7 ms·
Don't think this is accurate. I feel here "this time it's different" does apply. Previous automation waves focused on taking a mechanical task a human was doin
by Havoc 9y ago
Don't think this is accurate. I feel here "this time it's different" does apply.
Previous automation waves focused on taking a mechanical task a human was doing and making it faster by mechanical means. Compare that to the modern AIs outperforming highly trained doctors in diagnosis. Fundamentally different - instead of optimising a physical task we're replacing the need for a thinking human.
It's also worth noting that the response to previous waves was a shift to knowledge work. Now that is being threatened too. Short of moving to basic income I don't see where humans are going to go next.
- danbolt 9y agoI feel the same way with the "this time it's different" bit, but I want to play the devil's advocate. While knowledge tasks are different than mechanical ones, you could argue that they're both "tasks" that are being "done" and require thinking in some way, but just more (or different) thinking is involved this time. I don't know if machines could automate things like ingenuity, although at the same time I don't know if the demand for ingenuity is large enough for people to have a stable job.
- ThomPete 9y agoYou'd be wrong IMO: https://twitter.com/Hello_World/status/861735184990961664 https://twitter.com/Hello_World/status/861735184990961664
- krapht 9y agoThere's human ingenuity in applying computer optimization techniques to a given problem. Actually using finite or boundary element analysis software requires a great deal of expertise in order not to get garbage results out. Strong AI is not coming anytime soon. People will still be necessary.
- ThomPete 9y agoThats really not true. Humans don't know how it works. Just that it does.
- krapht 9y agoI'm not sure how much effort I want to spend arguing this. Have you ever used FEM/BEM software as a practicing engineer? In many cases, the design effort moves from making the structure to writing out all the possible structure constraints the design needs to satisfy. PS. This is not new at all if you look at the literature.
- ThomPete 9y agoYou claimed human ingenuity was necessary yet here is an example of a solution that no human could have thought up or would have needed years to calculate yet it's better than any humans could have come up with. I.e. computers are already used for ingenious solutions and that's really all that matters here.
- krapht 9y agoHuman ingenuity is needed for sanity checking the design output and ensuring the search space is sufficiently constrained so that a valid design is output. To say humans have no place in this is the same as saying in the future, my compiler will be so smart human programmers won't be necessary. Frankly that design is cool but not mind blowing. If you knew how much angst computer optimization packages can put you through, you might be more cynical. Even before NNs were all the rage there were (ex: genetic techniques, hill-climbing, simulated annealing, Monte-Carlo, pick-your-optimization-technique) to do exactly the things which that Twitter post referenced.
- ThomPete 9y agoI think you are mixing two different discussions. 1) AI is not there 2) Human ingenuity is necessary. With regards to 1) of course it's not there yet. But the progress is astounding and quite a few things will be possible even before we get to strong AI. With regards to 2) already today there are several areas where technology is better than humans to detect mistakes. In fact we use computers in many cases to test our work. This is an increasing trend and will result in end to end computer design and implemented solutions.
- danbolt 9y agoThat is super cool! I love how the design isn't conventional at all, or not what I expected. My immediate response is to think that the machine was told to "do something that does X, Y, and Z while W". The computer kicked ass at this in your link, but if XYZ were more fuzzy, or required more lateral thinking, how would a machine fare? I guess a machine could try keep trying, but the number of options to take might explode over time. Thoughts?
- ThomPete 9y agoThey all had the same information available. I will see if I can find the original paper.
- staticassertion 9y agoI think what's interesting about this is that the one on the right looks ridiculous and complicated. At some point a human being has to know how to use it, and rather than a trivial interface, it's a complex one. Many would likely still reach for the far left. A human would have to make a call saying "Wow, this is not a compromise we can make, other humans are not going to use this". That said, it's a moot point - I am in full agreement that the two revolutions are different.
- kbenson 9y ago> At some point a human being has to know how to use it You might find the first two comments interesting: @russell usability of the node on the far right is a nightmare. Where do the humans attach the cables? No empathy in AI design. What makes you think humans will have to attach the cables?
- ThomPete 9y agoThats not really a good argument as if humans was intended to install it also here you could have machines to make sure it got properly connected just like we have for sillicon chips. The argument actually speaks against what he was trying to say. He is basically saying that humans aren't clever enough to utilize optimal solutions which is a weakness not a strength.
- majewsky 9y ago> He is basically saying that humans aren't clever enough to utilize optimal solutions Your phrasing is too harsh since "optimal" is not defined thoroughly.
- ThomPete 9y agoAny solution that uses less material and is just as strong is considered optimal in my book. It is indeed defined as the blurp also says. I am trying to find the original paper again.
- coconut_crab 9y agoThe optimised process was only for weight and strength, not manufacturing. How would you craft the right one? by 3D printing? Good luck with the required strength, while the human designed one would cost a few dozens bucks here, hand welded. As a mechanical engineer, when you design something, think of the poor guy who will have to make manufacturing plan for it. Yes, maybe one day computer will be able to do both, but the problems still come from human, and human is notoriously known for undefined behaviour.
- wvenable 9y agoA fair amount of jobs require ingenuity; the few people who responsible for that are then surrounded by a dozen people who support them. Even without AI, as software developers we have been removing the need for those support people.
- sitkack 9y agoWhat happens when our ML models are trained off people, but we replace those people with models. How do we make new better models? Who trains the next wave of models? Or are we stuck with whatever we had right now? Honest question.
- Jemmeh 9y agoYes, but it doesn't take as many people to improve those models as it does to have a thinking person at every single end point of a system. If it did, programming wouldn't be profitable. And it's obviously highly profitable to automate things. And that's a good thing. Less work for humans is a great thing, but we have to have new economic models to keep up with it.
- Havoc 9y ago>How do we make new better models? Who knows? Just because you can't imagine the step after the one I describe doesn't make the one I describe less valid. Maybe it'll become self-improving? We already need AIs to help build CPUs so doesn't seem far-fetched.
- zzalpha 9y agoWe've already crossed that threshold: https://www.wired.com/2017/05/googles-alphago-levels-board-games-power-grids/ https://www.wired.com/2017/05/googles-alphago-levels-board-g... But then, about a year ago, DeepMind redesigned the system. In essence, they built the new AlphaGo without help from human moves. They trained it entirely from games where the machine plays against itself—part of a continuing progression toward AI techniques that truly learn on their own. “AlphaGo has become its own teacher,” says David Silver, the project’s lead researcher. Answer: The machines just train themselves.
- PeterisP 9y agoIt might be more accurate to say not that the ML models are trained off [decisions of] people but that ML models are trained off past experience, the results of decisions that were taken. When introducing a new ML system, it's generally trained off of people, but in most (though not all) contexts you get much, much more data once the system starts getting used in practice and you get feedback on the decisions that the machine made. The human knowledge is generally needed as a seed to make the model sufficiently not-bad so that we'd allow it to be widely used in practice and learn on its own.
- johngalt 9y agoCounterpoint: even if it is different this time, it's different in a good way. As the need for human labor decreases, costs fall to the floor. This is specifically true for knowledge work that is not constrained by resource inputs. Imagine how much HN would cost to run if it was a newspaper or magazine? This may seem like it's still bad news for humans, because where do the new jobs come from? Yet we regularly see an increase in consumption to compensate for the reduction in production costs. This increase in consumption generally overruns the job reductions due to efficiency. The number of newspaper writers may drop, but the total number of employed writers goes up. We also see this trend in technology. Early computers took a highly trained team to keep a single computer operating. Now technician to computer ratios can be 1000:1 or more. Yet we seen an increase in total employment rather than a reduction. > the response to previous waves was a shift to knowledge work. Computer assisted knowledge work may be precisely the answer to the unskilled and unemployed. Ubiquitous computing made it so that your banker only needs 1/10th the knowledge in order to be effective (when was the last time you calculated an amortization schedule by hand?). Make it so the "knowledge work" is done by the expert system and the human simply operates the expert system.
- Avshalom 9y agoThe increase in consumption seems coupled with an increase in debt. And of course in the face of a combo of increased consumption and automation the jobs created are part time entry level retail.
- euyyn 9y ago> The increase in consumption seems coupled with an increase in debt. How? And what would be bad about that?
- ThomPete 9y agoNo we aren't seeing an increase in employment. We are seeing an increase in temp work which means among other things no employe paid healthcare. Furthermore cost doesn't necessarily fall to the floor just because cost of production does. Many products are being kept up artificially and even more is being subsidized.
- rhino369 9y agoIs there really a lot of AI replacing highly trained professionals? There is a lot of hype around the idea, but I don't see any evidence of it happening in large scale. Using an AI tool will increase efficiency of course, but we've always had new developments that increase efficiency. People undervalue how radically the word processor and spreadsheet changed how offices are run. But it didn't lead to a mass layoff of office workers. I bet not even a single doctor has lost his job to AI yet.
- ThomPete 9y agoIt's not happening yet because it takes time to mature but it will happen and the more it happen the faster it will go. Under Bush and Obama almost all the jobs created were temp jobs, they even had to expand the definition of what constitutes a job.
- jml7c5 9y ago>they even had to expand the definition of what constitutes a job. Can you cite a source for this? The Bureau of Labor Statistics etc. has been using the same U1-U6 unemployment metrics for decades.
- ThomPete 9y agoYeah but which one they focus on matters. They focused on U3 rather than U6. I will see if I can find a more official source than this: http://www.washingtontimes.com/news/2015/feb/10/donald-lambro-obama-hides-jobs-failure-by-not-coun/ http://www.washingtontimes.com/news/2015/feb/10/donald-lambr...
- jml7c5 9y agoThat's a (valid) critique of using U3 instead of U6. But the US government has been using U3 as the "official" unemployment since at least 1994. You asserted that "they [the Obama and Bush administrations?] even had to expand the definition of what constitutes a job," which is not supported by that article. I have had this debate before, and I have yet to find any evidence for the claim.
- calibration263 9y ago"Compare that to the modern AIs outperforming highly trained doctors in diagnosis." This sounds exactly the same as automation replacing skilled labour. Lots of things today that we take for granted used to highly skilled markets. Things like clothing, pottery, furniture, etc all used to be highly skilled trades where people would learn the craft through apprenticeship over potentially decades. The cost of these items was massive compared to today, most of them being items a families would save up for years to a afford and keep for a lifetime. While now the production of these goods are available cheaply to everyone, and there's massive industries around them. Imagine a world where all you needed for an accurate medical diagnosis was a smartphone? Maybe the only doctors that remain are some highly trained specialists, but the number of nurses needed would skyrocket. The barrier of entry for a general practitioner could be much lower. The families who debate wether or not to feed their family or go to a doctor to ask about that cough they've had for 2 months which won't go away will disappear.
- ThomPete 9y agoYou are mixing to different effects. Yes technology makes things that used to be expensive and required highly specialized people much cheaper and thus accessible. But Technology goes further than that and replaces, in the beginning the need for as many and with time for anyone. It's not about how the market react it's what is outside the reach of AI that it can't do better than humans which can be turned into a big enough industry to compensate for the jobs lost. Thats the challenge here.
- calibration263 9y agoIt's not just about wether a machine can do something better than a human. Firstly you need to account for costs. If a machine can do it perfectly, but a human can do it 80% as well for 40% the cost, the human still has a job. The cost to hire the human is heavily tied to the economy and their cost to live, which actually ties back in to how expensive/accessible things are. You'll notice that there's still a lot of manual assembly in manufacturing today for this reason. Secondly there are a massive number of social issues. When a person does something wrong it's a mistake, when a machine does something wrong it's broken/unreliable/unsafe. So it can't just do something better than a human, it has to do it exponentially better and much cheaper(including maintenance/development/switching costs). Maybe that's possible, but this blossoming we have as AI today is nothing like that. Knowing how we do AI today and the fundamental difference between general purpose AI, which is very stupid, and extremely specialized AI, which is very smart but only useful for very specific things, gives me zero confidence that the recent AI developments are headed towards an AI that's cheaper/smarter at everything than your average joe. Take a look at all the voice assistants. They're very good at some specific tasks(they aren't general purpose, and they fail at super basic things. These failures can be addressed, but Siri isn't going to learn them on it's own, some one needs to program new features still. As an example, say my wife name is Jane, and there's another Jane in my address book. If I asked a human assistant to call Jane and let her know I had a meeting and I'd be home late, and they said "Do you mean Jane Doe or Jane Smith?", or forced me to say "tell my wife" otherwise she wouldn't do it, I'd fire her. The human assistant understands that "home" is where I live, and the only person who would care when I got home named Jane is my wife. Also when they called my wife and my wife asked when I would get home, they would think to check my calendar for the meeting I had mentioned in the message, if it was there tell her how long the meeting ran, and then even people able to figure out that there'd be some transit time. They might even think to ask wether or not I was going to grab something to eat before I got home since they might know that sometimes when I stay late I get a sandwich at the cafeteria and that my wife doesn't need to wait for me to have dinner. Identifying that "text" means send an SMS, and that the next could words need to be searched in my phone book, and the remainder is the message, is not general purpose AI. It's some voice recognition with some fuzzy matching and is not that far off from SmarterChild. All the connections a human assistant can intuitively figure out in todays "AI" need to be programmed in. They need to know to look for those variables, and make relations between them. If we stop developing Siri, she will never learn on her own to make those connections or look at those variables. Finally, the idea that if machines can do everything we do today, there will be nothing left to do is the definition of the lump of labour fallacy. Just like everything else with technology in the past, reducing the cost to produce X increase the availability and accessibility of X, reducing the number of people who had to work on X, and allowing people to instead start working on new things Y, or to do interesting awesome things with X that weren't possible when X was scarce, creating industry Z. Unless once AI is invented there is literally nothing else new we can create, I think we're okay. And I don't think AI will be out-inventing or out-creating human beings in me or my great grand kids lifetime. There's just no precedent I can see that this is happening. tl;dr: our AI suck, AI will make things(and subsequently labor) cheaper, AI will enable us to focus on new things, AI will create new industries of jobs.
- skookumchuck 9y agoEven the automation of "thinking" jobs is replacement of drudge work.
- j4kp07 9y ago"Compare that to the modern AIs outperforming highly trained doctors in diagnosis. Fundamentally different - instead of optimising a physical task we're replacing the need for a thinking human." Which in turn leads to the (cheaper) diagnosis of more illnesses, which leads to an influx of treatments, that will require more Doctors and Nurses...which leads to more jobs. "Short of moving to basic income..." There it is.
- njarboe 9y agoThis idea of basic income seems so feudal to me. Who gives me this income? How about basic capital? Everyone owns and looks over part of the robot army doing the work for everyone.
- rainbowmverse 9y agoPeople seem to have an easier time accepting patches on capitalism (welfare) than a conversion to communism (cooperative ownership of the means of production).
- njarboe 9y agoSeems like capitalism today is based on "cooperative ownership of the means of production" in that all (exceptions?) current public companies (and most private ones over a certain size) do not have a single real person as majority shareholder (>50%). A few hot tech company founders have been able to push through voting share structures such that they remain in control with minority ownership (Google, Facebook, Snap). Others have a founder who is CEO for a long time and given defacto ownership control (Musk, Gates, Bezos, etc). But these cases are the exception to the rule. Maybe the problems that are occurring with our "capitalistic" system today is that it is not really capitalism at all, but a lighter, stealth form of communism.