Monday, May 23, 2011

Human vs. Computer Memory and Language

Gary Marcus in his book Kluge which disses both  God and evolution claims that human memory is inferior to computer memory.  Odd, since humans constructed human memory.  Well, that may not be germane, but the question of whether computer memory is superior is.

Computer memory, which Marcus calls Postal Code memory, assigns each item in its memory to a specific location.  If you search for a word or phrase that has  been input into that memory, you will get it virtually instantly. This can be handy, especially when searching for trivia like the dates of given movies or for information about Zachary Taylor.  It works less well when you're looking for concepts.

One problem is that someone has to key the item in memory.  Humans develop memories this way, too, of course, but they also acquire them from personal observation.  That way, nobody is just a product of his or her parents' or caregivers' knowledge.  Computers are limited to what programmers have input into their memory.  Furthermore, once basic programs are written, it is very hard to change them so we're, in effect, wedded to early programs which have been improved a little over time, but not substantially changed.  Humans, on the other hand, can change their minds completely, can reorient themselves, transform themselves, and, in the process change their memories.  Computers can't do any of these.

Lanier, in his very startling book, You Are Not a Gadget, tells the story of MIDI, the computer music program (p. 22).  It seems that its creator Dave Smith wrote it strictly for a synthesizer, but that is inadequate to "describe the curvy, transient expressions a singer or a saxophone player can produce."  In other words, it is not useful for a composer of serious music of most genres.  However, because that program has become entrenched, all subsequent music programs have been written on the MIDI platform.

Imagine humans limited in music making by a primitive, non-sophisticated internal explicable program instead of having our messy creativity, which can't be written into code.  Of course, someone someday may write a program suitable for orchestras and 12 string guitars, but computers, dependent on the MIDI model, wouldn't be able to run the program unless computers themselves are changed.  However, experience so far has shown that computer languages conform to the platforms already generated, as limited as they may be even after changing a line or two of code.

I am not a musician, but I am a scholar, and I do make use of computer memory in my research.  Is it superior to human context driven memory?  That is, human memory--and probably all mammalian memory--is triggered by the context.

Three years ago, when I first started researching a book I'm supposed to be writing about dogs, I dutifully searched the college's electronic databases for "dogs, intelligence tests," I got a lot on dogs, including dingos, dholes, wolves, New Guinea singing dogs, but nothing on intelligence testing, although I knew several studies had been conducted and published at that time.  I tried other key words: canis lupis familiaris, canis familiaris, intelligence testing, tests, test, laboratory, clinical, reasoning--the list went on. I couldn't find one article about intelligence testing of dogs.

While writing this post,  I just googled "dogs intelligence tests" and got one overview article, but not all  the studies on the subject. I did find articles pertinent to my search, both then and now, but never easily.  One study done with a border collie selecting symbols on a computer touch screen has never come up in a search, but I saw a National Geographic TV program showing this dog's prowess.  As for the studies I've amassed over the past three years, they all had to be culled from a mass of irrelevant stuff that happened to use one of the key words plus my own reading in many, many books and journals, none of which were retrieved from a computer search..

If you search human memory, you will (if you're not schizophrenic) retrieve only what fits the context you're interested in.  If you try to remember something about dogs, meaning domestic ones, you won't get a lot of information about wild dogs or extinct ones, but you do in a computer search.  Yes, sometimes we search our memory and don't get a result, as Marcus notes. However, frequently the word will come to you out of the blue.  At my age, it sometimes takes a few hours, but it shows that one's mind is still searching even when you're no longer actively thinking about the topic.  Forgetting can be a small price to pay for a flexible context-dependent memory.

Actually, if we did have a computer memory as Marcus wants, we'd have to remember the address of the neuron that contains the word in memory.  That's how computers work. You put in a word, and google just goes to the computer address where that word is stored forever.  In human memory, as you have more experiences or solve more problems, words forge networks with those novelties.  Our memories aren't static. Nor are they dependent on what someone else thinks.

Well, I did get too much information on computer searches, so why is that bad? It's not that it's so bad.  It is messy . Looking at schizophrenic speech, one easily sees what's wrong with unwanted associations when you're trying to create in words. Some schizophrenics, those called Speech Disordered or SD,  (see Chaika, Elaine Psychotic Speech:Beyond Freud and Chomsky) respond to questions this way:

(Asked what color a plastic chip is) "Looks like clay. Sounds like gray. Take you for a roll in the hay. Hay day. May day. Help. I need help."

(Asked mother's  name) "My mother's name was Bill--and coo? St. Valentine's Day is the start of the breedin' season of the birds. I love birds. Especially parakeets.  They work hard..."

The difference between such associational speech and computer searches is that human memory also accesses words on the basis of shared sounds (rhyme) and concepts.  Computers aren't that smart.  Bill and coo is an old idiom referring to lovers' sweet talking. It evokes both love and bird  The idiom is based on "lovebirds," doves, who touch bills and coo when mating.  St. Valentine's Day is the holiday for celebrating love. Hence that surfaced.  SD schizophrenics aren't able to control unwanted associations between words when they talk or write.  Non-schizophrenics automatically retrieve only words that fit the context, except for the occasional slip of the tongue, or the sudden, "Oh, that reminds me..." phenomenon.

Computer retrieval is not smart enough to allow for context, so if you search "Adam Gerber and World War II" you get articles about Gerber (not necessarily Adam--I made the name up), but also "Public Opinion in World War II," articles about Ansel Adams, the photographer (I input only Adam), Rabbi Adam Mintz, a Gerber World War II knife, Flags of our Fathers and Adam Beach... on for 5 pages of varied references.  The computer won't limit its search to the combination of words you use.  It will give you references to each word you input.  It doesn't recognize names, so that Adam and Gerber are treated as separate items. If there is a plural of either, the computer will select that, too.  However, World War II is apparently one whole item in computer memory.

I love technology, and am at a computer often for hours a day.  Certainly, I welcome googling for information, although, especially with the availability of smart phones, it has made people trivia laden.  Too often, they mistake trivia for knowledge or analysis. Even so,  once one has  information from a computer search,  human memory allows one also to access articles and books one knows about and, as one writes or talks, one remembers again something  read or something heard in a discussion. Human contextual memory is not limited to the first internal search.  It expands so long as one is on the same topic

To sum up: a computer is only as good as what is input into it.  It cannot learn by its own observations as humans can.  Therefore, its memory is always restricted.  Moreover, because computers don't know meaning, they only match words.  But, every word in every language has more than one meaning and more than one use, a fact Gary Marcus also bemoans--but he's wrong, as I'll show in my next post

5 comments:

Anonymous said...

Hey - I am really delighted to find this. great job!

Brenda F. Bell said...

While current search engines are supposed to search according to relevance, the process of making them "user friendly" made it less difficult to define the parameters of one's search. I used to be able to use the syntax "Adam Gerber" AND "World War II" and retrieve only those items that had both the specific string (name) "Adam Gerber" and the specific string "World War II". The removal of Boolean search in favor of free-text search has, paradoxically, made finding "exactly what one is looking for" more difficult.

John Cowan said...

Elaine: Did you search in Google Scholar? It's a much better search engine for scholarly books and articles than either mainstream Google or the fairly limited search tools provided by electronic publishers like JSTOR.

Brenda: "AND" is no longer necessary. Searching for ["Adam Gerber" "World War II"] will get what you want (I write the square brackets for clarification; you don't have to type them.) You can also use | for OR and - for NOT. Google will sometimes find pages that don't match all your search terms: you can mark them as absolutely required by preceding them with +.

Brenda F. Bell said...

I've had issues with Google giving me "or" searches when I've had multiple terms or strings in the search line, despite their early assurances that their default was an "and" search.

The prepended "+" was early Yahoo Boolean syntax for "and" and for "required"; I've never known Google to accept that.

Brenda F. Bell said...

John: While Google has always maintained that its default is an AND search, I've had too many OR results come up, and it's failed to find any results when I've used double-quotes to specify an exact string.

The prepended plus- and minus-signs were part of Yahoo's Boolean expressions; once again, I've never found them to work in Google.