and comparing his models to data from biology to demonstrate that his choices were consistent with that. This makes for a enlightening contrast to what we just got from Gary, especially when it comes to code that is claimed to have *amazing* consequences for the study of biology. Tofu prime rib, anyone?
|
And there was the whole thing about not needing any stinking neurons.
From my response:
| Quote |
There is a difference between talking about an artificial neural system and implementing one. There's plenty of discussion in the artificial neural system community about how biologically relevant particular models are, and there's essentially a continuum from those with higher biological relevance to those with no biological relevance. (I remember hearing a talk by Hecht-Nielsen about how glad he would be to take all the insight for ANS models from things like spin glasses and leave the realm of biology entirely.) Gary can't just invoke "neurons" and be given a pass for having established biological relevance. And Gary can't just say that stuff he does in his code is sorta kinda like "neurons", therefore it is "neurons", therefore it is completely biologically relevant.
|
Gary's further response shows clearly that he presumes the biology conforms to his notions, not that his notions should show biological plausibility.
Gary:
| Quote |
The only difference is the number of data locations the RAM has. It barely even matters, especially where it uses the good-guess trick that I explain to add the properties of neurons to a digital RAM. And RNA, DNA and other memory systems have gene-like data that is easily modeled in a digital array, which the model has to easily include. Neural networks became more of an exception, than the rule, and boils down to their being just another kind of RAM anyway.
|
Which after a few more iterations leads to Gary's straight-up declaration of independence from biological plausibility:
| Quote |
Hey Bozo, the video explains why the model does NOT need to use one of the hundreds of unnecessarily complex algorithms that you demand are necessary in a K-12 level model of how any intelligence works.
|
And all of that because Gary couldn't bear to admit he was wrong to claim this:
| Quote |
That is part of the COGNITIVE MODEL that has NEURONS that they can make virtual ROBOTS with, like I do.
|
--------------
"You can't teach an old dogma new tricks." - Dorothy Parker
|
