markpasc (markpasc) wrote,

  • Mood:
  • Music:

Neural McCarthy language grooming

But perhaps [McCarthy] knew—as, I guess, few of the rest of us then did—that he could afford to be profligate. The people who were rallying around him didn't really care whether he was technically right or wrong about X or A or B or Z. In their judgment, it was mere splitting of hairs to distinguish between a Communist, ex- or otherwise, and a liberal or internationalist in the government service. They were all dogs to be beaten with any stick, and McCarthy was the first really handy man with a stick to have come along.

I had been meaning to post that or something like it for a while. Here's another thing I'd been meaning to post (my emph):

However, if you look at humans we [like other primates] maintain a group size (of about 150) which matches our neocortex size [similar to other primates'] but which is way over the time budget we allow for it (we'd need to spend about 42% of our time on one-on-one grooming, but it doesn't seem we do).

Dunbar contends that humans evolved vocal grooming (language) as a more efficient form of bonding. Assuming that our closest ancester, the chimpanzee, has hit the time budget limiting factor, and that our extra efficiency has all come about with the transition to vocal grooming, this means language is 2.8 times more efficient for bonding than the mechanism nonhuman primates use.

In other news, I finally got the neural network software (conx) working, but apparently a 112 node wide network with 689 training cases isn't the best tack to take: it's taking a very, very long time, and the RMS error actually went up between training epochs 100 and 200. So it's probably taking literally forever to train, which says to me I should try not using the 112-node wide network shape. I suppose I'll wait and see what happens by epoch 300 though.

  • Post a new comment


    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened