Claiming that some mysterious and hard to define property that we can't measure even in principle "exists" in some meaningful way strikes me as the stronger claim than the skeptical take does.
Why do you think the burden of proof should be inverted? The mere fact that most humans intuitively feel "something" doesn't count for much of anything, especially once you stipulate that p-zombies would vote the same way.
He's also famous for the Chinese Box thought experiment, widely derided by everyone apart from his own students as the most high profile, idiotic, uninformative, trivially debunked thought experiment of all time, which teaches us negative information (in that it actually wastes time bringing up useless shit that otherwise wouldn't receive scholarly discussion except that he's an old white guy that was in the field early).
Searle is an absolute waste, nobody should engage with his drivel, ever, period.
Because of one thought experiment? It certainly generated a lot of conversation for being so trivially debunked. But it's not like it's the only thing he's ever talked about. And the old white dude reference is unnecessary.
But anyway, I think Searle had a point about semantics not being syntax with the Chinese Room. They system doesn't understand anything other than how to translate from A to B. And that's not what understanding language is about (see the later Wittgenstein or any philosophy of language).
However, as Daniel Dennett pointed out in his rebuttal, although one can produce a somewhat convincing fake to some people, similar to "passing" the Turing Test with ELIZA or any bot we've created so far, a genuine Chinese room would have to know the nuances of language at such a level that there would be no question that it understands what Chinese words mean. So Searle was wrong in his setup of the thought experiment, becuase it assumes the room is only following syntatic rules, instead of understanding the web of context and meaning that words take place in.
And I imagine many more machine learning tools will take the same name in the years to come, since it's about the most obvious one you could think of other than "brain".
ImageNet Roulette deliberately uses a terrible categorization scheme that has long been acknowledged as so poor as to not admit meaningful results in order to make the highly political point that ML should never be applied to people. There's a reason most people scrub that whole piece of the taxonomy before training.
Good Resnet models trained on ImageNet (the good parts, not just people) tend to result in state of the art results for almost every transfer-learning domain they're applied to.
There is absolutely no reason to think the brain is non-algorithmic in any way, to the extent that you can even define such a nonsense statement without waving your hands about quantum idiocy like Penrose in his senility. The default assumption in science is that any phenomenon is explainable and predictable, not the opposite: you don't get to invert the burden of proof on that front just because it would make your point (intelligence involves non-algorithmic woo) easier to make.
Even neuroscientists, who are more pessimistic about the prospect of AGI than anyone else, generally agree that the brain is ultimately not doing anything involving woo (with the exception of a few notably crazy religious ones), and is effectively just a computer. Neuroscientists think it's doing more involved computations than AI researchers hope, but it's still just crunching data.
FWIW, Penrose first posited the idea of QM having some role in consciousness about 30 years ago. I don't remember much about his argument but it seemed plausible and not easily dismissed. BTW Penrose was also a guest on the Rogan experience.
Whatever the case, we are still very very very far from understanding the brain when it comes to consciousness. There's room for people to explore possibilities.
> The default assumption in science is that any phenomenon is explainable and predictable
Two points:
a) Not everything is explainable and predictable, and thus under the domain of science.
b) There's a huge (maybe infinite?) class of things that are explainable and predictable and yet aren't algorithms.
An 'algorithm' is a very specific mathematical concept with a very specific definition. It is quite possible that the brain is explainable and predictable and yet isn't an algorithm.
> ...but it's still just crunching data.
Only if you expand 'data' to mean every possible physical phenomenon under the sun, which is disingenuous. (Are hormones 'data'? Is electromagnetic radiation? Etc., etc.)
I'm saying that the inputs to the brain are data, and the outputs are data. The brain transforms that data in some way, and we have mathematical theorems that say yep, most of the ways data can be transformed can be expressed as an algorithm in any Turing-complete language.
If your argument is that the brain leaps past normal computation into hypercomputation or something like that, then you're making an extremely bold claim that doesn't match what we know about the physical universe (there is a long history of arguments about the physical possibility of hypercomputation, and most people don't think it's possible even in theory).
I know it sounds expansive to say that everything in the physical world (at least the bits accessible to our experimentation) can be modeled by an algorithm, but that really is the mainstream scientific view, and the edges where people argue about the fringe possibilities most definitely do not apply to the energy/time scales involved with the brain.
Processing was the closest thing that I found. That was my first and simplest post-QBasic programming, and it served as a great jumping-off point into a real career in the industry.
I still reach for p5.js when I need to throw up a quick visual demo of something.
I think you're covered by the parent's "besides learning to code" caveat. That's a real smooth career transition, I did it as well as a ton of friends who had the same background. ML is a particularly good fit because physics majors just LOL at the "difficult" math that trips everyone else up.
I feel like any bachelors degree in the sciences should have you capable of doing any math needed.
I am quite concerned that Doctors in the medical field have limited to no understanding of statistics. IMO, stats should be second nature to anyone with 200 level math courses.
This might be overstating it a bit. Just this week I've been studying the AdaNet paper and grappling with Noise Contrastive Estimation. I agree that with a B.S. in physics you'll at least get the calculus you need, but I think a graduate level degree really deepens your understanding since it's not the first time through and you're already familiar with the basic concepts.
Regarding doctors and stats, I share your concern. But I disagree with the statement that stats should be second nature to anyone that's gone through the courses. If I've learned one thing while going deeper and deeper into stats, it's that there's a lot more nuance than I originally understood, and I'm still not there. Just when I think I have a thorough understanding of p-values and the like, I'll read some "I can't believe everyone doesn't understand THIS" blog and see that there was more to the story still again. It's hard to know what you don't know.
0.1% are beyond that, but at that point, hopefully your 200 level skills have taught you enough to learn about solving that.
In my lifetime, I only had 1, beyond 200 level problem that required research on math to understand. And technically, it was optional, but I volunteered.
Stats is theory + math not just math. Maybe not even any math at all, just light programming using stats libraries. If you apply them wrong though, that's a big problem. Biological experiment design is a grad course, so 500 level. Still I agree that a doctor understand the theory.
Even if every driver knew for certain that the income wouldn't cover the longterm cost of wear, there would still be drivers who needed money now enough to trade the future value of their car.
Which I imagine happens a lot. I expect a lot of/most drivers have at least a vague sense that a decent chunk of their earnings are going to get eaten up down the road in new tires, maintenance, etc. But they have rent due next week so...
You miss the obvious driver who is not an Uber driver only but one who is an Uber driver as a side hustle. This person owns a car for their own reasons, they want to drive around and have access to a vehicle. They 'rent' that expense to Uber which comes out of their compensation.
Here is another way to look at it. Someone who rents their apartment on AirBnB for one week a month which generates enough revenue to pay their rent, and they live in the apartment the other weeks of the month. This arbitraging of costs is similar to what Uber drivers do.
If Uber were to shift completely to self driven cars, all of their arbitrage ability would be lost, as the full costs of the cars would then be on their books. Now if the business model of having cars available 24hrs a day and there was no drivers fee allowed them to be profitable, that would be an interesting model. Of course one of the things that is going to happen in self driven cars will be things like drunk people throwing up all over the car, resulting in the car being out of service while someone cleans it. Or being defrauded when someone steals a phone and tells a self driving uber to drive them to the next state so they can sleep in the back seat. Then you are back to a security guard who is watching cameras of cars, but how many do you need of those? What do they get paid, are they 'contractors' or employees?
I'm rooting for them to change the world for the better but I've been unable to guess at a business model that would allow them to be profitable yet.
The cost equation for AirBnB and robo-Uber--when used in the sense of renting out personal property as a side hustle--are different though.
If you're away for the weekend or a business trip, there's very little $$ cost associated with renting out your condo for the weekend (ignoring the risk of a bad renter). It's arguably (almost) free money, again, depending on how you quantify the risk.
The cost of a car, on the other hand, is mostly related to how many miles it's driven, not over how many years it's driven. There are some time-based costs (taxes, insurance mostly, rust in snow states, having older tech) but it's mostly distance.
I agree with all of this, but I'd warn that unless you really know what you're doing and what can go wrong, jumping from weekly billing to project billing can be super dangerous and you're taking on a lot of risk. Most people aren't great at locking in scope for projects, let alone estimating the work it will take once the scope is locked; once you agree to per-project billing, you've committed and are on the hook for errors on both fronts.
Why do you think the burden of proof should be inverted? The mere fact that most humans intuitively feel "something" doesn't count for much of anything, especially once you stipulate that p-zombies would vote the same way.