Sunday, January 8, 2017

Google Ain't No Minnie Earl Sears

Image result for minnie earl sears

Not to pick on Kevin Drum, who I genuinely like and respect but he has had two recent posts that seem to me to exhibit far too much credulity for the superstition of machine intelligence.  The first one is more complex as it deals with his noting that Google Translate has "gotten better" recently due to changes in the algorithms used.  The implication is that Google, wherever or what ever that artificial entity is, is better at understanding human languages and transferring information from one to another.   That involves a lot of complicated and unanswerable questions about language and, tempting as it was to research and write about those and why Google the great and powerful didn't understand anything, the complexities in making that case didn't appeal to me with the world on the verge of having its biggest superpower become the plaything of the Russian Neo-Tsar and ruled by a fascist mad-man in his power.  I suppose that's the best prospect under the Republican-fascist regime that is taking hold, that at least the senile 2-year-old our system has elected will be under the control of an adult, though an evil adult.

Anyway, Kevin Drum gave me another chance to address the alleged understanding of computers and computer programs, even big, big ones in a later post.

A little while back I mentioned that Google Translate had gotten a lot better overnight when they switched to a new machine-learning algorithm. Their voice recognition got better too. And so did its question-answering capability.

I was chatting about this at Christmas with my family, and we all decided we should test it. But not with anything boring. We know that Siri and Google and other digital assistants can find nearby coffee shops or tell us the weather in Berlin. How about something harder? The conversation then morphed into something about pencils, and my mother said she only trusted erasers that are pink. But why are they pink, we wondered? Why indeed?

So there you have it? Not only did Google understand me, even with a cold, but it also understood the question and provided a brief and precisely on-point answer, which it read off very nicely. Impressive!

Anyway, this strikes me as close to Watson-esque. The thing is, this is not as simple a question as it seems. It requires a fairly sophisticated understanding of context and meaning. And finding a source that matches the question perfectly is also pretty amazing. If my phone can do that, how long before it can drive a car too?

The answer that first came to me and a number of his other commentators was that, far from Google doing any thinking, the results of its new algorithms are based on the thinking and choices of the many, many thousands of people who asked the same question, probably in a wide but limited range of forms the question could take and their choices in which of the results on older versions of Google they clicked on to find an answer to it.  The machine didn't do any thinking, the people who came up with the rules for matching and listing a string of characters input to it with an existing set of strings of characters did a lot of it and the rest of the thinking involved was in the choices people made about clicking the lists that came up from the entire line of entirely human thought.  The engineers who came up with the original means of doing that and the, hopefully, improved means of doing that, no doubt based on the choices people made in the past are the ones who did the thinking.

The machine no more thinks than the index in a book or a card catalog in an uncomputerized library does.   A well indexed book or a well cataloged library had more work, more thought, more insight put into the information retrieval aids than ones which are less well indexed or cataloged.  The same goes for a computerized library catalog, the main advantage of that is that the number of subject listings might be larger because the computer records won't take up room in a card catalog or be as expensive to print on card stock.  Though the numbers of those subject headings will be of variable usefulness.  I always found those in Sears Subject Headings to be very well thought out, like Google, they were chosen in consultation with a large number of librarians who did a good job of cataloging their small libraries.  The alternative of using the far more detailed Library of Congress headings would have been too expensive for a small library and most of them would probably not have been commonly useful.

It's even more basic than that, though, the choice of pixels or other means of symbolically recording thought are no more akin to the human acts of thought and articulation than the ink on a page that is given semiotic significance by the people who do that.  As far as the machine goes, it's far more like paper and ink sitting on a table, any information recorded with them depends on human action and any consumption of that information depends, as well, on human action.

The disabling effects of materialist ideology, not only preventing understanding of these mundane realities but creating their materialist phantoms and materially based ghosts in machines they generate the most stunning of ironies.   And a lot of the most educated and intelligent people among us fall for them.

I never thought about the time I spent working in a small, rural library - under the supervision of a professional (retired) librarian - as perhaps inoculating me against that superstition but I've never bought it.   I knew that to come up with useful subject headings to type on cards you had to consider it and even then it sometimes wouldn't work.  It worked best for mundane questions, not as specific as "Why are erasers pink?" that was more the work of a book indexer or an encyclopedia compiler.   I never mistook a book or a card catalog as an intelligent object, just a useful one, depending on how useful it had been made.

Try asking Google how Americans got duped into voting fascists into power and see how clear the answer is.  Answering that one takes some real insight that computers will never have.

By the way, I don't think the Youtube of the audio response Kevin Drum put up as the right answer is really an answer as to why [SOME] erasers are  pink, never mind why his mother didn't trust erasers that aren't pink.


That doesn't tell anyone why even the "Pink Pearl" is pink, which would involve who made the decisions that went into them not changing the color to something else.  I'm pretty sure they put some coloring in their products, Pink Pearls aren't a natural color.   They could have chosen to make it a different color.  Lots of erasers aren't pink.   Back when I did lots of music manuscript, I preferred a white eraser with a texture that wouldn't erase the staff lines or leave a stain that might show up when the page was photocopied.  But, then, I always worked in a soft pencil that looked like pen when it was photocopied.  I hated it when one of my teachers insisted on ink.  Friggin' mess, totally pointless, the guy was a jerk.  

As to why his mother didn't trust erasers that aren't pink, she's the only possible source of that information.  If Drum had asked Google to give him the answer,  "Why does my mother only trust pink erasers?" I wonder if it could come up with an answer that satisfied everyone.   I also wonder if maybe Faber or someone else might have Google bombed to make their product come up high in such a search.  Tell me when Google can protect itself against Googlebombing, that might impress me.  

2 comments:

  1. Not intended to be an appeal to authority, but I really think David Bentley Hart makes some very powerful, but also head shakingly obvious points about the worthlessness of comparing human thinking to computation.

    https://www.youtube.com/watch?v=FIAZoGAufSc&t=467s

    I really don't think it takes a genius to understand that comparing thinking and computation is a bad idea. I think we've become too enamored with computers and not enough with our own capabilities.

    ReplyDelete
    Replies
    1. Especially considering that all acts of computer computation are the product of human choices and human actions of thought.

      It is instructive, how gullible even highly educated, highly intelligent people are when the machine "doing" the calculations are electronic or manipulate symbols instead of beads in a frame. No one ever seems to have made the mistake when it's an abacus that's involved, just when it's fast and automatic. It doesn't take a high level of complexity before even really smart people stop considering and start gaping.

      Delete