[moxious]

> Some people sound like they're planning how to invent enough bullshit jobs to provide everyone a regular 9-5 schedule and a supervisor even after machines are doing all the strictly necessary labor.

I think that's mistaken on many fronts. First, new jobs that get invented aren't bullshit. Just as the internet destroyed DVD manufacturing jobs, it created smartphone app jobs, which weren't make-work. Second, the day will never come when machines are doing all of the strictly necessary labor; this is an affectation of the group-think that happens on this site. The firmly established pattern in humanity is that if you automate all of the existing work, humans will start wanting new things, which creates more work. It just....never....stops.

So while I know there are techno-enthusiasts who think AI is so imminent that it can take over even high value knowledge work in the near term, and while I know that "elimination of the middle class through robotics" is a fashionable viewpoint right now, let me just say that I have been reading articles about how strong AI was just around the corner, just a few years off...for the last 20 years. I feel exactly the same way about driverless cars. Every time I pointed that pattern out, someone was always quick to say "this time is different" without really having much evidence why it was. So I guess I'll wait for my thread reply that this time, no, it's really different, AI is going to eat all of the jobs.

I'm enthusiastic about all tech development, but prognostications about the future are pretty much always wrong. Isn't that intuitive? Does anybody really believe they can predict the future?

It's like Star Trek and the 1960s view of what today would be like. Everybody expected matter transporters and flying cars. They didn't get either. But they did get the tricorder.

[philipkglass]

If the market economy keeps producing new jobs fast enough to compensate for technology eliminating old ones, then I agree that we don't need to plan for mass unemployment at all.

But I want to stake out a position early on that if the market economy plus automation does not actually produce enough jobs for job-seekers in practice, then we should accept that mass employment was a historically contingent phenomenon that can be let go. It's not something that governments should try to keep shambling around in zombie form after the original economic rationale has died. (Bullshit jobs invented just to keep people provided with employment/income is one kind of bad response that I'm worried we'll see from protective governments. Another bad response, from neglectful governments, would be to ignore mass technological unemployment observed in practice, should it come to pass, because they're too wedded to a theory that enough new jobs will always arise in the private sector to offset jobs eliminated by automation.)

[computerex]

Everyone knows the AI community failed to deliver on strong AI, the modern boom in automation and machine learning is not about strong AI, rather the exact opposite. Instead of trying to create artificial general intelligence, researchers are now focused on solving very specific problems.

Like speech recognition, or machine vision, or predictive analytics. What's awesome is that deep learning seems to be coming out with the best results across many different disciplines and problem domains, leading to optimism about it being a general technique that can be applied to many different types of complex problems.

So yea, modern renaissance of machine learning and AI is not the same as the promise of strong AI, very opposite.

I am totally with you in that the future is hard to predict. But this isn't quite the same thing, there are notable differences here that'd suggest that this isn't like the promises made by the community since last 50 years.

Speaking specifically about deep learning, never before in history have we been able to work with the sheer volume of data that we now have and can easily work with. GPU computing shits on the CPU and is faster than the CPU by orders of magnitudes for things like matrix multiplication[1]. Advances in the field like dropout, transfer learning, ensemble learning, boosting, convolutional neural networks, unsupervised pre-training, etc have also led to breakthroughs.

Finally, researchers have more access to large freely available datasets like ImageNet, which has had an enormous impact on the field of machine vision. Freely available tools like Caffe, TensorFlow, Theano, Lasagne, Keras, Torch also make it easy for engineers and not machine learning experts to utilize the state of the art techniques to build awesome software.

[1] https://github.com/tensorflow/tensorflow/issues/3320

[moxious]

I understand that modern techniques are different than older approaches and I'm enthusiastic about their potential.

But going from saying that these approaches will improve things to claiming that they'll destroy the middle class and effectively end employment everywhere is a crazy leap that the article and excessively optimistic people often make.

We can agree that the tech is wonderful, but disagree on the future it leads to.