My AI Timelines Have Sped Up (Again)
alexirpan.com
alexirpan.com
Is the author trying to say that all that work falls into the 5% left over after you replace the stock brokers, programmers, artists, etc.?
I understand no offence is meant, but this feels extremely naive. At best the work programmers do is in the minority of "economical valuable" work that we strongly rely on for our world economies to keep chugging along.
Quick and dirty searches show that some 12% of employed people are in healthcare in the USA, 9% in leisure and hospitality, and 2% in education. 23% of people in the USA are employed in economically valuable work that is very personal, and very difficult for technology to fully replace.
Sorry if I'm rambling, I think I just did not enjoy the complete lack of mention of blue-collar workers here. I don't think I'm yet in the camp betting on AGI coming out of the woodworks in the next few decades.
Accelerating programming and information jobs also means accelerating the creation of robots that can do these trade jobs
I think they'll ultimately go hand in hand - this is more just a question of what we're defining AGI to be and whether robotics should be mandatory as part of the stated definition around doing 95% of work.
edit: he can also turn the crank to keep it powered maybe
It could be used to let gig workers perform surgery, play in an orchestra, be actors in a movie.
Sure, it'll be one more thing that eats into bits and pieces of what we need trained labour for, but it won't eliminate it.
[meme time]
Oh god, oh fuck, we're going full Manna.
[/meme time]
https://marshallbrain.com/manna1
For those have have not read Manna, this is a huge portion of the premise of the story that AI helmets turn us into automatons that allow the AI to further automate our jobs until we are all unemployed and destitute.
I hadn't seen manna before, thanks for sharing. There are some good thoughts there. The bit about girls liking Manna because it doesn't hit on them was excellent. The dad character had some good points too.
The examples do seem a bit over the top (likely to drive home the premise). But honestly the verbosity seemed excellent for a new hire scenario. As long as it dynamically scaled down the verbosity of the instructions to match what was needed by each individual user would it really be awful if the managers got replaced?
It is a fine line, the tech being used to add to our knowledge vs making us dumb robots. If it could be kept to teaching, certain types of management, and gamified checklists it wouldn't be the downfall of the human race. However I could easily see it getting to the point where customers are talking to the AGI mounted on your head instead of you and that is scary.
I mean, this is OpenAI today "By using our product you are training our product to be you, please insert $$$ to avoid this scenario"
* You don't have the tools or supplies on hand (just let that pipe keep flooding your basement until an amazon delivery arrives!), and many of them are expensive and sufficiently specialized that a homeowner might only need them 2-3 times in a lifetime.
* You don't have any training for using the tools if you had them.
* You might not be physically capable of doing the task even if you had the tools and new how to use them, you might easily injure yourself (or others) or make the problem worse, etc.
Think of it like youtube - for most household repair tasks, you can probably already find a youtube video of someone explaining how to do the fix. Do roofers and plumbers and electricians still exist, even though a youtube video can show anyone how to do most of those tasks? AGI glasses seem like they might make less of an impact for most homeowners than youtube has.
* Every time I've encountered a home owner with a burst pipe or flooding scenario their immediate problem was lack of knowledge on they different ways to shut off the water supply. And if the problem is something as simple as lack of a wrench then you could always go get one faster than an emergency plumber could get onsite.
* The AGI glasses don't need to solve 100% of problems to be disruptive to the skilled trades. No reason they can't refer you to a human specialist. An (outsourced) human could join your AR glasses session (for a cost) and see if they can provide more instruction to both you and the AGI or if referral to a local human was needed. A Google supplied augmented reality AGI could charge users for specialized assistants AND then charge human service providers for higher placement in the referral listing.
* Local tool rental is a common thing, no reason hardware stores couldn't function as a same day fulfilment instead of amazon. And a lot of tools are cheaper to buy even if they only get used once than paying for a single service call.
* Re: tool training - I'm sure the AGI could recommend approaches to avoid tooling where possible. For example recommending shark bite connectors instead of brazing pipes, etc. But yeah, some people just won't have the dexterity to use a hammer. The issue of being able to plaster or finish drywall nicely that vidarh pointed out in another post is also a good example. It might be a boon to the local AR/AGI enhanced handman.
* Agree on roofers and electricians... where the possibility of self harm is high these scenarios are too risky without basic training. Maybe you'd have to get a DIY electricity safety license to unlock that ability in the AGI. Or maybe only apprentice electricians can use the electrical AGI, the result would still be less need for master electricians.
* Re: YouTube - As a DIY homeowner I love youtube but having to wade through hundreds of videos until I find the one video that applies to my scenario is exhausting. An AGI would already be trained on much of that knowledge. It'd lower the barrier. Again it doesn't eliminate the need for the trades but it does disrupt them further than youtube already has.
Think of it more like adding a self-checkout line to the trades. It helps in a lot of scenarios, the casher job role doesn't go away but a lot of people are unemployed or making less money all of a sudden.
Some examples:
Something goes wrong even though you did what it told you to do, but it didn't account for something specific to your situation, but the fine print says you were supposed to provide info on any non-standard circumstances, but you aren't qualified to know what needs to be said.
Or it asks you to do something surprisingly difficult that you don't know is outside your limit until you're in the thick of it - like holding something down with a lot of pressure and your hand slips causing damage or injury.
Or you do the whole thing and everything seems fine, and then later it breaks and causes damage/injury, the cause is linked back to your work so now fault needs to be assigned/split between you and the AI.
I think you nailed part of the problem here. For no less that 30% of the population the first step in the googles would be. Wear googles and go to gym for at least 6 months, along with diet guidance. "You are about to eat a snickers bar, this will delay your air conditioner repair by 16hrs"
I once tried to show a white collar friend how to spray for bugs himself, he needed a break after hand pumping the chemical sprayer. Im pretty sure in the end he just paid someone anyway.
There will be zero work, humans will be completely useless economically and we'll just have to hope that some how, all of this wealth is fairly distributed and we don't lose our minds from feeling completely useless.
What you can't imagine happening now, will be possible by deploying the AGI. Things we can't even dream of will be possible. It will be all chaotic and completely unpredictable over night. Eternal life ? No problems, pickup the phone and place your order.
What's appealing about this? I don't know, but billions and upon billions are being poured into it and many "smart" people are striving for it.
Men's way to "give birth"?
That's delusional on a disturbing level. It's as if those people have no clue about the real world, or are willingly lying. All the wealth we have now, is not fairly distributed, and just having more to distribute won't change this. Unless they speculate on putting enough stress on the system to break it, and think it will lead to something better, because...?
Maybe it's for the best that we start meeting each other in meatspace.
I'm not saying it's correct or fair that a trader sat at his desk can contribute £20 million to a country's "output", while a builder, working his backside off could maybe add £100k, but that's the world we live in...
But anyway about it, if there is suddenly an influx of manual/physical labor poured into the market, labor prices will drop far below the living cost and all hell will eventually break loose.
Is
"The Physically Embodied Bear" "95%+ will include taking physical, real-world actions."
not a mention of "construction, manual labour, mechanics, welders, painters" and "blue-collar workers" ?
Others seem to mention it is a sort of feedback look, once we get to 50% AGI I guess that AI itself can help us overcome the physical limitations. I'm sceptical that it is as simple as that, but at the same time can see it as very much possible.
How much do current AIs contribute to their own development? Likely quite a bit, but then again I don't know any AI researchers to ask.
Really a pretty fair amount these days as AIs can contribute to their own training data. In addition they've got pretty good at measuring their own success. In the past for example, if you had a robot hammer hitting a nail, you'd have a person judging the quality of the strike, but now you can face a few cameras at it and self judge far better. Yea, you still need some people around, but the amount required has dropped a lot.
I mean, why can I not get my SuperSmart5000 robot to self improve and then use it's intelligence make my situation better?
Personally, I think this is why open source AI efforts are so important. If we're going to get wiped out by AI, I hope it's an open source AI that does it.
The following is a recent video from the youtuber "Two Minute Papers" about some researchers (Clemens Schwarke, Victor Klemm et al.) who have created a highly mobile autonomous robot running software trained via simulated reinforcement learning models.
https://www.youtube.com/watch?v=Nnpm-rJfFjQ
i think it demonstrates the potential for the kind of motile finesse that could be used by robots to complete trades-like tasks.
I personally don't think they'll be economically viable in the foreseeable future, but they do represent a possible bridge to ubiquitous robot labor.
A good chunk of the value in the "trades" is in the experience and in the management.
If I can make anyone a painter by giving them a headset [1] and having an "AGI" model tell them exactly how, where, what to paint, how much of the value is the model collecting and how much of it is left for the human being?
[1] https://www.ptc.com/en/resources/augmented-reality/infograph...
I have the strange feeling that it will be those jobs that will be automated further. There is no need for manual workers in construction - you can have bots build houses, and bots painting them inside. Mechanics are like to be phased out by EVs, there's not even a need to manually change wheels and tires. Everything they do can be automated, and will be automated.
I'm pretty sure 95% in that definition would include trade workers.
5% could be something like teachers, babysitters, because you want children to have that human to human connection.
I think it could be thought of as "95%+", meaning even if computer is more intelligent, we'll still want humans because they are humans for some things, because we, ourselves are humans and that's what we want.
Things like healthcare should be fully automated, except for a small percentage again. You have a health problem, you go in, you don't need that human connection unless you have a very specific need, some of the times.
Not an easy problem, but the assumption is that AGI makes solving such problems tractable.
Japan has been working on healthcare robots for decades.
I think the key word is that the robots "could" replace 95% of economically valuable work, not that they will. Using an example from today: we can and have replaced fast food order takers with kiosks, but there are still a lot of people employed as fast food order takers because that's what customers prefer.
There are unknown issues down that road. Given 1000 years AI could probably replace all work, but I think we’re all on the same page of “could AI replace 95% of work within a timespan that’s meaningful to people alive today?”, and in that term the realization of an AI-powered machine doing all manual labour (even an AGI one) still has a too many unknowns for a meaningful answer.
We were “going to have flying cars in X” for a long time, but it turned out that building a flying car was not actually the problem that needed to be solved in order to make them viable.
I do think social mobility from being able to work is something people want. I think AI researchers want to feel important and like they have purpose. As they start to realize that will all vanish once they invent AGI, it will be interesting to see how they proceed.
I think human augmentation is far more attractive and will be a more popular field of endevour, if we workout how to get better at it.
In short -- current AI approaches are great at interpolation and still perform poorly at most extrapolation exercises that a human can take on.
Joseph P. Kennedy, 1929 (paraphrased)
When applied to the correct domains, AI is incredible.
Not disagreeing, just curious what makes you think this is so incredible ? Did you not have enough to watch before ?
This will empower fan fiction authors, web comic authors, indie creators, et al. to all reach new heights that they couldn't before.
Good web comics and fan fiction require creativity. They're low cost, but high effort endeavors that are achievable by solo creators.
Now there's a path appearing that will give this same class of people the tools to reach new creative heights that most could never achieve by themselves.
This is a good thing. It's a tool. You still have to be an artist to use it effectively.
Let's circle back on this soon.
Everyone is attributing high level reasoning and intelligence to LLMs and quite frankly there is little evidence for this being the case.
There is almost certainly an AGI winter coming.
Even at today levels of AI a very large amount of work can be optimized if the domain experts knew how to use the ai tools. As people become more used to using them I don't see an ai winter coming.
Ignoring the fact that wild science fiction has become totally mundane within the last 10 years is putting your head in the sand.
Yes it’s easy to guess the job the involves mostly moving information around would be automated but that’s not all jobs by any means.
His AI timelines haven't sped up, only clarified
Is there a way to actually mathematically quantify this better? When people make predictions with percentages like this, I understand what they’re trying to communicate, even if the math isn’t precise. There’s a reason all his percentages are in clean multiples of 5.
I don’t take it as him trying to put on a veneer of precision on what he’s saying. In fact it feels kinda obtuse to assume that’s what he’s doing.
Everyone is providing the greatest training set you can now find (since all other venues for training data will gradually disappear behind walls and fees)
http://www-formal.stanford.edu/jmc/history/dartmouth/dartmou...
The target of 95% is still 2070, the 50% is still 2045, means the progress between 2035 and 2045 is now less than before.
So the timeline has sped up and slowed down at the same time
Let me do his job for him: 90% AGI will happen by 2030.
Apple used to use Helvetica, and even an ultra thin version of it, but I wouldn't say they started the trend of using Helvetica.