SoylentNews

This Speech Recognition Code is 'Just As Good' as a Pro Transcriber

posted by martyb on Thursday October 20 2016, @04:45AM   
from the nature'll-anguish-wreck-ignition dept.

Phoenix666 writes:

Microsoft on Tuesday said that its researchers have "made a major breakthrough in speech recognition."

In a paper [PDF] published a day earlier, Microsoft machine learning researchers describe how they developed an automated system that can recognize recorded speech as well as a professional transcriptionist.

Using the NIST 2000 dataset of recorded calls, Microsoft's software performed slightly (0.4 per cent) better than the error rate the company attributes to professional transcriptionists (5.9 per cent) for the Switchboard portion of the data, in which strangers discuss a specified topic.

There goes your bright future as a court recorder...

---

Original Submission

• Re:first post for basic income Re:first post for basic income (Score: 0) by Anonymous Coward on Thursday October 20 2016, @07:03AM

  by Anonymous Coward on Thursday October 20 2016, @07:03AM (#416511)

  Are you really fucking stupid enough to compare board games, where each player has perfect knowledge of the state of game, to legal proceedings, where one party has motive to fabricate evidence and the other party has opportunity to conceal evidence? I hope you get charged with terrorism for secret reasons that cannot be divulged on account of national security, and I hope you die in jail before you are allowed to see a lawyer.

  Parent

• Re:first post for basic income Re:first post for basic income (Score: 0) by Anonymous Coward on Thursday October 20 2016, @09:26AM

  by Anonymous Coward on Thursday October 20 2016, @09:26AM (#416540)

  There's also Watson winning in Jeopardy. That's clearly unlike Go or Chess. And if the transcription software works as well as Microsoft claims, that's also something you cannot simply do though simple rule-evaluation.

  And sure, AI is not yet ready to do lawyering. But there's nothing indicating that it won't ever be.

  Parent

  • Re:first post for basic income (Score: 0) by Anonymous Coward on Thursday October 20 2016, @11:07AM

    by Anonymous Coward on Thursday October 20 2016, @11:07AM (#416566)

    How well does Watson play if you lie to it, change the rules in the middle of the game, and deliberately set it up to lose? Because that's how the game is played in real life, by stonewalling, railroading, and moving goalposts. Let's see Watson deal with the kind of shit that goes down in a court of injustice.

    Parent

• Re:first post for basic income Re:first post for basic income (Score: 0) by Anonymous Coward on Thursday October 20 2016, @05:25PM

  by Anonymous Coward on Thursday October 20 2016, @05:25PM (#416807)

  Lawyers will always have a huge disadvantage against Software AI, Humans will never have perfect knowledge of the law and all relevant case-history. Think about who will outmaneuver who.

  Parent

  • Re:first post for basic income (Score: 0) by Anonymous Coward on Thursday October 20 2016, @06:58PM

    by Anonymous Coward on Thursday October 20 2016, @06:58PM (#416882)

    An AI lawyer would first have to be able to construct a decent LIE.

    That is arguably a function of General or Strong AI and not something we are even close to yet by any estimate. (Even Elon Musk's optimistic bullshit)

    Parent

• Re:first post for basic income (Score: 3, Informative) by urza9814 on Friday October 21 2016, @05:04PM

  by urza9814 (3954) on Friday October 21 2016, @05:04PM (#417314) Journal

  Are you really fucking stupid enough to compare board games, where each player has perfect knowledge of the state of game, to legal proceedings, where one party has motive to fabricate evidence and the other party has opportunity to conceal evidence?

  Ironically, evidence discovery is the one part of law that *has already* been automated, although only in limited test cases so far. Lawyers don't *gather* the evidence, that's going to be detectives / law enforcement in criminal cases or possibly the client themselves in civil cases. Lawyers mostly review what others have collected, and reading and preparing documents is a pretty good target for automation. Especially legal documents, since they usually have stricter standards of language and formatting.

  And your concerns about fabricating/concealing evidence is something that cops might do, but lawyers generally can't. You're required to provide the other party with the evidence you are going to use at trial -- even if it isn't requested (see: Duty to Disclose). You can't just keep something secret until the day of the trial. And the attorney would have trouble fabricating anything too because they aren't the ones generating it. It's gonna be pretty obvious when the coroner report from the coroner disagrees with the coroner report provided by the other attorney. Maybe they could bribe a "witness" to lie, but I don't see any reason a computer wouldn't be able to pick out discrepancies between the evidence and the witness testimony. Probably even better than a human could, assuming the computer can properly parse the statements.

  Previously on SN:

  Lawyers have been described as the canaries in the coal mine in the face of a wave of automation now beginning to displace highly skilled white-collar workers as the increasing reliance on so-called "e-discovery" software in lawsuits raises the specter that $35-an-hour paralegals as well as $400-an-hour lawyers could fall victim to programs that could read and analyze legal documents more quickly and accurately than humans.

    - https://soylentnews.org/article.pl?sid=16/01/04/2224232 [soylentnews.org]

  There's also this one, though the scope is much more narrow:

  An "automated lawyer" chatbot service has successfully challenged and overturned more than $2.5m in parking tickets in New York and London, according to its inventor.

    - https://soylentnews.org/article.pl?sid=16/06/28/2148259 [soylentnews.org]

  Parent