Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
AI Network Technology

DeepMind AI Beats Humans At Deciphering Damaged Ancient Greek Tablets (newscientist.com) 29

An anonymous reader quotes a report from New Scientist: Yannis Assael at DeepMind and his colleagues trained a neural network, a type of AI algorithm, to guess missing words or characters from Greek inscriptions, on surfaces including stone, ceramic and metal, that were between 1500 and 2600 years old. The AI, called Pythia, learned to recognize patterns in 35,000 relics, containing more than 3 million words. The patterns it picks up on include the context in which different words appear, the grammar, and also the shape and layout of the inscriptions.

Given an inscription with missing information, Pythia provides 20 different suggestions that could plug the gap, with the idea that someone could then select the best one using their own judgement and subject knowledge. "It's all about how we can help the experts," says Assael. To test the system, the team hid nine letters of a Greek personal name from Pythia. It managed to fill in the blanks. In a head-to-head test, where the AI attempted to fill the gaps in 2949 damaged inscriptions, human experts made 30 per cent more mistakes than the AI. Whereas the experts took 2 hours to get through 50 inscriptions, Pythia gave its guesses for the entire cohort in seconds.
The arXiv paper is available here.
This discussion has been archived. No new comments can be posted.

DeepMind AI Beats Humans At Deciphering Damaged Ancient Greek Tablets

Comments Filter:
  • Or does it only work with languages already deciphered?
    • by AHuxley ( 892839 )
      Fast look up needs the correct data sets to start with :)
    • Several years ago they were already able to determine with a simple mathmatical formula based on character patterns and word counts that the Voynich manuscript almost certainly does not contain any actual language, ciphered or otherwise. It is most likely an expensive hoax sold by an artist to some royal or rich noble.

      • by AHuxley ( 892839 )
        Try it on the Zodiac ciphers?
        • Yea, I think the Zodiac ciphers passed the test. Dolphin squeaks pass the test. Even *encrypted* data streams pass the test. So the fact the Voynich manuscript does not, shows that the character patterns in the document are too even and regular to represent language. Someone put a lot of thought into making the characters look like they made up words and sentences, but this person was not a linguist, and in fact there is no proof they were even literate. They may have simply studied a lot of books they

          • Aren't encrypted data streams supposed to fail any such test?
            • by Zeroko ( 880939 )
              If the test is trying to detect being sufficiently non-repetitive, encrypted streams should have no problem passing, because they should appear to be maximum-entropy. Whereas people are notoriously bad at being random, so gibberish will tend to be less random than meaningful content in some real language.
              • Ah, I apparently misunderstood the fourth sentence in the comment I was responding to. Somehow I thought the exact opposite was being tested for here. Thanks.
    • It might be able to predict what the next page of the Vms would be, if there had been another page.

  • Now use it to decipher the Mohenjo-Daro tablets.

    • It doesn't decipher anything; it predicts what missing letters are. Missing letters are common in old manuscripts, tablets, etc. When the Greeks conquered Troy, if there had been any manuscripts, they would at best now be fragmentary, and probably at least partly burned. Clay tablets might have fared better--that's why there are so many of them still around--but even tablets have a tendency to get broken when walls fall on them.

      As for the MD tablets, it's not even clear that these represent a writing sys

  • by SuperKendall ( 25149 ) on Friday October 18, 2019 @08:03PM (#59324074)

    Pythia provides 20 different suggestions that could plug the gap, with the idea that someone could then select the best one using their own judgement

    People worry a lot about being replaced by AI. But the truth in most cases is way more likely to be similar to this - many people with abilities strongly augmented by AI and robotics.

    • OK, people still work on farms too. But in 150 years the percentage of people working on farms has dropped from 50 percent to 2 percent. So you could say machines are just augmented people's effort, or you could say 96% of the people who used to work farms were displaced. In the overall market, you could say unemployment is super low, or you could say the labor share of GDP keeps plummeting and people are therefore working for less and less. Both are true.
      • In the overall market, you could say unemployment is super low, or you could say the labor share of GDP keeps plummeting and people are therefore working for less and less. Both are true.

        The unemployment rate doesn't count people who've been out of a job for more than a year or so. It doesn't count people who have given up and are stealing or doing odd jobs or sucking dicks to be able to afford food. It doesn't count people staying in abusive relationships because the person they're with is paying their bills. It doesn't count people with a job that doesn't pay their bills, going steadily further into debt even though they're working. It doesn't count people who have retired early because t

        • by Geekbot ( 641878 )

          Do you think the homeless are employable though? I mean, they don't have jobs: true. But could they hold one down? I think there must be a percentage of on-the-streets-homeless that are out of work but aren't they mostly mentally ill drug addicts? The California ones anyway?

          • Do you think the homeless are employable though? I mean, they don't have jobs: true. But could they hold one down?

            Most of them yes, if they had some place to sleep and take showers.

            I think there must be a percentage of on-the-streets-homeless that are out of work but aren't they mostly mentally ill drug addicts? The California ones anyway?

            Most of them come to California because it's possible to live here without a fucking igloo.

          • Chicken and egg mostly. If you lose your job and then lose your housing, its easy to have a mental crisis and turn to drug abuse. That's why the "housing first" approach to homelessness tends to work better than most other solutions. Get the homeless a stable place to sleep with an address, and you can then begin providing the rest of the services they need.

            The broader question is even if they can get sane and sober if they're employable at all. A number of the homeless in my area are in their 30s through 5

    • People worry a lot about being replaced by AI. But the truth in most cases is way more likely to be similar to this - many people with abilities strongly augmented by AI and robotics.

      The effect is that a old team of 10 people is now replaced by one person with AI augmentation. The other 9 are still left without a job.

  • Considering that I am a typical human, I would say pretty much anything can beat humans at deciphering damaged ancient Greek tablets
  • The tail of the article says humans must still piece together the text first. Why is this so? Can't we place the pieces on a glass plate so that it can be laser scanned and then virtually reassembled by AI too? One would think if the AI was already trained it could perform the most grammatically correct assembly even when it had to guess fit.

  • ...when it can decipher any Perl code of our time
  • The experts had to decode "2949 damaged inscriptions"? Maybe the fact that they had so many to do was the cause for the errors and they were rushing? Not that people trying to hurry could ever be blamed for causing errors, right? /s
    2 hours is a really short time to analyze the data. That is around 2.4 seconds per inscription. The article doesnt state how many experts were involved, but even if there we 30 of them (doubtful), one per minute is still a fast rate for a 2 hour period.
  • "The AI, called Pythia"

    So IOW, the backend is Oracle and the frontend written with Delphi?

Time is the most valuable thing a man can spend. -- Theophrastus

Working...