Bay 12 Games Forum

Please login or register.

Login with username, password and session length
Advanced search  
Pages: 1 ... 3559 3560 [3561] 3562 3563 ... 11056

Author Topic: Things that made you go "WTF?" today o_O  (Read 14853206 times)

alway

  • Bay Watcher
  • 🏳️‍⚧️
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53400 on: July 06, 2014, 11:46:18 pm »

Right, so AI.

Game AI is terrible, and doesn't represent the state of modern AI programming. The goal of a game AI is not to win. The goal of a game AI is to provide a reasonable challenge for a player to overcome, and do so with as tiny a computational budget as possible while taking as little developer time/dollars as possible to create it. This is why early game AIs, especially in strategy games, were allowed to cheat. They aren't playing to win or some similarly navel-gazing objective like that, they exist to give the player a challenging experience. You generally can create game AIs better than any player which learn and adapt over time. It's just that nobody cares, because that takes a larger amount of time (Money) better used elsewhere, takes a larger amount of computational power better used elsewhere, and the end result is a game AI which gives players a shitty experience that will make them go play something else.

This is why game AI uses simple scripts. You're supposed to be able to learn its moves and countermoves, and thus gain mastery over the system and beat it. Also because it is highly predictable (easy to debug on a short deadline), can be tweaked by designers via config files, and takes very little time to implement.

History of AI:
But all of those are just ancient techniques from before the first AI winter. Initially, there was optimism that such state machines with their 'if x do y' tables could create powerful AI; back at the very dawn of the field half a century ago. Of course, then came notions like 'combinatorial explosion' which is where most people's modern misconceptions about AI come from (the whole 'the just read scripts' thing seen in this thread). At that point, it became very clear that such techniques could not, in fact, do much of anything. And so came the AI winter (1970s). A time when research funding sort of dried up and the field was much more subdued.

Then came the real progress that led up the the field of today. The earlier problems were beginning to be addressed with new techniques. The theory behind Neural Networks came about; a technique which attempted to solve problems through learning algorithms based on the workings of neurons in a brain. Algorithms which could be trained to act on imprecise, noisy, and previously unseen information. Likewise, the idea of 'embodied cognition' came into popularity; the belief that the best path to powerful AI was to simply have it explore the world around it, figuring out things as it went. So that was the 80's.

At this point it must be mentioned that at this time, there were two main camps in the AI field; referred to loosely as the 'messies' and the 'neats.' The messies were all about things like neural networks; essentially believing that powerful AI would be achieved through use of algorithms which worked, but whose full functioning couldn't really be understood. In a neural network, you can train it to do something, but picking apart how it does it is effectively as difficult as figuring out someone's thoughts by watching their neurons firing. You get some general ideas, but not a comprehensive, detailed version. The neats, on the other hand, wanted more mathematical rigor, essentially believing that powerful AI would be such a complicated system that a messy approach would end up being too difficult to get anything useful out of. As is obvious by this point, the two sort of went back and forth for a while in the field. The neats pretty much owned the early days, as simple models were more powerful when constrained to 60s era computational power. The messies became more popular after the events and lessons leading up to the AI winter brought the previous efforts of the neats to naught (well, aside from game AI, where the 60s era AI is still very much in vogue today).

Then came the 90s, when everything changed in a big way, creating the modern field of AI. Statistical and economic theory were brought in, resulting in "The Victory of the Neats." Statistical reasoning, with roots in things like Bayes' Rule for updating beliefs, became central to the field of AI. This gave not only the ability to figure out information from a set of data, like the neural networks had done a decade earlier, but to do so in a way which was both mathematically rigorous for which error values could easily be calculated, but which were also fully transparent. Suddenly, you could not only prove that your algorithm would actually work, but you could even say exactly why it was working. You now had things like Bayesian Networks, which could not only learn like neural networks, but through virtue of its internal transparency, you could now easily create AI which could reason about their own reasoning. From economic theory came the idea of the utilitarian "intelligent agent" or the "ideal rational agent," which gave a firm foundation to the theory behind how an AI should act and react to its environment; including unknown factors it had previously been unaware of. Thus, all the benefits of the techniques used by the 'messies' were incorporated into the modern, comprehensive techniques of the 'neats,' while still fitting within an overall framework of mathematical rigor and provability espoused by the neats. They can learn, reason, understand natural language, and investigate. They can even learn and reason about their own ability to learn and reason!

This has been the paradigm over the past 20 years. The result? Self driving cars. Search engines which categorize and find information from exabytes of data on the web. An AI which beat the human champions of Jeopardy, went on to become better at diagnosing some illnesses than doctors and invent cooking recipes. A stock market which is increasingly run by financial corporations' AI. A multibillion dollar industry of information brokers trading anonymized information for companies to plug into their marketing algorithms. And so on, pervading more or less subtly in nearly every aspect of our lives.

Not that you would know any of this from the media. Typical descriptions of AI are still your ridiculous 60's era AI, even in much of sci-fi. Which is why things like Watson or self driving cars have come as such a shock to people. The general public isn't aware of just how far things have come, and how fast things are going now. And it's actively to their detriment, as they simply can't understand contemporary events like the news about the NSA without such knowledge of where AI is at... and just how utterly intense its benefits and dangers are in the modern world.

As for AI consciousness... well, they already think about thinking and reason about reasoning, so there's that. The Watson AI, for example, had hundreds of individual learning algorithms (look up IBM's videos about Watson for details; they're definitely worth watching, especially the 45+ minute long ones which go really in detail) at the lower levels. Other algorithms were keeping track of how well each of these 'thought processes' worked for varying categories of questions (of course, the categories themselves had to be figured out too, so it's definitely non-trivial), and adjusted how much Watson trusted each of these for different questions accordingly. So that's not terribly difficult; and by that measure, it is, strictly speaking, conscious. Though, coming back to the embodied cognition idea from earlier, it's certainly a different kind of consciousness, as there is much less there to be aware of. It doesn't have millions of rods and cones, millions of pressure and heat sensing nerves, ears, hunger or thirst, or a health and body whose protection it was created to ensure.

edit: oops, I accidentally a 1300 word essay >_>
« Last Edit: July 06, 2014, 11:51:12 pm by alway »
Logged

SalmonGod

  • Bay Watcher
  • Nyarrr
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53401 on: July 06, 2014, 11:59:49 pm »

That was a great read, alway.  Thanks for putting that together.

Not that you would know any of this from the media. Typical descriptions of AI are still your ridiculous 60's era AI, even in much of sci-fi. Which is why things like Watson or self driving cars have come as such a shock to people. The general public isn't aware of just how far things have come, and how fast things are going now. And it's actively to their detriment, as they simply can't understand contemporary events like the news about the NSA without such knowledge of where AI is at... and just how utterly intense its benefits and dangers are in the modern world.

This is so frustrating to me these days.  General public awareness of technological capability is so very far behind reality, and it's hard for me to tell sometimes whether people's disbelieving reactions are just denial or genuine ignorance.
Logged
In the land of twilight, under the moon
We dance for the idiots
As the end will come so soon
In the land of twilight

Maybe people should love for the sake of loving, and not with all of these optimization conditions.

Arx

  • Bay Watcher
  • Iron within, iron without.
    • View Profile
    • Art!
Re: Things that made you go "WTF?" today o_O
« Reply #53402 on: July 07, 2014, 02:06:02 am »

RIOT


Maybe this is the strange one.



That was an interesting 1300 word essay, alway.
Logged

I am on Discord as Arx#2415.
Hail to the mind of man! / Fire in the sky
I've been waiting for you / On this day we die.

EnigmaticHat

  • Bay Watcher
  • I vibrate, I die, I vibrate again
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53403 on: July 07, 2014, 02:52:54 am »

http://en.wikipedia.org/wiki/Arimaa

For those who don't want read, its a game specifically designed so that a 4 year old could understand the rules, but a computer would have trouble winning.  There have been yearly human versus computer championships with a $10,000 for the programmer of the first computer to win, for 9 years and counting, and the prize has never been claimed.

Computers are stupid.  Believe it or not, I already knew about self driving cars and Watson, and I am still not impressed.

In particular the self-driving cars use a combination of pre-built, extremely detailed maps of the area and sensors which track distance to other cars.  They have no concept of the space around them, they ARE following a script, which is "follow this line, adjust speed based on what your sensors say".  The problem with this as an AI is that it has no concept of what any of what it does actually means.  It brings to mind a magazine article I read on one of the earlier attempts to make self-driving cars, which was based solely on cameras for sensory data.  This attempt involved having the computer record camera data from a human driver driving around, then having the computer take over.  It drove flawlessly until it reached a bridge and then it swerved off to the side and forced the human driver to take control and prevent an accident.

The reason?  It believed that keeping the color green (the grass) to the immediate right of the car was the most important thing.

To go further with it, a human looks at a room and sees walls and floors and tables.  This is because a human divides the world into objects and knows what each of those objects is.  They have nothing but the sensory perceptions of light, but they construct an entire imaginary world with three dimensional space around them, and this (as far as any human can know) matches up with the actual real space around them, which cannot be perceived by humans.  A human looks at an area of green with blue above it and divides it up into "this is grass, this is sky".

Meanwhile a computer has nothing but the sensory information.  It can be programmed (as face recognition software does, albeit badly a lot of the time) to recognize specific objects.  But this has to be done on an object by object basis, because the computer isn't actually thinking, its just doing what its programmer told it to.  So sure, you could program it to divide the world that those grey pixels on its video feed are road and the yellow bits are the line it shouldn't cross.  Then you show it at a night and will have no idea, so you have to program it to account for lighting differences.  Then take it off road, or to a dirt road, and it will again need to be reprogrammed.  This is because it will never actually understand the concept of road, because it doesn't think, it just reads from a script.

As for Watson, trivia contests are not the same thing as, say, Starcraft 2.  Building a huge database of information and then accessing it on the fly plays to a strength of computers, which is performing a series of simple tasks thoughtlessly and very, very, very quickly.  Its creating the illusion of thought using brute force and processing speed.  Try to invent a program that takes two arbitrary related facts and reaches an intelligent conclusion using them, I guarantee a supercomputer running said program will be less profound than the average twelve year old.  Because computers are fucking stupid.

Reading from a script is fundamentally how every modern computer program works.  No computer program can go beyond its script.  Sure, you could write a program that edited its own code and the relaunched itself, but that would still be limited by the ways in which it was programmed to rewrite itself.  Computers do not think.  If you believe a computer is thinking, that's because its programmer designed it to imitate a specific human behavior using smoke and mirror tricks.
Logged
"T-take this non-euclidean geometry, h-humanity-baka. I m-made it, but not because I l-li-l-like you or anything! I just felt s-sorry for you, b-baka."
You misspelled seance.  Are possessing Draignean?  Are you actually a ghost in the shell? You have to tell us if you are, that's the rule

Mech#4

  • Bay Watcher
  • (ಠ_ృ) Like a sir.
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53404 on: July 07, 2014, 03:02:25 am »

I have a LEGO Mindstorms kit that I used to build a robot car with. It could follow a black line on a piece of paper and keep going around and around in circles without making a mistake. Take it off the line and it just drove straight forwards until it hit a wall or followed a join in the floor boards.

An A.I. that can extrapolate would be quite a jump forwards me thinks.
Logged
Kaypy:Adamantine in a poorly defended fortress is the royal equivalent of an unclaimed sock on a battlefield.

Here's a thread listing Let's Players found on the internet. Feel free to add.
List of Notable Mods. Feel free to add.

hops

  • Bay Watcher
  • Secretary of Antifa
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53405 on: July 07, 2014, 03:45:22 am »

I really love how you guys are using toy AIs and crappy amateur work for your argument. It makes them very convincing.
Logged
she/her. (Pronouns vary over time.) The artist formerly known as Objective/Cinder.

One True Polycule with flame99 <3

Avatar by makowka

Xantalos

  • Bay Watcher
  • Your Friendly Salvation
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53406 on: July 07, 2014, 03:51:37 am »

I really love how you guys are using toy AIs and crappy amateur work for your argument. It makes them very convincing.
Watson and self-driving cars are crappy amateur work?
Logged
Sig! Onol
Quote from: BFEL
XANTALOS, THE KARATEBOMINATION
Quote from: Toaster
((The Xantalos Die: [1, 1, 1, 6, 6, 6]))

Mech#4

  • Bay Watcher
  • (ಠ_ృ) Like a sir.
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53407 on: July 07, 2014, 03:54:00 am »

Now now, I'm trying to present something I think is similar for people, like myself, who might understand it better with a more common comparison. These talks can get rather complicated, so presenting the same thing in an easier to understand manner is important so people don't get too lost.

The idea that A.I hasn't advanced as much as it has I would put up to that, for the average person, complex A.I doesn't factor into their life. Plus I would assume what comes to mind when you mention A.I. is more something like a robot or a house that talks to you rather than a car that can tell where it is on the road.
Logged
Kaypy:Adamantine in a poorly defended fortress is the royal equivalent of an unclaimed sock on a battlefield.

Here's a thread listing Let's Players found on the internet. Feel free to add.
List of Notable Mods. Feel free to add.

EnigmaticHat

  • Bay Watcher
  • I vibrate, I die, I vibrate again
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53408 on: July 07, 2014, 03:59:46 am »

I really love how you guys are using toy AIs and crappy amateur work for your argument. It makes them very convincing.

Its an argument about video game AI :/

Anyway, add something better.  I'm interested (not sarcasm).
Logged
"T-take this non-euclidean geometry, h-humanity-baka. I m-made it, but not because I l-li-l-like you or anything! I just felt s-sorry for you, b-baka."
You misspelled seance.  Are possessing Draignean?  Are you actually a ghost in the shell? You have to tell us if you are, that's the rule

Molly Renata

  • Bay Watcher
  • Just an obnoxious fangirl.
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53409 on: July 07, 2014, 04:00:28 am »

Of course, Weird Al is releasing probably his last studio album this month and I'm waiting for that too.

nooooOOOOOOOOoooooooooooooooooooooooooooooo.......

YOU WILL BE MISSED, WEIRD AL. ;_;
Logged
Molly Renata, also known as gyppygirl2021.
Fangirl. Beware random references to old and obscure games.
I'm a dumb hipster who plays older Dorf Fort versions. Mostly DF2010 and 40d. I also make and play a lot of silly mods, and find my own ways to have Fun.

Empiricist

  • Bay Watcher
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53410 on: July 07, 2014, 04:06:48 am »

My opinion on the matter is as follows: stereotypical "true" AI is unlikely to occur at best. At the end of the day, people generally don't NEED an AI that thinks and acts like a human, after all for all the research costs and whatnot involved, you could just train and hire someone for far less money. What people need is an AI that performs it's purpose. Even if the purpose involves maintaining a verisimilitude of humanity, it does not necessarily need human-like thoughts or cognitive processes.

Furthermore, it may be argued that intelligence, cognition and whatnot are dependent upon the hardware so to speak, the physical housing and processes in addition to the processes that created it. Brains and whatnot are nevertheless different from a computer in terms of structure and processes. Furthermore, a brain is something that arose from the random mutations and natural selection of evolution due to it helping organisms with them surviving and reproducing; or a gift from God; or some combination/compromise between the two. Comparatively, the hardware used for an AI was designed specifically for it's purpose, which generally isn't mimicking human thought.

Of course, it is entirely possible that our minds are not necessary the same as the underlying processes that govern our intellects, mere conscious shadows of some light we fail to see. I acknowledge that perhaps the consciousness of an AI may be such, that is, not the same as the programming. However if that is the case, then it is still not something we are capable of recognizing, perhaps we are too blinded by our knowledge of what dictates their actions to see any inherent thought within them? Much like someone who has seen the fire in Plato's cave is unable to believe the viewpoint of the prisoners?
Logged
Quote from: Caellath (on Discord)
<Caellath>: Emp is the hero we don't need, deserve or want

Loud Whispers

  • Bay Watcher
  • They said we have to aim higher, so we dug deeper.
    • View Profile
    • I APPLAUD YOU SIRRAH
Re: Things that made you go "WTF?" today o_O
« Reply #53411 on: July 07, 2014, 04:55:49 am »

My opinion on the matter is as follows: stereotypical "true" AI is unlikely to occur at best. At the end of the day, people generally don't NEED an AI that thinks and acts like a human, after all for all the research costs and whatnot involved, you could just train and hire someone for far less money. What people need is an AI that performs it's purpose. Even if the purpose involves maintaining a verisimilitude of humanity, it does not necessarily need human-like thoughts or cognitive processes.
AI psychiatrists and friends seems like a very useful thing for AIs to act like humans for. Oil rig divers, submariners, astronauts; being isolated can drive humans crazy, having a humanlike aboard could keep them more or less sane enough.

Empiricist

  • Bay Watcher
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53412 on: July 07, 2014, 04:59:11 am »

My opinion on the matter is as follows: stereotypical "true" AI is unlikely to occur at best. At the end of the day, people generally don't NEED an AI that thinks and acts like a human, after all for all the research costs and whatnot involved, you could just train and hire someone for far less money. What people need is an AI that performs it's purpose. Even if the purpose involves maintaining a verisimilitude of humanity, it does not necessarily need human-like thoughts or cognitive processes.
AI psychiatrists and friends seems like a very useful thing for AIs to act like humans for. Oil rig divers, submariners, astronauts; being isolated can drive humans crazy, having a humanlike aboard could keep them more or less sane enough.
Yes, however, that does not mean they have to be fully human-like in thought processes either, it is possible to mimic human-like qualities without having human thought patterns.
Logged
Quote from: Caellath (on Discord)
<Caellath>: Emp is the hero we don't need, deserve or want

hops

  • Bay Watcher
  • Secretary of Antifa
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53413 on: July 07, 2014, 06:20:03 am »

Logged
she/her. (Pronouns vary over time.) The artist formerly known as Objective/Cinder.

One True Polycule with flame99 <3

Avatar by makowka

Sergarr

  • Bay Watcher
  • (9) airheaded baka (9)
    • View Profile
Re: Things that made you go "WTF?" today o_O
« Reply #53414 on: July 07, 2014, 06:23:10 am »

My opinion on the matter is as follows: stereotypical "true" AI is unlikely to occur at best. At the end of the day, people generally don't NEED an AI that thinks and acts like a human, after all for all the research costs and whatnot involved, you could just train and hire someone for far less money. What people need is an AI that performs it's purpose. Even if the purpose involves maintaining a verisimilitude of humanity, it does not necessarily need human-like thoughts or cognitive processes.
AI psychiatrists and friends seems like a very useful thing for AIs to act like humans for. Oil rig divers, submariners, astronauts; being isolated can drive humans crazy, having a humanlike aboard could keep them more or less sane enough.
But what about AIs going insane themselves? If they can act like humans, then they can go insane, too.

Also I think that AIs on current hardware (digital values/central processing unit) are impossible, or at least would compute very, very slow.
A switch to (analogue values/decentralized computing) would help.
Logged
._.
Pages: 1 ... 3559 3560 [3561] 3562 3563 ... 11056