A.I secretaries : Pros and cons

I'm talking about if it fires people for statistical issues like X amount of widgets produced or some shit. Sometimes people have shit going on that affects their work performance, and I don't want an AI making those decisions over anyones career.

You think HR doesn't keep a list of issues and fire you if you hit the arbitrary limit? I assure you, they do.

I'm not seeing how having human HR would make a difference there Barkey. An AI tool does exactly what you tell it to do, just like an HR department. If you fire people over that today, you'll continue to fire them over that tomorrow. You won't suddenly start firing people over such silly things in the future because you bring in AI. Like deboss says, if the AI is designed to study the problem, then it most definitely gives you a different solution to decrease in worker productivity than firing people.

I just don't trust an artificial intelligence when it comes to terms of proper judgement. If we're talking about prepared arbitrary numbers, you don't need an AI for that.

Speaking as a programmer, not so much. Or at least, it's very good at doing what you tell it to in relation to random and seemingly insignificant white noise, or combining long lists of what you tell it to do in unforeseen ways that like to end in disaster.

edited 27th Jul '12 10:14:52 PM by Pykrete

For the most part, the response for a lot of issues doesn't take much judgement by HR people, it's just a conditioned response to specific stimuli. As long as you keep the managers from getting any input and it gets tweaked by IT guys, it should be good.

And also, speaking as the child of someone who nearly got fired over bureaucracy, and then after he left had to have three people hired in his place, HR's own rubrics kind of blow goats themselves and shouldn't be taken as even more blindly and divorced of context as an AI would do.

edited 27th Jul '12 11:07:58 PM by Pykrete

That's not really how it is like at all. If a corporation is firing you over arbitrary rules, it's firing you over arbitrary rules. The AI didn't decide anything.

A business decides how it wants to run its company, how it wants to use new tools to help it make its decision and what data it thinks is relevant.

The whole point of using an AI is that it would make better decisions than a human would. If it doesn't, then there isn't any point to it (I'm assuming true AI, not just an expert system).

I could see a learning diary that could learn that an 8 AM meeting the day after a night out with the boys was impossible and rebuke you if you tried to schedule it that way.

Unless it knew your meeting was a no-brainer that didnt require much mental effort on it's owner's part.

And what if humans disagree with those decisions the AI decides are better?

If the AI predicts you will be bent over a toilet bringing your guts up it won't let you schedule a meeting.

Which comes down to how it will perceive these details. AI is easy, the challenge is in perceiving the world in a way which allows the AI to learn about its environment and draw conclusions. We still haven't listed the human senses yet so knowing which ones a machine needs is a way off.

@Barkey: Depends on whether the AI has been instructed to pay attention to the dissenting opinions. If it has, then it will find a better compromise position (according to whatever criteria it has been programmed with) than those same humans could have come up with.

And if it hasn't been programmed to pay attention to different opinions, then those other people are just SOL.

@Michael: Actually, it just requires that the owner share his/her perceptions of events with the secretary. Over time, it will build a profile of events accurate enough for forecasting. If you kept your daily journal on this thing, that would do it.

edited 28th Jul '12 11:16:58 AM by DeMarquis

I thought the point of using an AI instead of a human was not having to pay them.

That's the point of using a computer, but you dont need AI for that.

The problem with the human sharing their perceptions is that humans lie. What is the chance of a human admitting to its AI secretary that it drank too much and spent the morning vomiting?

Edit: Or more likely the human will be lazy and report incomplete information.

edited 28th Jul '12 11:33:05 AM by Michael

Well, if you do that, you're wasting your own money.

So you end up employing a secretary to keep your AI secretary fully informed.

Not unlike hiring someone to manage your peoplesoft application.

One thing that occurred to me is that this A.I could be used by people with mental problems, from autism to dementia a little AI helper could dramatical improve peoples lives.

In the webcomic Last Resort one of the character has special glasses that help minimise her autism, something like that could potentially be used as a learning aid for such people, allowing them to notice and process thing they would normally miss and hopefully pick up good habits.

A.I used in this way could supplement existing help systems and allow these people to lead more normal lives.

I just don't like the blunt assumption that an AI automatically and always knows best. I make it a habit of constantly questioning human judgement, and I don't like some of the blind trust people say they would put into an AI in that "Well it will come up with better solutions than a human" vein.

I'm sure sometimes it would, but I'd just be paranoid about ever letting an AI make its own decisions when it comes to anything based on people.

It's the perspective people are taking.

From your perspective, from what I can tell, is that an AI will be "apply rule, someone loses job, end of story".

From an engineer's perspective it is "the AI can collect data, figure out what works better or not based on statistical analysis and then help employers form better policies inside their corporation".

Of course, if they do no such thing (improve their policies through the use of superior tools such as AI) then they won't get better. But then, they'll lose out to businesses that do handle their workforce in a superior fashion. Or at least, we hope so. The drive to have the American-style "yes/no" rejection of candidates in employment has to do with employers not bothering with facts.

Well I guess as an upside, an AI couldn't decide not to call me for an interview because I'm a National Guardsman.

Unless that was programmed in, which would be a felony.

To prove it you'd have to be able to decompile the program, and the messes that makes are so ungainly they could pass it off as a glitch anyhow.

Hell, even if you got your hands on the source code (trade secret), they'd just keep it obfuscated and uncommented for the sake of deniability. I mean if you're curious how crazy and impenetrable a half-decent programmer can make a simple task look, they actually have a contest for this.

edited 28th Jul '12 10:35:15 PM by Pykrete