Anybody in here ever mention Musk said universal basic income is inevitable? Just last year, even.
what happens when AI replaces all our jobs?
- Thread starter sleewell
- Start date
This site may earn a commission from merchant affiliate links like Ebay, Amazon, and others.
KentBrockman
Bite back with Kent Brockman!
I’m a structural engineer. I design bridges for a living and manage other engineers, so that will be the context of my comment.
Structural analysis software that started getting more usage in the 90s and onwards didn’t eliminate structural engineering jobs, but it did change the nature of the job - the analysis became easier and things that were once pretty complicated became a case of inputting into software, getting results out, and verifying the output. You cannot forget that last step ever! You cannot just plug stuff into a machine and blindly accept its results.
One other thing that has been happening is that we are continuing to learn and the codes are getting more complicated as a result. The bridge design code in the US used to be maybe 100 pages. Now, it is closer to 2000 pages and that doesn't even include the supplemental codes for seismic design, isolation bearings, etc. In addition, each state has its own design requirements. So as the analysis got easier but the design requirements got more rigorous. There is software out there that can do both the analysis and SOME of the detailed design (such as choosing plates for steel plate girders which is generally an easier task to do yourself anyway), but it's usually not great. It will output results that are believed to be safe, but the results are weird and not as optimized as we would have liked...but you will get close enough for you to take it over from there. Remember, in any case, an engineer needs to check it.
And where is the AI getting the information for it to learn? Most books end up being outdated or not having it done in the way the specific DOT wants it and a lot of the examples I use are not publicly shared - they are made either by me or other engineers that I work with. They don't get uploaded to Libgen or whatever.
Also, no one wants a computer to determine a bridge layout, explain the permitting implications, traffic implications, (all of which are ever changing) either. Those are things with so many different opinions and there is not "right" solution. I'm sure an AI can summarize the pros and cons of them, but by the time you write the input prompt for the AI, you've probably figured it out yourself anyway. That's probably true for a LOT of complicated tasks.
The industry, especially on the transportation side), is very cautious about new technology. I get plenty of emails and calls from software vendors that want to tell me how amazing their new software package is and how it can automate design and analysis. Often, these packages are expensive and, more importantly, are a giant pain in the ass to use (usually, we like the analysis tools but aren't hugely interested in the design capabilities). We are struggling to automate a lot of the seemingly straight forward things - e.g. my DOT is struggling with automating some CAD linework for a simple span bridge.
Most importantly, an engineer has to stamp the work per state laws. I don't know a single bridge engineer that would be willing to stamp some automated output without it being checked themselves. Our companies' insurance policy covers professional liability insurance (PLI) for the engineers, but I am skeptical that any PLI provider would want to cover an AI instead of a human. That doesn't mean it cannot be used as a tool, just like regular finite element analysis is used as a tool - the question is whether the engineer has verified the output and is willing to take a responsibility for it. If I were a client, I would not hire a firm who just assigned the project to an AI to do the job. I would want the assurance from someone willing to take legal responsibility for the output from the automation tools (AI or not). That is consistent with what is happening right now - tools are used to make the job more efficient, but an engineer has to be willing to put their stamp on it.
Long story short, I think the job will change in the same way it's been changing over the past few decades.
Structural analysis software that started getting more usage in the 90s and onwards didn’t eliminate structural engineering jobs, but it did change the nature of the job - the analysis became easier and things that were once pretty complicated became a case of inputting into software, getting results out, and verifying the output. You cannot forget that last step ever! You cannot just plug stuff into a machine and blindly accept its results.
One other thing that has been happening is that we are continuing to learn and the codes are getting more complicated as a result. The bridge design code in the US used to be maybe 100 pages. Now, it is closer to 2000 pages and that doesn't even include the supplemental codes for seismic design, isolation bearings, etc. In addition, each state has its own design requirements. So as the analysis got easier but the design requirements got more rigorous. There is software out there that can do both the analysis and SOME of the detailed design (such as choosing plates for steel plate girders which is generally an easier task to do yourself anyway), but it's usually not great. It will output results that are believed to be safe, but the results are weird and not as optimized as we would have liked...but you will get close enough for you to take it over from there. Remember, in any case, an engineer needs to check it.
And where is the AI getting the information for it to learn? Most books end up being outdated or not having it done in the way the specific DOT wants it and a lot of the examples I use are not publicly shared - they are made either by me or other engineers that I work with. They don't get uploaded to Libgen or whatever.
Also, no one wants a computer to determine a bridge layout, explain the permitting implications, traffic implications, (all of which are ever changing) either. Those are things with so many different opinions and there is not "right" solution. I'm sure an AI can summarize the pros and cons of them, but by the time you write the input prompt for the AI, you've probably figured it out yourself anyway. That's probably true for a LOT of complicated tasks.
The industry, especially on the transportation side), is very cautious about new technology. I get plenty of emails and calls from software vendors that want to tell me how amazing their new software package is and how it can automate design and analysis. Often, these packages are expensive and, more importantly, are a giant pain in the ass to use (usually, we like the analysis tools but aren't hugely interested in the design capabilities). We are struggling to automate a lot of the seemingly straight forward things - e.g. my DOT is struggling with automating some CAD linework for a simple span bridge.
Most importantly, an engineer has to stamp the work per state laws. I don't know a single bridge engineer that would be willing to stamp some automated output without it being checked themselves. Our companies' insurance policy covers professional liability insurance (PLI) for the engineers, but I am skeptical that any PLI provider would want to cover an AI instead of a human. That doesn't mean it cannot be used as a tool, just like regular finite element analysis is used as a tool - the question is whether the engineer has verified the output and is willing to take a responsibility for it. If I were a client, I would not hire a firm who just assigned the project to an AI to do the job. I would want the assurance from someone willing to take legal responsibility for the output from the automation tools (AI or not). That is consistent with what is happening right now - tools are used to make the job more efficient, but an engineer has to be willing to put their stamp on it.
Long story short, I think the job will change in the same way it's been changing over the past few decades.
gabito
Advanced power chords user
Well, I don't know, I think it's working very well so far.
MFB
Banned
wake up babe, new rambo avatar unlocked
wheresthefbomb
noise wizard
- Joined
- Jul 30, 2013
- Messages
- 7,266
- Reaction score
- 14,535
MFB
Banned
If the Kawasaki robot is to be believe - and I'm aware of how big an IF that is - we're getting dangerously close to Horizon Zero Dawn style machines
wankerness
Well-Known Member
MFB
Banned
Meet me at the sit-down pizza hut buffet (I'll be wearing the SJ Sharks starter jacket), first round of TMNT Turtles in Time is on me
narad
Progressive metal and politics
Seems like the only possible option as an increasing percentage of jobs become automated. But it seems like something Musk would regurgitate without actually believing, or liking when not considering where that money comes from, being so overly capitalist.
It seems like a job that's pretty automatable. I think the misconception here is about "where is the AI getting the information for it to learn". The ability to read comes from the huge amount of pre-training, and that data will often be out-of-date to some respect. That's the "throw infinite data and GPUs at it" part of training. But you could just feed in (at test time, in the context of the prompt) the DOT standards and examples of previous designs, and that will be used in generating the design.
Probably the safest aspect is just that human liability, but if someone is just verifying an AI-generated bridge design meets DOT requirements or state laws, then that's not really the same job is it?
Church2224
Guitar Whore
I will travel back in time to ensure rhe survival of Sarah Conner in 1984.
Food, healthcare, and housing? Nope. We need trans kids in their own bathrooms. Vaccine safety warnings. And tax cuts. Those things aren’t that hard so we need government spending cuts too. It’s basic really. The debt is too big to afford a military. Hell I should run for office. I got this.
CanserDYI
Yeah, No, Definitely.
They did it. They finally fucking did it. They made The Horse 2.
wheresthefbomb
noise wizard
- Joined
- Jul 30, 2013
- Messages
- 7,266
- Reaction score
- 14,535
Horsinator 2: Judgement Day
narad
Progressive metal and politics
The lame thing is that they didn't fucking do it. It's like concept art that they "unveiled" with a physical version of it that doesn't move... Given what chinese robotics companies are doing these days, this is like children playing.
AwakenTheSkies
Life is like a box of chocolates
I watched a bit of a Rick Beato video about AI, since that seems to be all the rage now.
Scroll through the comments and everyone’s either alarmed or sad or telling others to get with the times, same old shit.
I really don’t get it. First of all as of now it still sounds kind of trash, but as they say it will evolve and you won’t be able to tell the difference eventually. (same as with the amps and modellers huh?
)
But if you’re a musician, building your song from the ground up gives you a level of control that writing to an AI can never give you.
Basic stuff, like changing a snare, reamping a guitar, changing parameters on an effect, fucking changing a fill. How would you go about doing that on an AI generated track?
It seems like something that would be useful for someone who has no fucking clue what they’re doing, okay, but how it this a threat or an advantage to a musician who knows their stuff?
I can see for example potential in the AI to use it as a synthesiser-like thing, because it comes up with ridiculous sounds trying to imitate the real thing, but other than that idk.
Scroll through the comments and everyone’s either alarmed or sad or telling others to get with the times, same old shit.
I really don’t get it. First of all as of now it still sounds kind of trash, but as they say it will evolve and you won’t be able to tell the difference eventually. (same as with the amps and modellers huh?
)But if you’re a musician, building your song from the ground up gives you a level of control that writing to an AI can never give you.
Basic stuff, like changing a snare, reamping a guitar, changing parameters on an effect, fucking changing a fill. How would you go about doing that on an AI generated track?
It seems like something that would be useful for someone who has no fucking clue what they’re doing, okay, but how it this a threat or an advantage to a musician who knows their stuff?
I can see for example potential in the AI to use it as a synthesiser-like thing, because it comes up with ridiculous sounds trying to imitate the real thing, but other than that idk.
Crash Dandicoot
» B E H O L D
I would love for AI to pull an Eaton 18-speed out of a derelict Peterbilt for me any day of the week 
narad
Progressive metal and politics
Same as it would be a threat or advantage to an artist who knows how to draw/paint. It has no bearing on an artist creating art for their own enjoyment. It has enormous impact on so-so artists that want to take commissions.I watched a bit of a Rick Beato video about AI, since that seems to be all the rage now.
Scroll through the comments and everyone’s either alarmed or sad or telling others to get with the times, same old shit.
I really don’t get it. First of all as of now it still sounds kind of trash, but as they say it will evolve and you won’t be able to tell the difference eventually. (same as with the amps and modellers huh?
But if you’re a musician, building your song from the ground up gives you a level of control that writing to an AI can never give you.
Basic stuff, like changing a snare, reamping a guitar, changing parameters on an effect, fucking changing a fill. How would you go about doing that on an AI generated track?
It seems like something that would be useful for someone who has no fucking clue what they’re doing, okay, but how it this a threat or an advantage to a musician who knows their stuff?
But changing a snare, reamping a guitar, changing parameters on an effect -- those are all really easy to imagine being accomplished via AI. I think regenerating some modality in a different style is something that is easily accomplished with the model architectures that already exist today, it just takes time for people develop models and interfaces tailored to specific domains, like editing audio.
