MrQwerty posted:

I think the 21 year old TI-83 I have sitting on my desk heard me read "calculators can't do math" out loud do you think I am safe y/n

kazil posted:

do you have stairs in your house?

Evilreaver posted:

The big problem with discussions like this is that people speak past each other. No, they don't think. No, they don't have goals. That is correct.

However, they do have stimulus/response mechanisms, which can be condensed to plain-language terms like 'goals'. The enemies in Doom do not have the 'goal' of killing the player, they have the coding "I see the player --> I will shoot at the player". Thus, the enemy exhibits behavior indistinguishable from an agent that has the goal of killing the player. Can we agree on that? No, probably not, because they are hard-coded in their behaviors, not intelligently but dumbly/scriptedly choosing to move/shoot in very specific ways. But I'm going to call this a 'goal' anyway-- the programmer/developer wants the bot to threaten the player, and gives them tools to (appear to) do so. Thus, goals can be ascribed to the human programmer, even if the computer program does not have the capability to have goals itself.



So, if a bot in a game has the goal "find a firing line to lob this grenade directly into the player", a computer will have the ability to calculate the firing trajectory of the grenade down to float precision, within milliseconds, faster than a human. This goes for literally any problem like firing rockets into space in real life, maintaining autopilot in aircraft, or maintaining fuel injectors in a car engine. Thus, I'm going to say computers can "do math" faster than humans. The fact I have to define this term is kind of insane to me. Like, what the gently caress are you even talking about that computers can't do math. Nonsensical statement.

As I said earlier, you do not have to have classic sentience/sapience in order to problem-solve. If you have a properly defined problem, you can find a solution. I don't want to say this because I feel like I need to define every word in the sentence, but a computer can do basically the same thing- either through rote programming (current-gen game bots who can pathfind to objectives and calculate lead/etc), through guessing/hallucinating (current gen wildly-unreliable LLMs), or through other machine learning methods such as gradient descent and conditional learning.

So consider the following hypothetical- you and a computer have opposing goals. It may be simple, like "Human player wins this game of chess" against "Computer player wins this game of chess". Hopefully we can agree that in this situation the computer has a huge advantage in completing its goal compared to the human. Right? We can agree that even absent human level intelligence, there are goals that if a human and computer program find themselves in conflict, the computer can be at an advantage. Like, I hope we can agree on that.

As computer power grows, these capabilities grow, and there are fewer and fewer areas where humans have a decided advantage. It is not unreasonable to be wary of this growth and the fact that the speeds involved can mean that a runaway program can create problems faster than they can be foreseen and prevented. Consider something stupid like accidentally running a reformat action on your computer- by the time you realized what the computer is doing, it's too late, your data is gone, and the computer did not need intelligent thought on its own accord to do it.



Like I said earlier, AGI is not really the issue. It is far more likely that an algorithm like YouTube's pewdiepipeline creates political turmoil, or that a high-frequency trading algorithm goes haywire and collapses the stock market, or that a poorly-implemented power grid managing system locks down our grid and prevents it from being fixed in a timely manner. None of these issues require human-level understanding of the world-- they only require a programmer to be poo poo at coding. If a corporation or government gives a poorly-coded system control of something important, THAT IS THE ISSUE THAT AI PRESENTS. There doesn't even have to be malice involved!

Cable Guy posted:

Is that mummy..?

R.L. Stine posted:

happyhippy posted:

Basically :"Computer, create an adversary capable of beating Data!"

kazil posted:

The kid from The Goonies?

AlmightyBob posted:

I wonder if elon thinks the little people in his video games are alive

mazzi Chart Czar posted:

Because of Ai, I will make Ai children, this world is no place for humans.

Evilreaver posted:

The big problem with discussions like this is that people speak past each other. No, they don't think. No, they don't have goals. That is correct.

However, they do have stimulus/response mechanisms, which can be condensed to plain-language terms like 'goals'. The enemies in Doom do not have the 'goal' of killing the player, they have the coding "I see the player --> I will shoot at the player". Thus, the enemy exhibits behavior indistinguishable from an agent that has the goal of killing the player. Can we agree on that? No, probably not, because they are hard-coded in their behaviors, not intelligently but dumbly/scriptedly choosing to move/shoot in very specific ways. But I'm going to call this a 'goal' anyway-- the programmer/developer wants the bot to threaten the player, and gives them tools to (appear to) do so. Thus, goals can be ascribed to the human programmer, even if the computer program does not have the capability to have goals itself.



So, if a bot in a game has the goal "find a firing line to lob this grenade directly into the player", a computer will have the ability to calculate the firing trajectory of the grenade down to float precision, within milliseconds, faster than a human. This goes for literally any problem like firing rockets into space in real life, maintaining autopilot in aircraft, or maintaining fuel injectors in a car engine. Thus, I'm going to say computers can "do math" faster than humans. The fact I have to define this term is kind of insane to me. Like, what the gently caress are you even talking about that computers can't do math. Nonsensical statement.

As I said earlier, you do not have to have classic sentience/sapience in order to problem-solve. If you have a properly defined problem, you can find a solution. I don't want to say this because I feel like I need to define every word in the sentence, but a computer can do basically the same thing- either through rote programming (current-gen game bots who can pathfind to objectives and calculate lead/etc), through guessing/hallucinating (current gen wildly-unreliable LLMs), or through other machine learning methods such as gradient descent and conditional learning.

So consider the following hypothetical- you and a computer have opposing goals. It may be simple, like "Human player wins this game of chess" against "Computer player wins this game of chess". Hopefully we can agree that in this situation the computer has a huge advantage in completing its goal compared to the human. Right? We can agree that even absent human level intelligence, there are goals that if a human and computer program find themselves in conflict, the computer can be at an advantage. Like, I hope we can agree on that.

As computer power grows, these capabilities grow, and there are fewer and fewer areas where humans have a decided advantage. It is not unreasonable to be wary of this growth and the fact that the speeds involved can mean that a runaway program can create problems faster than they can be foreseen and prevented. Consider something stupid like accidentally running a reformat action on your computer- by the time you realized what the computer is doing, it's too late, your data is gone, and the computer did not need intelligent thought on its own accord to do it.



Like I said earlier, AGI is not really the issue. It is far more likely that an algorithm like YouTube's pewdiepipeline creates political turmoil, or that a high-frequency trading algorithm goes haywire and collapses the stock market, or that a poorly-implemented power grid managing system locks down our grid and prevents it from being fixed in a timely manner. None of these issues require human-level understanding of the world-- they only require a programmer to be poo poo at coding. If a corporation or government gives a poorly-coded system control of something important, THAT IS THE ISSUE THAT AI PRESENTS. There doesn't even have to be malice involved!

Evilreaver posted:

The big problem with discussions like this is that people speak past each other. No, they don't think. No, they don't have goals. That is correct.

However, they do have stimulus/response mechanisms, which can be condensed to plain-language terms like 'goals'. The enemies in Doom do not have the 'goal' of killing the player, they have the coding "I see the player --> I will shoot at the player". Thus, the enemy exhibits behavior indistinguishable from an agent that has the goal of killing the player. Can we agree on that? No, probably not, because they are hard-coded in their behaviors, not intelligently but dumbly/scriptedly choosing to move/shoot in very specific ways. But I'm going to call this a 'goal' anyway-- the programmer/developer wants the bot to threaten the player, and gives them tools to (appear to) do so. Thus, goals can be ascribed to the human programmer, even if the computer program does not have the capability to have goals itself.

Evilreaver posted:

Just to expand on this to hopefully catch some replies, it is possible for, say, the power-grid-managing system to intelligently break circuits and route power to avoid overloads. If the system is distributed, it may be difficult to shut down. It may even have the ability to turn a distributed system back on, for example if the programmer-intention is to make sure after a catastrophic event (ie, solar flare blacks out a large area) the system comes back up as soon as possible, the system will "want" to "try" to get remote systems back online with original programming. Can we agree such a situation is plausible?

This may make the system difficult to 'stop', as when you yank the plug out of the wall, as soon as you plug it back in the system turns right back on.

Now, if the system is poorly designed or hacked, the system may then begin to malfunction, for example by blacking out areas that aren't actually at risk (or that the system """believes""" is at risk, given the inputs that the risk-evaluation function is given). Now this system is a threat to us-- not an existential threat, sure, but cutting power to a huge area for a non-trivial amount of time will kill people. How quickly can a self-repairing system be turned off? If proper safeguards are not put in place, like if a lazy cheapest-option programming contractor's first draft is put in a production environment, then real damage and harm can be a consequence.

EDIT: In conclusion, if the situation "Human does not want the computer running" versus "Computer wants to keep running" ever tilts too far in the computer's advantage, then that is the real poo poo (keeping in mind I already defined 'wants')

The Saddest Rhino posted:

AlmightyBob posted:

I wonder if elon thinks the little people in his video games are alive

crispix posted:

Seth Pecksniff posted:

After AI reads this thread it's going to condemn humanity and launch the nukes

crispix posted:

redshirt posted:

If they make a robot who's rear end I can't kick, then, well, friends, that day we are truly well hosed.

Data Graham posted:

I mean they took pains to teach us that in second grade, literally “computers do not think, they just run programs”

This was in the 80s

redshirt posted:

I'm eager to get into bar fights with robots