897 karma · joined May 28, 2012
Anyone interested could also take a look at Popper (https://github.com/logic-and-learning-lab/Popper) or this overview of the first 30 years of ILP (https://arxiv.org/abs/2008.07912)
"This is the first time two Black people were spotted out in public together."
"They were there for hours and no one could tell them apart."
That's not OK. A "joke machine" that regurgitates unfunny stereotypes and racism is worse than broken.
There are some things we can't do half baked. AI comedy is one of them.
It's appealing to think we can just make a program that learns programs and then use that to learn to do anything computable. But this is a well studied field and it turns out that when you generalize a learning problem that way you make the learning problem a lot harder.
The space of programs that could possibly.identify dogs in images is much much much larger than the space of images that contain dogs. The images are bounded by the number of pixels in the image times the color depth. What is the space of programs bounded by? 10TB? That's roughly 256^10000000000 programs. That's just a stupidly large number.
Obviously not every 10TB string is a valid program. You can reduce that number. But what current research in program synthesis tells us is that you can't reduce it as much as you might hope.
So the point is that, just like for humans, it's easier to learn to do a thing than it is to learn to write a program to do a thing.
I felt like the part that wasn't in line with Kuhn was the idea that there was something wrong with a field if incremental improvement couldn't lead to a breakthrough like AGI. You're right. He's arguing Kuhn's point. But he seems to use it to conclude that machine learning is a dead end when it comes to AGI. Further, he seems to think this means AGI won't happen any time soon.
But, if I'm not misinterpreting Kuhn again, knowing that a revolution is necessary to overturn the current dogma (which I would argue is deep learning) doesn't tell us anything about when the revolution will occur. It could be tomorrow or 50 years from now or never. So, specifically, it doesn't tell us anything about machine learning in general, whether AGI is possible, or when AGI will happen.
Many people are very resistant to the idea that their particular work can have a negative impact or that they should take responsibility for that. See Yan Lecun quitting Twitter (https://syncedreview.com/2020/06/30/yann-lecun-quits-twitter...)
Other people are very aware of the dangers of their work. But, when the money gets big enough, they take their concerns to the bank and their therapist. See Sam Altman's concerns about the dangers of machine intelligence before he invested in OpenAI (https://blog.samaltman.com/machine-intelligence-part-1) Contrast that with his decision to become the CEO, take the company private and license GPT-3 exclusively to Microsoft. (https://www.technologyreview.com/2020/02/17/844721/ai-openai...) He had reasons. He posts here. He might defend himself. But to me it seems like the kind of moral drift I've seen happen when people in silicon valley have to make hard choices about money and power.
There are also applications of ML that are generally safe and can be of benefit to society. See the many medical uses including cancer detection.(https://www.nature.com/articles/d41586-020-00847-2) Most of the work being done to expose the risks and biases of ML is being done by researchers who are at least somewhat within the field. In my math and computer science program, two and a half of the 25 students are doing their thesis in safe ML. (I'm giving myself a half because I'm working on logic based ML.) I don't think it's fair to believe that every person working in ML is participating in something negative for society.
Ultimately, I think we need some reasonable regulation and a lot more funding for research into safe ML. Corporations and governments want ML for purposes that can be unethical. Unfortunately they also control a lot of the research grants. So they have a disincentive to fund AI ethics or safe ML over pushing the boundaries of what ML can accomplish.
Finally, I think many engineers would like their work to be positive for society. Unfortunately, with what we know now, a lot of the edge cases we run into are unfixable. When Google Photos started classifying black people as gorillas, Google just removed primates from the search terms. Years later, they hadn't fixed it. (https://www.wired.com/story/when-it-comes-to-gorillas-google...) I'm sure most engineers on the project knew that was a hack. When faced with an unfixable issue like that, the engineer either tries to get the company to stop using ML for that problem, compartmentalizes and ignores the issue, or they quit. Where do you draw the ethical line? It's good to hold people accountable but it's unrealistic to expect that to solve the problem.
> I share the skepticism towards any progress towards 'general AI' - I don't think that we're remotely close or even on the right path in any way.
This isn't how science works though. Quoting the wikipedia page for Thomas Kuhn's "The Structure of Scientific Revolutions" (https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Re...):
"Kuhn challenged the then prevailing view of progress in science in which scientific progress was viewed as "development-by-accumulation" of accepted facts and theories. Kuhn argued for an episodic model in which periods of conceptual continuity where there is cumulative progress, which Kuhn referred to as periods of "normal science", were interrupted by periods of revolutionary science."
I think this is the accepted model in the philosophy of science since the 1970s. That's why I find this argument about AI so strange, especially when it comes from respected science writers.
The idea that accumulated progress along the current path is insufficient for a breakthrough like AGI is almost obviously true. Your second point is important here. Most researchers aren't concerned with AGI because incremental ML and AI research is interesting and useful in its own right.
We can't predict when the next paradigm shift in AI will occur. So it's a bit absurd to be optimistic or skeptical. When that shift happens we don't know if it will catapult us straight to AGI or be another stepping stone on a potentially infinite series of breakthroughs that never reaches AGI. To think of it any other way is contrary to what we know about how science works. I find it odd how much ink is being spent on this question by journalists.
I am.
> The example Cory puts on policing
My most upvoted comment on this website was discussing this exact scenario. https://news.ycombinator.com/item?id=23655487
Could you perhaps clarify the generalization you're making about me and people like me so I can understand it?
Every marketing manager has engaged them privately to boost their numbers. Every developer secretly works for them on the side.
But no one anywhere ever talks about it until one day a former consultant notices an expired NDA.
Mistborn is a complete trilogy although he continues to publish other novels set in the same world.
Way of Kings is ongoing. It's on book four now. Each book is over a thousand pages so there's a lot there. I don't think it's a problem to start. Books 1-3 are great and stand alone pretty well. It's started to drag with book 4 in my opinion. Like so many other huge epic fantasies, it has too many characters, too many plotlines, too huge of a world, and it's difficult to maintain the epic feel with all that sprawl. I'm worried for book 5.
> Mistborn and The Way of Kings take place in the same universe?
They take place in the same universe (literally) but they are on different worlds. So they don't have anything (much?) to do with each other (yet?)
It's an important stage. But people who stay in this stage too long become intolerable technical architects who yammer away about best practices and high level company goals, produce technical documents no one reads and plan team building events no one wants to attend.
To me stage 3 involves thinking critically about tradeoffs. Applying best practices in relevant contexts and rejecting them in others is a good example.
I always end up linking this silly blog post. https://medium.com/@webseanhickey/the-evolution-of-a-softwar... It sums up the seniority growth cycle better than I can in words.
I knew enough not to feed him. Kea are trolls. You don't feed the trolls.
Here's a story about Kea I heard from a park ranger when I was in New Zealand. The rangers have traps for non-native predators scattered throughout the park. The traps are basically a big mouse trap in a box (I think.) When going around to reset the traps they found that someone was setting them off with nothing inside. They thought it was hikers. So they set up some cameras.
Instead, it was the keas. They would come to the trap with a stick in their beaks. Then they'd use the stick to trigger the trap. It made a loud boom when it closed. They'd laugh and move to the next trap. They actually have a distinctive laugh (https://www.youtube.com/watch?v=N37rN29nUIc)
I think they're one of those species, like octopi or maybe elephants, that are a lot smarter than we understand or know how to quantify. It's cool that they're getting the recognition they deserve. Even though they're little bastards.
When no one wants to make a suggestion, propose the worst possible one you can come up with. People will race to improve it because, while they don't know what they want, they know it's not THAT.
Plus their idea can't be worse than mine. So I've stolen all the embarrassment and disdain for myself.
Of course some absurdity lies in making believe your corporation has a conscience. You may have a conscience. But ultimately, no matter what corporate culture you instill, your corporation is a psychopath.
So your corporation will be Dexter and you do your best to operate it like it cares about people and issues and is a good citizen (since after all, it is legally a person.) If you do this poorly, it is revealed to have no soul and punished. If you do it well, you get a brand.
If covering for a psychopath whose only motivation is money and whose opinions are the amalgamation of half-formed thoughts from hundreds or thousands of different individuals isn't your idea of a good time, then maybe running a business isn't the best idea.
Then we rationalize it by telling ourselves that we use it ethically. It's almost always true . . . except when it's not. If 99% of the time the data is used ethically, it's easy to write off that 1% even when the 1% is all that matters.
I think it's easy to blame the symptoms of social media for the issues we have. But the internet, video quality bandwidth, smartphones, and social media together have combined to dramatically change who publishes and who finds it over the last 10-15 years.
In my mind, it is going to take us time to adapt, maybe a couple generations. Things will be difficult during that time but I hope we don't make regressive changes in our values based on what is fundamentally (in my opinion) an issue adapting our approach to free speech to the rapid advancement of publishing technology.
EDIT: "most people are not equipped to understand what is reasonable" -> This is the kind of dramatic conclusion I don't think we should be drawing.
I had a company do this and it went really poorly for me. It was a python app and at the time I wasn't very expert in python. They told me to use my laptop and all I had for python was a vim environment. I knew enough python to do white-boarding problems. But debugging a complex app exercises a whole different set of skills I just hadn't developed in the language. I was also learning vim on the side. I looked like an incompetent amateur.
So there's a language and tooling issue. The bigger problem is that debugging an unfamiliar code base under time pressure is almost as niche as doing algorithm problems under time pressure. I've been on enough company wide on calls that I actually think I'm pretty good at debugging unfamiliar code under time pressure. But if I have someone who understands the code looking over my shoulder, they're usually at least trying to help.
It's still time pressure. It's still a somewhat contrived problem and code base. The interviewer is still comfortably omniscient. And it introduces some more complexity due to unfamiliarity of the language, tooling, code base or problem space.
Ultimately, I'm not sure it's any better than white-boarding.
A lot of these problems are solved with a take home question but then all the problems with take homes rise in their place.
The Greeks were really into geometry using the compass and straight edge so they actually did a lot of math without really needing numbers at all. They viewed calculation as less worthy of mathematicians and my understanding is that we don't have a lot of evidence for how merchants and engineers did basic calculations since most of the great Greek math texts ignored it.
Algorithms and algebra probably existed in some informal way but they weren't really formalized until the Arabs did it with the help of Arabic numerals.
So, while you can do some calculations in roman numerals or using an abacus, the interesting question to me is: Did the Greeks (and the Romans) not develop algebra or use Arabic numerals because they weren't that into numbers as compared to geometry? Or was it the other way around? Did the clumsiness of doing calculations in Roman numerals keep them from developing more complex systems of numerical calculation?
I'm not an expert on the subject at all but it's always interested me. It makes me think of Bret Victor's Media for Thinking the Unthinkable (http://worrydream.com/MediaForThinkingTheUnthinkable/)
Prior to this era in which we're constantly online and connected, for a message to be heard it usually had to go through a professional editor at a newspaper, radio station, publishing company, or TV station. If you had an opinion you wanted to share, you had free speech but you probably didn't have a platform. Now messages are global, immediate, viral, and permanent.
Max Wang's video makes a strong case that Facebook's content policy and Zuckerberg's concern with adherence to it is a stringent ideology which lacks the flexibility to address clear cases that it should cover but doesn't. In other words it lacks good judgement.
I think it's going to be a long time before an automated system can exercise good judgement. Obviously having a professional editor review every tweet is impossible. So in the meantime, I think we will see a continual elevation of tension between people who value free speech at any cost and people who think that certain messages shouldn't be given a platform.
Lots of examples I've seen have phrases like "see table below". Of course there's no table and it's hard to imagine how there could be.
But GPT is trained on internet content and the internet is full of terrible writing that never gets to the point. I doubt there's any way to know how much is "not actually understanding the subject matter" vs. "learning bad writing from bad writers". I'm inclined to believe the majority is the former but there's got to be a little of the latter sprinkled in.
I have advocated before for epistemic learned helplessness (https://scienceforsustainability.org/wiki/Epistemic_Learned_...).
> thinking for ourselves is over-rated in most cases. In most cases, for most of us, good science and pseudoscience, good history and pseudohistory are going to be equally convincing. Bayesian logic suggests that sticking with mainstream experts and consensus thinking is a safer bet than rolling the dice on the Galileo Gambit.
So PG seems to be arguing that orthodox privilege is bad and independent-mindedness is important while Scott Alexander seems to be arguing that conventional mindedness should be the default. They're both very convincing essayists. Who is saying something true that can't be said and who is saying something popular but false?