Comic Discussion > QUESTIONABLE CONTENT

AI Rights

<< < (11/12) > >>

Is it cold in here?:
How does Bubbles's situation fit into this?

Morituri:

--- Quote from: bicostp on 02 Jan 2012, 08:19 ---But are those phone menu systems really 'intelligent'? They're essentially talking flowcharts with semi-accurate speech recognition. They leave all the logic up to the human on the phone.

--- End quote ---

See, here's the thing; you can't program a computer for semi-accurate speech recognition.  All the systems that pick words out of speech with any kind of decent usable accuracy are copies of things that would not exist if it weren't for machine learning algorithms.  And a fair number of them learn online; if enough people with an accent which it can *barely* recognize keep calling, it'll adapt and start understanding them better.  A call center that serves a Latino community is likely to start recognizing a Spanish accent with better accuracy.  Which still sucks if you're in the small percentage of people whose first language was Cantonese and the system's not adapting to you, but that speech recognition IS an AI-driven function, even if all the logic of what you're actually doing and how the company has set up its system is all mapped out by humans.

We're using AI learning algorithms already to fill in all kinds of gaps in our ability to program.  You think some Engineer at google taught their photo search system how to recognize dogs and tennis players?  Nope.  You think the systems that do real-time load balancing and prevent transformers from exploding in lightning storms learned exactly what they have to do, which is different depending on how many microseconds down the line every component happens to be,  from a human?  Nope.  You think the systems that coordinate, synchronize, and adjust traffic lights in every major city to make traffic flow as smoothly as possible are  some static program that has to get updated every time a traffic light gets installed because the patterns at that particular corner are different, or because the patterns change over time?  Nope.  If New York went back to timed traffic lights the city would self-destruct; they've got more cars on the same streets now than they had back when cars were getting stuck in gridlock for days in the 1970s. 

There are many, many pieces of basic infrastructure, from ad servers to waste processing, that contain bits and chunks we could not possibly have programmed.  These things aren't human-level AI, no.  Some of them are smarter than cockroaches.  But they're ubiquitous.

Morituri:

--- Quote from: techkid on 12 Jan 2012, 06:43 ---So looking at that, how would AI be configured to run? Whether it ends up being an executable or a file, compatibility would still be an issue. If it runs on its own, then the problem you face would be operating system (and possibly file system) support, and if it runs through an intermediary program, then system compatibility might not be so much of a problem, but the program will have to be upgraded and file compatibility is not always guaranteed.

--- End quote ---

My anticipation is that a lot of the early ones are going to get the ability to interact with unix command-line shells.  It's easy to interface, it's powerful, and its idioms are discoverable. 

Controlling machines in real time though would be either through some different virtualized interface, or by direct use of the runtime environment like any other program (Or possibly more like an operating system than a program).

With the ability to rapidly adapt to very different hardware (bank server to human-form robot to submarine to jet fighter) and benefit from CPU and other hardware upgrades?  I'd be betting on a virtualized interface of some kind.  The Bash shell, for example, runs on ALL kinds of hardware and knowing how to handle it remains useful in spite of different hardware choices, kernel upgrades, runtime library mods, and OS changes.  Some kind of crazy-advanced analogue of that would be a standard way for AIs to interface with all kinds of hardware. It would be like you know how to drive a car and when you sit down in a truck or a farm tractor or a cotton picking machine.  It's different and it handles differently and there are some levers over there that do things the car couldn't do, and some things the car could do that this machine can't, and maybe you have to handle gear shifting differently ... but the steering wheel is still a steering wheel, the gas pedal still accelerates, the brake pedal still slows down, the gauges on the panel that show things like speed and RPM or whatever still mean the same thing, and within a few hours of taking things slow and careful, you get used to it.

Storel:

--- Quote from: Morituri on 13 Jan 2017, 12:22 ---We're using AI learning algorithms already to fill in all kinds of gaps in our ability to program.  You think some Engineer at google taught their photo search system how to recognize dogs and tennis players?  Nope.  You think the systems that do real-time load balancing and prevent transformers from exploding in lightning storms learned exactly what they have to do, which is different depending on how many microseconds down the line every component happens to be,  from a human?  Nope.  You think the systems that coordinate, synchronize, and adjust traffic lights in every major city to make traffic flow as smoothly as possible are  some static program that has to get updated every time a traffic light gets installed because the patterns at that particular corner are different, or because the patterns change over time?  Nope.  If New York went back to timed traffic lights the city would self-destruct; they've got more cars on the same streets now than they had back when cars were getting stuck in gridlock for days in the 1970s.

--- End quote ---

Wow. The traffic light example reminded me somehow of a computer game called SimTower back in the '90s; you had to build a skyscraper floor-by-floor, laying out the infrastructure as you went, finding different tenants (retail, offices, apartments), and one of the biggest deals was placing the elevator shafts and programming the elevators to handle the traffic at different times of day. After reading what you said, I'd be willing to bet money that modern elevators now use AI learning algorithms to reprogram themselves just like traffic lights -- or if they don't yet, they damn well should.

Thrudd:
In the Metropolitan area next to the residential city I live in, the opposite is true. I have encountered a lot more smart elevator systems than traffic signals. Forget about integrated, since city officials over-ride the engineers since most of the council is anti-car and don't have more than a few neurons amongst the lot of them. They also bring in outside consultants, quote a dew from the states, to analyse the issues and provide suggestions to be ignored. The sad but laughable thing is most of these specialists were students of a Master on the subject who happens to work at a local university. That person has never been spoken to or even acknowledged since he was just some local professor on the subject. Politicians should be damn glad the populace has a thick skin and isn't as hands on as they were a century back, else there would be a hands on with tar feathers and rope as well the the occasional pitchfork and torch.

Navigation

[0] Message Index

[#] Next page

[*] Previous page

Go to full version