Dex-chan lover
- Joined
- Dec 26, 2018
- Messages
- 569
Just curious about MD users opinion. I'm not asking here about "consciousness" or whether it will be "human", just pure utilitarian ability. Personally, I'm giving us 2 years tops.
How many years until AI takeover?
- Thread starter Niewaznekto
- Start date
Dex-chan lover
- Joined
- Nov 14, 2019
- Messages
- 1,297
Ai will never takeover, it's just bullshit. imo
Ai can not do certain things humans can do, it'll take many more years or a decade to "takeover".
Ai can not do certain things humans can do, it'll take many more years or a decade to "takeover".
Dex-chan lover
- Joined
- Nov 18, 2018
- Messages
- 10,133
I don't really know about when, but I'm hoping it's after my lifetime. 
Dex-chan lover
- Joined
- Dec 26, 2018
- Messages
- 569
yeah, i don't get people who are cheering for it, I like life, why the hell would I want non-humans to take overI don't really know about when, but I'm hoping it's after my lifetime.
Dex-chan lover
- Joined
- Dec 26, 2018
- Messages
- 569
Every day it seems the number of things it can't do to be dropping, and personally at least I see no logical reason why it couldn't
Dex-chan lover
- Joined
- Nov 14, 2019
- Messages
- 1,297
I am confused by this first part "it can't do to be dropping"...? Could you rephrase that a bit ^^"
Supporter
- Joined
- Apr 26, 2020
- Messages
- 2,536
Depends on what you mean by 'take over.' (Text wall incoming.)
We have AI that can win against the best human chess and go players, every time; those are pretty small universes to operate in. We have AI that can ingest large amounts of preexisting art or statistical data and spit out something to match a prompt for a 'new artwork' or correlative tables; that's a pretty narrow range of operation. It isn't that hard at this point to create a narrow AI that can do one thing well, within a well-defined set of parameters. I expect that AI will continue to get better at writing code, and term papers, and ordering produce for the local grocery store, and auditing taxes, and a bunch of other individual tasks. And eventually large swathes of these more mundane things will be handled by narrow AI.
We don't have AI that can operate a car safely nearly as well as human drivers (even the young ones). Yes, it can handle basic things like keeping between the lines and navigating under normal circumstances. The problem comes when it's presented with an edge case, particularly one it's never seen before - someone flipping a traffic cone onto the hood, or coming up behind a vehicle being towed backwards, or emergency vehicles trying to navigate around it, or road closures, etc. Then it seizes up, or otherwise reacts inappropriately for the situation. Admittedly, so do some human drivers, but a human can look around and make a decision how to deal with a situation on the fly in a way the AI can't, and is much more likely to be able to take direction from a cop or some other traffic control officer.
And that's the second biggest problem with AI - being able to respond correctly to 95% or 99% or 99.9% of situations is simply unacceptable in some applications, particularly when operating in meat space around humans. As a human driver of (coming up on) 30 years, I've been involved in exactly five accidents. Two were clearly my fault (one following too closely, one pulling out in front of cross traffic I didn't see); one was nominally my fault (driving over a downed power pole in a thunderstorm - I didn't see it because the streetlights were out due to, y'know, downed power pole); the last two were another driver's fault (one rear-ended due to them following too closely, and one front-end collision due to them blowing through a stop sign - for that one, I got slowed down enough to keep the damage to light body work). That's at most five momentary lapses, each resulting in an accident I could walk away from, in... maybe eleven thousand hours of driving (an hour a day for 30 years)? Never had an issue with knowing how to handle emergency vehicles; never driven off the road because I freaked out behind a tow truck; always been able to get out of traffic when a mechanical failure disables the car (even if I had to jump out and push).
The biggest problem with AI is that, even if we solve the above issues and can get to human-plus levels of accuracy within the field of narrow AI, it's still going to be a set of discrete applications. A general AI model - one that can choose which narrow AI application to use in every situation, and that can interface with people in natural language - is a challenge that has been worked on since the late '50's, and still seems quite some distance off from being able to pass the Turing test. Meat space adds an additional level of complexity to this - not only does the AI need to avoid people and other AIs, but it has to predict how someone will react to it (think about the awkward dance of two people trying to pass in a hallway, when both of them go the same direction initially).
So if the question is 'will AI take on certain tasks in the next X years?' Yes, yes it will. But if the question is 'will AI be handling all the jobs and driving us around and flying our planes and making our food and washing our clothes and generally letting all humanity live a life of 24/7 leisure?' Not anytime soon.
We have AI that can win against the best human chess and go players, every time; those are pretty small universes to operate in. We have AI that can ingest large amounts of preexisting art or statistical data and spit out something to match a prompt for a 'new artwork' or correlative tables; that's a pretty narrow range of operation. It isn't that hard at this point to create a narrow AI that can do one thing well, within a well-defined set of parameters. I expect that AI will continue to get better at writing code, and term papers, and ordering produce for the local grocery store, and auditing taxes, and a bunch of other individual tasks. And eventually large swathes of these more mundane things will be handled by narrow AI.
We don't have AI that can operate a car safely nearly as well as human drivers (even the young ones). Yes, it can handle basic things like keeping between the lines and navigating under normal circumstances. The problem comes when it's presented with an edge case, particularly one it's never seen before - someone flipping a traffic cone onto the hood, or coming up behind a vehicle being towed backwards, or emergency vehicles trying to navigate around it, or road closures, etc. Then it seizes up, or otherwise reacts inappropriately for the situation. Admittedly, so do some human drivers, but a human can look around and make a decision how to deal with a situation on the fly in a way the AI can't, and is much more likely to be able to take direction from a cop or some other traffic control officer.
And that's the second biggest problem with AI - being able to respond correctly to 95% or 99% or 99.9% of situations is simply unacceptable in some applications, particularly when operating in meat space around humans. As a human driver of (coming up on) 30 years, I've been involved in exactly five accidents. Two were clearly my fault (one following too closely, one pulling out in front of cross traffic I didn't see); one was nominally my fault (driving over a downed power pole in a thunderstorm - I didn't see it because the streetlights were out due to, y'know, downed power pole); the last two were another driver's fault (one rear-ended due to them following too closely, and one front-end collision due to them blowing through a stop sign - for that one, I got slowed down enough to keep the damage to light body work). That's at most five momentary lapses, each resulting in an accident I could walk away from, in... maybe eleven thousand hours of driving (an hour a day for 30 years)? Never had an issue with knowing how to handle emergency vehicles; never driven off the road because I freaked out behind a tow truck; always been able to get out of traffic when a mechanical failure disables the car (even if I had to jump out and push).
The biggest problem with AI is that, even if we solve the above issues and can get to human-plus levels of accuracy within the field of narrow AI, it's still going to be a set of discrete applications. A general AI model - one that can choose which narrow AI application to use in every situation, and that can interface with people in natural language - is a challenge that has been worked on since the late '50's, and still seems quite some distance off from being able to pass the Turing test. Meat space adds an additional level of complexity to this - not only does the AI need to avoid people and other AIs, but it has to predict how someone will react to it (think about the awkward dance of two people trying to pass in a hallway, when both of them go the same direction initially).
So if the question is 'will AI take on certain tasks in the next X years?' Yes, yes it will. But if the question is 'will AI be handling all the jobs and driving us around and flying our planes and making our food and washing our clothes and generally letting all humanity live a life of 24/7 leisure?' Not anytime soon.
Dex-chan lover
- Joined
- Jan 8, 2023
- Messages
- 1,933
people are too lazy my man. i talk from my experience being an art graduate. the majority of the ai image generating people, hasnt even tried to learn to draw anything. ask them about color theory and they know sht. people are just running around trying to find ways to earn money without putting in any effort.
Dex-chan lover
- Joined
- Oct 10, 2018
- Messages
- 1,150
Its just fears peddled by American Hollywood. Like a lot of things peddled by Americans, its horse shit.
Dex-chan lover
- Joined
- Jan 25, 2023
- Messages
- 9,820
When this site almost becomes reddit, it will be AI chatbots instead of real humans.
Dex-chan lover
- Joined
- Jan 25, 2023
- Messages
- 9,820
Tbf you cannot distinguish a person from a bot in any forum.
First bots do simple things like emojis, then they copy posts from the past.
Then they shitpost or post random links that is taken down in 2 years.
Supporter
- Joined
- Apr 26, 2020
- Messages
- 2,536
I have yet to see any bot that can perform quality shitposting at the level of the GFG.
(Paging @sterven , @bigtiddyoneesan , and @pandascepter to come administer a shitposting Turing test. I'm starting to think @TestAddress is a bot.)
Dex-chan lover
- Joined
- Jan 25, 2023
- Messages
- 9,820
Turing test is a myth.
Group Leader
- Joined
- Jul 15, 2019
- Messages
- 7,478
What is Turing?
Supporter
- Joined
- Apr 26, 2020
- Messages
- 2,536
Group Leader
- Joined
- Jul 15, 2019
- Messages
- 7,478
cool, but dw im not a bot
Supporter
- Joined
- Jan 19, 2018
- Messages
- 4,467
Dex-chan lover
- Joined
- Mar 17, 2019
- Messages
- 9,815
gif credit to halo
while we are all human and still retain most of our humanity, there are rumors that haikubro aims to create, maybe even become, a haikubot in the future. will he succeed? only time will tell
Similar threads
