About The Bots - part the first

Well, about time I had an essay on the rampaging chatbots. Don't you think? I promise this isn't generated by one of them. Ha.

Today's report in the NYT is the lengthiest I've seen. Which doesn't mean a whole lot, since I've not been interested all that much. The angst and consternation over what these AI programs produce appears to be happening almost in lockstep, one version with the others. In other words: they're all fucked up in the same way. What might that same way be?

Well, you already know the answer (if you're a regular reader, anyway). The general claim is that these bots are driven by neural nets, which are made out to be really, really complicated thingees.

A commonly used cost is the mean-squared error, which tries to minimize the average squared error between the network's output and the desired output.

If the clowns pushing this three-card Monte on the public were honest, AI (as of today) is just standard inferential stats - minimizing squared differences, albeit with the sheen of 'complexity' to distinguish it from your Stat 101 class exercises. They ain't be no diff; it's all the same. And, of course, that's all there will ever be for AI, because correlation exercises, no matter how complex they are presented as being can't do anything else.

Which is why they can never be reliable.

It is just generating text using patterns that it found on the internet. In many cases, it mixes and matches patterns in surprising and disturbing ways.

We already know, don't we, that the RRW are far more prolific on the innterTubes than the Woke Crowd, so the likelihood that a bot will find itself using RRW memes and such tends to the RRW mindset (if one is allowed that anthropomorphism); that's just what correlation hunting gets you. Remember, all these AI machines are just calculating correlations; i.e. minimizing squared differences. A lot of them very fast, but not, qualitatively, any different from your multiple regression or ANOVA class. The bots, alas, don't publish the p-values of what they generate. One can only hope.

As one of the interviewees put it

This is terra incognita. Humans have never experienced this before.
-- Terry Sejnowski [from my dead trees edition of today]