The AI Misinformation Epidemic

Tesla has a market cap larger than Ford. Tesla also has been on the edge of bankruptcy every couple years and recently raised $1 billion in capital.

From what is really going on, nearly all of it looks like (1) in some cases a lot of new data, (2) some new, faster processor hardware, e.g., based on graphical processor units although the x86 processors are astoundingly fast anyway, able to manipulate the new data, (3) for what manipulations to do on the data, some tweaks of some of the work of L. Breiman and his CART -- Classification and Regression Trees, (4) fitting with S-shaped (neural) sigmoid functions, and (5) the radar, etc. engineering of autonomous vehicles. E.g., the neural networks might be able to simulate the operation of a neuron in a worm.

What I don't see is (A) much progress in better methods for how to manipulate the data, that is, the basic applied math and (B) progress in working with concepts and causality -- in the history of science, progress with concepts and causality did well where the new methods would need Nevada full of disk drives of data. E.g., space flight navigation is based, first, on Newton's law of gravity and second law of motion, not on fitting massive of data via deep learning.

Uh, when the AI people use classic regression analysis as in Draper and Smith, etc. and the IBM Scientific Subroutine Package, SPSS, SAS, R, etc. to find some regression coefficients, they claim that their machine learned the coefficients. Gee, I didn't see that in, say,

C. Radhakrishna Rao, Linear Statistical Inference and Its Applications: Second Edition, ISBN 0-471-70823-2, John Wiley and Sons, New York, 1967.

or Breiman's CART.

Again, the rest I see looks like hype.

E.g., it appears that there is a basic, clever, publicity, hype idea: Whenever do some technical work, give it a catchy name. Then just use the name in the hype and ignore what is really going on in any math or science.

E.g., a while back I published some work in statistical hypothesis testing that is both multi-dimensional and distribution-free. The intended use was for high quality hypothesis testing of zero day problems in computer networks and server farms. Alas, I neglected to give the work a catchy name!

The OP seemed surprised that a big, famous company might say things where they know better. Why not? Just assume that they are trying to sell something to some people with money.

Similarly for the news media: They want eyeballs for the ad revenue. That the newsies are willing to write junk to get eyeballs goes back at least to Jefferson's remarks as in

http://press-pubs.uchicago.edu/founders/documents/amendI_spe...

and are not nearly new.

We have long had a good filter to apply to the writing of the newsies: Does the writing meet common high school term paper writing standards for solid references and primary sources? Rarely is the answer yes.

My view of the printed news is that it can't compete with Charmin, not even with cheap, house-brand paper towels. For the electronic versions, they are not useful even as fire starter, shredding for cat litter, or wrapping dead fish heads -- that is, are useless.

So, don't read them.

And don't debunk them, either -- debunking wastes your time and is something the newsies long since have ignored. The newsies have no shame.

Ignore the newsies. They ignore the debunking efforts.

My startup manipulates some data, and how is from some applied math I derived. From what I've seen of AI, my work would qualify as quite good and innovative AI -- besides the work is solid theorems and proofs. Still, I see no good reason to give my work a catchy name or call it AI. E.g., I'm not trying to fool anyone. Or, why would I want to associate my good work with a lot of hype to fool people?