Sunday, September 30, 2018

ML Again

I'm back to attending to Siraj, meaning I'm falling asleep, sometimes intentionally, while getting back into the namespace of Machine Learning, wherein I visit new-to-me nooks and crannies.

Siraj is one of my Youtube teachers, one of a cast of celebs that have come to frequent the privileged homes.  In recordings of live chat sessions, I see a lot of kids from Bangalore and elsewhere, chiming in.  They're learning about TPUs as a feature offered by the Google Mind.

This renewed focus on ML is apropos for other reasons.  The Philosophy of Mind debates go well with Indian food.  Dwaraka rocks.  We yakked about such marvels as The Turk in the Age of Napoleon.  AI was in some dimensions self booting and retains some of its carnival ambience.  Sophia and so on.

The idea of an AI Brothel in Texas came up, which I think distills the memes in a wicked brew that maybe Texans are able to handle.  Shades of Westworld.

Speaking of Westworld, as a test of Google's picture matching I slid the picture below from file manager to Google image search.  I'd snipped it from a Youtube of the opening credits. Google search recognized right away where it came from.

Carol (89) and Celeste (94) are both at the symphony, is the current hope.  Supporting evidence would be no evidence to the contrary.  I just dropped Carol off in the Nissan, just a block from the venue.  Celeste was arriving by rail.

Long before I worry about a Singularity, I'm going to be studying the anthropology of what people believe.

Siraj is suffering from no serious delusions about robots about to be real humans, yet already many humans have been branded as "bots".  The need for a term other than "subhuman" has been filled.  The "Russian Bot" T-shirt has been popular in some eCommerce circles.

Long before we say AI has closed the gap, we'll see ourselves in the mirror as highly advanced AI.  We're entitled to some vanity, as we admire ourselves as machines, cyborgs, not because anything has changed, but because we've closed a lot of gaps in the narrative.  We see ourselves as learning much the same way machines learn, by perpetually re-weighing what marbles we're given, by God's grace.

What's changed?  In part it's that our machine models now converge without becoming clones of one another, because of differing initial conditions and sometimes randomness in the algorithms themselves.  Two strong classifiers need not always classify the same way, and that matches our experience of intelligent humans disagreeing.  AI becomes more believable in proportion to its not always seeing eye to eye with itself.  Intelligence means nothing if not contentiousness.