Stories
Slash Boxes
Comments

SoylentNews is people

posted by chromas on Tuesday September 10 2019, @10:11PM   Printer-friendly
from the 🗹⠀I-am-not-a-robot dept.

Deep learning excels at learning statistical correlations, but lacks robust ways of understanding how the meanings of sentences relate to their parts.

At TED, in early 2018, the futurist and inventor Ray Kurzweil, currently a director of engineering at Google, announced his latest project, "Google Talk to Books," which claimed to use natural language understanding to "provide an entirely new way to explore books." Quartz dutifully hyped it as "Google's astounding new search tool [that] will answer any question by reading thousands of books."

If such a tool actually existed and worked robustly, it would be amazing. But so far it doesn't. If we could give computers one capability that they don't already have, it would be the ability to genuinely understand language. In medicine, for example, several thousand papers are published every day; no doctor or researcher can possibly read them all. Drug discovery gets delayed because information is locked up in unread literature. New treatments don't get applied, because doctors don't have time to discover them. AI programs that could synthesize the medical literature–or even just reliably scan your email for things to add to your to-do list—would be a revolution.

[...] The currently popular approach to AI doesn't do any of that; instead of representing knowledge, it just represents probabilities, mainly of how often words tend to co-occur in different contexts. This means you can generate strings of words that sound humanlike, but there's no real coherence there.

[...] We don't think it is impossible for machines to do better. But mere quantitative improvement—with more data, more layers in our neural networks, and more computers in the networked clusters of powerful machines that run those networks—isn't going to cut it.

Instead, we believe it is time for an entirely new approach that is inspired by human cognitive psychology and centered around reasoning and the challenge of creating machine-interpretable versions of common sense.

Reading isn't just about statistics, it's about synthesizing knowledge: combining what you already know with what the author is trying to tell you. Kids manage that routinely; machines still haven't.

From Rebooting AI: Building Artificial Intelligence We Can Trust, by Gary Marcus and Ernest Davis.

If Computers Are So Smart, How Come They Can't Read?


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 4, Insightful) by Immerman on Wednesday September 11 2019, @12:45AM (3 children)

    by Immerman (3985) on Wednesday September 11 2019, @12:45AM (#892470)

    OCR isn't reading though - it's just converting text from analog to digital format. No different in concept from digitizing a cassette tape to MP3

    From the summary (not even the article):
    > If we could give computers one capability that they don't already have, it would be the ability to genuinely understand language.

    Though I personally I think that's putting the cart before the horse - I suspect that to understand language you first must understand the world that the language represents, which is a size large slice of the holy grail of AI research. I'd settle for just the ability to actually understand anything. As far as I can tell the best AIs in existence today are pretty good at recognition. Maybe, *maybe* AlphaZero actually understands the games it learns to play - but it seems far more likely that it just recognizes strategic and tactical threats and opportunities.

    Starting Score:    1  point
    Moderation   +2  
       Insightful=2, Total=2
    Extra 'Insightful' Modifier   0  
    Karma-Bonus Modifier   +1  

    Total Score:   4  
  • (Score: 2) by Freeman on Wednesday September 11 2019, @02:14PM (1 child)

    by Freeman (732) on Wednesday September 11 2019, @02:14PM (#892681) Journal

    You can read, without comprehension. Which is essentially what OCR+Text-to-Speech is, reading without comprehension. The computer doesn't comprehend anything, it just does what it's programmed to do.

    --
    Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
    • (Score: 2) by Immerman on Wednesday September 11 2019, @03:21PM

      by Immerman (3985) on Wednesday September 11 2019, @03:21PM (#892725)

      Can you really?

      Okay, yeah, most dictionaries do have something along the lines of "to utter written words aloud" as one of the definitions of "read", but it's typically buried under a host of definitions making some reference to comprehension or understanding, including of non-written things.

      It's also quite clear from context that they're talking about comprehension in this article.

  • (Score: 0) by Anonymous Coward on Wednesday September 11 2019, @05:21PM

    by Anonymous Coward on Wednesday September 11 2019, @05:21PM (#892800)

    Maybe, *maybe* AlphaZero actually understands the games it learns to play - but it seems far more likely that it just recognizes strategic and tactical threats and opportunities.

    This is the classical philosophical question of "how do you define life (or sentience)?" How do I know you "understand" Go? How do you know I do?

    It's just like duck-programming. If something looks like it is sentience, then it should get the presumption of being sentient, with the burden of proof being on the other person suggesting it isn't. This is kind of the point of the Turing test. Admittedly that is a flawed benchmark which can be rules-lawyer-ed to death, but it is a reasonable concept to me in the abstract.

    AlphaGoZero can beat the best human players, and it provides quality assessments of each move in the form of statistical probability of winning. If that's not "understanding," then I don't know what is. If anything, I think it "understands" the game better than humans do (much like somebody who has a chart of every winning move in tic-tac-toe/noughts-and-crosses understands the game better than a 4-year old playing it for the first time). If you think that that is not "understanding," what benchmark are you using to define that term?