Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[faust]

I've been confronted with the "buying meat is ok because your contribution won't make any difference" argument about half a dozen times throughout my life.

But it's a non-starter! Purchasing meat is very unlikely to make a difference, but it has the potential to do an enormous difference! Maybe your purchase will push the store's numbers over a threshold, which will lead the company in charge of producing the meat to increase production.

So if X is the random variable measuring impact, then E(X) is the product of p (the probability that this happens) and let's call it z (the impact if it happens). And E(X) = p*z will come out as almost exactly the same impact that you would have if someone were killing animals for you personally. To see this, you need only imagine that a million people became vegetarians tomorrow. This would decrease meat production by about the amount that they consume, therefore the impact of each one is 1/1000000 * {amount that they consume} ≈ {amount that one person consumes}.

It's a few discrete jumps as supposed to a continuous curve, but it all nets out to a normalcy behind a veil of ignorance. Ditto voting (if you live in a swing state). So many people don't get this!

The whole individual responsibility argument is a scam. It is not an accident that the idea of a carbon footprint was popularized by BP.

Your math is flawed. Yes it would make a difference if a million people became vegetarians tomorrow. But it is not possible to deduce that if you become vegetarian, that is one millionth of the impact, because 999999 other people will not become vegetarians.

It is a fact that a vegetarian lifestyle, at least in the current environment for most people, is more time-consuming than an omnivorous one. At the very least, making the adjustment takes effort. It is fair to ask whether that energy is not better spent elsewhere. And it is! It is much more effective to works towards changing policy, because a lifestyle change means you only get to affect 1/{human population} of the problem, whereas by organising you can affect 1/{politically active population}, and that is a much higher number.

The best thing about being a vegetarian (or vegan for that matter; if your concern is the environment, then any animal products are problematic) is that you get seen being vegetarian and might influence other people in this direct, i.e. virtue signaling. I don't want to diminish that, and if it makes you feel more at ease with yourself to be vegan, then by all means do it! But it is counterproductive to claim some sort of moral high ground. I have influenced way more people towards becoming vegetarian by cooking them a good veggie meal than I have by arguing the moral superiority.

[silverspawn]

I agree with everything in your last two paragraphs, but this doesn't make the math flawed! The math says that your expected impact on meat production is roughly equal to the amount of meat you eat. Well, a google search says that the average american eats 124kg meat per year, or about 63% as much as one cow "produces". You can probably do a hell of a lot more than that by changing policy, so lifestyle choices are ineffective -- this follows from the math!

I met the the founder of proveg on the meditation retreat in Berlin, and he agreed that personally being vegan is probably not worth for him except for PR reasons (but he'll ofc stay vegan anyway). But the "your personal lifestyle has 0 effect because it won't affect meat production" argument is still 100% wrong, even if the conclusion happens to be directionally correct.

[silverspawn]

Your math is flawed. Yes it would make a difference if a million people became vegetarians tomorrow. But it is not possible to deduce that if you become vegetarian, that is one millionth of the impact, because 999999 other people will not become vegetarians.

Meat production changes over time, obviously, which means thresholds are crossed all the time. And you have no way of knowing whether or not you are the person who will tip the threshold, so your probability of doing that is 1/{number of people who are required to tip the threshold}.

[silverspawn]

The point of the 100.000 people stop eating meat thing was that a functionally identical thing happens the time when production changes for whatever reason. ANd it doesn't matter whether it goes up or down because preventing increase has the same EV as affecting decrease.

[silverspawn]

Apropos of nothing, Lalight are you still alive?

[silverspawn]

The authors find rather stunning that, given the importance of visual illusions for the vision science community, the neural inspiration of CNNs, and that so often the aim of CNNs is replicating human behaviour, there is virtually no work done on linking visual illusions and CNNs. To the best of our knowledge there are only two, very recent, publications in this regard. The first one comes from the vision science field [8], where a CNN trained to predict videos was able to reproduce motion illusions. In the second one, from the perspective of computer vision [9], the authors classify and attempt to generate new visual illusions using generative adversarial networks.

This is like a birthday present! Someone just answered exactly the question I'm trying to figure out! It's 3 years old, but still.

[Awaclus]

I bet that this has come up somewhere on f.ds before, too, though I don't remember any instance.

It has, in almost exactly the same terms, using vegetarianism as an example, in this thread, by you.

Ah, but if you need N votes on average for every seat, then voting for a party has chance 1/N to increase the number of seats by one. If you treat this as a random variable (which it is), then the expected value is just 1/N * (value of one seat), which is exactly the same as if every vote counts equally.

After all, say you take N people who change their vote from party X to party Y. If this happens successively, then the seat must flip after one of those N people. As long as you have no idea where in that process you are, again you have 1/N to flip the seat and (N-1)/N to do nothing, which yields expected impact of (1/N)*(value of one seat).

The same principle is true for every system that is comprised of many small inputs and reacts at a low resolution, like if you stop eating meat or reduce your footprint.

[silverspawn]

It's happening, I'm getting old and forgetful

[silverspawn]

I recently turned 28. If we measure similarity of numbers by edit distance, this is only one step away from 82. I should look into retirement.

But this is what I'm saying, the argument keeps coming back. The first time I've heard if was in middle school when trying to convince a friend to go vegeterian, then many times in between, and now someone touched on it on twitter, again about vegetarianism!

[faust]

I agree with everything in your last two paragraphs, but this doesn't make the math flawed! The math says that your expected impact on meat production is roughly equal to the amount of meat you eat. Well, a google search says that the average american eats 124kg meat per year, or about 63% as much as one cow "produces". You can probably do a hell of a lot more than that by changing policy, so lifestyle choices are ineffective -- this follows from the math!

I met the the founder of proveg on the meditation retreat in Berlin, and he agreed that personally being vegan is probably not worth for him except for PR reasons (but he'll ofc stay vegan anyway). But the "your personal lifestyle has 0 effect because it won't affect meat production" argument is still 100% wrong, even if the conclusion happens to be directionally correct.

I guess it's less the math being wrong than the phrasing being suboptimal. You invoked the 1 million figure to say that there was a measurable effect. But it is entirely possible for a thing to have a positive effect if a million people do it and yet no effect if only a single person does. But I am not trying to say that that's the case with vegetarianism.

[silverspawn]

I was saying there is positive *expected value*. There's likely 0 effect and maybe a large effect, and in expectation the effect is as large as it should be.

[Awaclus]

It is possibly worth noting that reducing your meat consumption is not a binary switch, and you can pretty easily contribute like 50% of the expected value for like 5% of the effort.

I used to try to have "vegetarian days" where I would not eat any meat that day, and that was kind of doable for a while. I usually had like at least one and sometimes two of them per week, and the idea was that I would get used to them and start to have them more and more often (but still at least sometimes eat meat, because it would suck if there was a war or other disaster and my digestive system had gotten used to a completely meat-free diet, which may or may not be available in that scenario). However, especially in the long run, it was substantially harder to come up with different foods I could eat on a low budget and low willingness to cook if I had to exclude meat completely, and that was eventually draining my energy more than the positive feeling from having successfully done something productive was benefiting me, and it ended up becoming never two but sometimes one per week, and then eventually never.

What I have been doing more recently instead is that I will still eat all the same foods, I will just put less of the meat and more of the other things on my plate. This takes no effort whatsoever and I am pretty sure I have reduced my meat consumption by at least a third, possibly more, and would be relatively easy to combine with occasionally eating vegetarian foods to get even more reduction.

[Awaclus]

I guess if your diet is already optimized to contain a healthy amount of meat, you can't actually do the thing without any downsides, but most people's diets have so much meat in them that doing it will certainly just make them healthier.

[silverspawn]

It is possibly worth noting that reducing your meat consumption is not a binary switch, and you can pretty easily contribute like 50% of the expected value for like 5% of the effort.

I used to try to have "vegetarian days" where I would not eat any meat that day, and that was kind of doable for a while. I usually had like at least one and sometimes two of them per week, and the idea was that I would get used to them and start to have them more and more often (but still at least sometimes eat meat, because it would suck if there was a war or other disaster and my digestive system had gotten used to a completely meat-free diet, which may or may not be available in that scenario). However, especially in the long run, it was substantially harder to come up with different foods I could eat on a low budget and low willingness to cook if I had to exclude meat completely, and that was eventually draining my energy more than the positive feeling from having successfully done something productive was benefiting me, and it ended up becoming never two but sometimes one per week, and then eventually never.

What I have been doing more recently instead is that I will still eat all the same foods, I will just put less of the meat and more of the other things on my plate. This takes no effort whatsoever and I am pretty sure I have reduced my meat consumption by at least a third, possibly more, and would be relatively easy to combine with occasionally eating vegetarian foods to get even more reduction.

fully agree with this btw

[silverspawn]

So DALL-E has a fantastic understanding of human art, specifically wedding dresses, but it cannot count to five. Maybe this will finally kill the stereotype that emotion and artsy things are hard for AI but reason is easy.

LOL just kidding I don't expect it to do anything to the stereotype.

[silverspawn]

Maybe once people start replacing illustrators for children's books. This shouldn't take too long. The best way to hide the fact that DALL-e is still incompetent at many areas is to just avoid those things. I expect you should be able to produce purely DALL-e illustrated books that look good.

[silverspawn]

So, Global Workspace Theory, the academically most prominent "theory of consciousness".

According to this review, which everyone says is really good, it says roughly that

- There is a lot of unconscious processing going on in the brain
- There's some mechanism that promotes important things into consciousness
- This is a discontinuous thing, things are either in or not in consciousness; they're never "sorta" in. For example, if you show someone an image of a digit for 400ms, they almost certainly won't remember it, but somewhere around 450ms (I think), they do, and if you look at their brain function, you either see a signal getting lost quickly or being strongly amplified, nothing in between
- We call this the "Global Workspace", and say then talk about what kinds of things this global workspace is and is not required for

[silverspawn]

My charitable interpretation of this is that it's quite nice! It studies what things do and do not make it into consciousness, which is a relevant question. And the research seems solid, too.

But it's not a theory of consciousness, like at all. It doesn't talk about why this unified workspace thing leads to first-person experience, and (unless that was just missing from the review), also doesn't talk about how this global workspace works.

So it tackles like 5% of the problem. Which is still great. It only becomes bad when you look at it and think "ah well, if we have some kind of a system that includes a filter for important information, and we let the important information be processed by a subsystem that we call 'global workspace', and then the entire thing shall be conscious!". This is like observing what happens to an egg when it fall onto the floor, building an item that behaves like that, and then expecting it to taste the same when you try to cook it.

(The uncharitable interpretation is exactly that, treating the theory like a blueprint.)

[Awaclus]

Maybe once people start replacing illustrators for children's books. This shouldn't take too long. The best way to hide the fact that DALL-e is still incompetent at many areas is to just avoid those things. I expect you should be able to produce purely DALL-e illustrated books that look good.

I wouldn't count on AI-generated illustrations in children's books any time soon. We're probably going to start seeing indie games (digital ones, that is) with AI-generated art in the near future because the art is often an afterthought for these games and it saves costs, and we're probably going to start seeing illustrators who actually use AI themselves e.g. to generate ideas, but the illustrations in children's books are often at least as important as the text and it's usually one person doing both the writing and the illustrations, or two people working together as a team such that the art also inspires the writing, not just vice versa. And when you're releasing a physical book, the cost of getting the books printed and distributed is pretty high so even if you would have to specifically hire an artist to illustrate your writing, it probably would not be a substantial part of the total cost.

[silverspawn]

Hm.

Maybe.

[silverspawn]

I honestly always expect the worst. Whenever I open my school mail account and see 5 new things, I go "oh crap did I do something wrong?" The answer is almost always no, there are mails all time about all sorts of stuff, but that's my reaction.

Or when my phone rings, always assume some trouble is incoming. Or when I see someone left a call.

[Awaclus]

I honestly always expect the worst. Whenever I open my school mail account and see 5 new things, I go "oh crap did I do something wrong?" The answer is almost always no, there are mails all time about all sorts of stuff, but that's my reaction.

Or when my phone rings, always assume some trouble is incoming. Or when I see someone left a call.

When I see new things in my e-mail (I have routed my school e-mail to my main one so I basically only have one nowadays, but this applied even for my work e-mail during my internship), I assume it's going to be something exciting. Usually it's nothing remarkable, but sometimes it is something mildly exciting, pretty much never bad news.

When my phone rings, I also assume some trouble is incoming. And I'm correct every time, because a call is incoming and calls are inherently troublesome.

[Awaclus]

During my internship it was actually bad news sometimes (e.g. someone replying back to a ticket I was hoping had already been resolved and could be closed due to inactivity soon), but oh well, they can't all be the best news ever.

[silverspawn]

When my phone rings, I also assume some trouble is incoming. And I'm correct every time, because a call is incoming and calls are inherently troublesome.

Amen

[silverspawn]

So, what tasks (that humans can do) are difficult for computers?

Though I would get roasted for this from most ML people, my answer is roughly "everything that consciousness is involved in". Since computers don't solve the Binding Problem, they're missing out on the effects of consciousness in the brain.

Two caveats to this:

- if human brains do something via consciousness, this probably means it's easier than with an unconscious algorithm. But it doesn't tell us if it's 5% or 5000% easier. This is really hard to know.

- An unconscious algorithm could also be hard for good old fashioned engineering reasons. Even if the brain does it more or less like a neural network, maybe it's still difficult.