Monday, July 29, 2024

The Birth Pangs of the AI Revolution

As companies move to use AI for routine tasks, they are finding that this sometimes creates more work for their human employees. I already mentioned here that companies are hiring more translators; some people are claiming that this is because editing AI-generated text is more time consuming that just making the translation yourself. Forbes is running a story with this headline:

77% Of Employees Report AI Has Increased Workloads And Hampered Productivity, Study Finds

Part of the problem is that firms are just telling their employees "use AI to speed up your work," without telling them how. But how to use AI efficiently without loss of accuracy is actually a difficult problem that requires some high-level thought to sort out. Sometimes the whole workflow model needs to be changed.

And this:

In tech, especially programming, we're finding that a large swath are getting their code generated in like 5 minutes where it'd take them like 3 hours to write it but they're now spending much longer debugging said code because it's crap and they didn't write it.

I actually just had a conversation with one of sons about this next question, specifically whether laying out an progress tree for a game mod should be called "programming." I say it should be:

Programming is 95% understanding the problem + formulating a robust solution then 5% sitting down to write the actual code.

At the moment AI can only do that last 5%. That will probably change, but it might not change fast.

Let me tell you sometime about running a novel with lots of dialogue through a grammar checker. . . .

2 comments:

  1. A illustrative anecdote.

    I've used computers since the early 90s, and we've had spell-checkers for more or less that entire time. I remember the earliest ones being clunky, having to be manually run as their own programs or subprograms, and not being very accurate.

    Over time, spellchecks got more accurate, and got integrated into word processing programs and functions, often running automatically in the background as you typed. But they were designed to be discrete and to respect human autonomy - they would highlight a word they detected might be incorrect, and if you clicked it might pull up a list of words that you could click on to choose from and autofill, but it more or less just flagged things for human review.

    Today, in my current job, I am stuck using computers where the spellcheck cannot be turned off - and, indeed, the default functionality is not a spell-check, but an "autocorrect", which far more frequently gets things wrong and introduces errors where none existed, than it ever does get things right and help avoid or correct errors.

    The alternate functionality which I have to resort to is a form of actual spellcheck, but an aggressive and overbearing one that does not respect user agency.

    If a word is flagged, and you click on it to correct your typo - perhaps expecting to be able to simply press a key to add a missing letter or press Delete to undo an extraneous keypress - the system ACTIVELY PREVENTS your attempts to edit your work manually, because it locks the keyboard and renders it non-functional.

    It instead FORCES the user to select an option from the pop-up suggestion menu, before the keyboard will be unlocked. These choices are often not helpful; and even when they are useful, it's still slower than and more frustrating to use than the system simply allowing the user to apply their own judgement and continue to actually type - you know, the primary and fundamental method of using a computer to compose text.

    ReplyDelete
  2. Sorry - the above message is mine, the new page format defaulted to anonymous without my noticing.

    ~~~

    Anywho!

    I've observed that this sort of thing has become the new norm in the modern world, and in corporate settings in particular. A sort of automated paternalism has been adopted, in which employees are seen as unfit to use their own human judgement when problems arise, and instead, company trust is wholly placed in "AI" and other automated computer systems.

    This might not be a terrible thing if these automated systems were designed to be able to meet the highest demands of rigor and offer the greatest degrees of flexibility - but the corporate desire to cut costs at every conceivable turn means that the automation systems that get adopted are the ones devised and created by the lowest bidder, and this is reflected in their numerous and severe flaws and limitations.

    In my current job, it is a daily routine to run into issues that are completely impossible to resolve in the proper and intended manner dictated by management, because the software we are provided with simply does not account for any deviation from "normal" expected behaviors and outcomes. We are constantly having to circumvent these limitations manually, by telling the system what it wants to hear even though it isn't true, and then relying on physical means to ensure tasks get completed the way they NEED to be completed, rather than the impossible way the computer THINKS it should somehow magically work.

    We have, naturally, reported these issues up the chain, and been met with resounding silence. The few times we've gotten any kind of response whatsoever, we have been informed by management that the problems we report either: A) aren't really problems, and we're either just imagining them or exaggerating them; or B) are real problems, but they are ones which must inevitably be the result of employee incompetence, because the machines are infallible.

    We have - I think understandably - given up on reporting anything, and just fudge the numbers, break the rules, and fix the problems ourselves, in order to ensure our customers are properly handled. We tell the computers what they want to hear, even when it's flatly false, and because our customers are kept happy by our actual human judgement and efforts, corporate doesn't notice and couldn't care less.

    It feels like we're living in a Soviet-style dystopia where reality is subjected to the whims of an unimpeachable inhuman bureaucracy, but oh well.

    We joke with our customers about the absurdity of the situation, and they commiserate with us and tell us they appreciate what we do for them.

    Meanwhile, thousands of miles away in some faceless monolithic corporate office, the executives are happy because they get to keep their heads in the clouds instead of having to stop and actually pay attention to reality on the ground. The company makes money, the customers are happy, and they get to take three hour lunches and then leave the office early to go golfing.

    "The system works!", they think to themselves. "And it's all thanks to the automation we set up!"

    ReplyDelete