Posts
Comments
Our discussion look like:
Me: we can do X, that mean do X1, X2 and X3.
You: we can fall on X2 by way Y.
Do you mean "we should to think about Y before realize plan X" or "plan X definitely fall because of Y"?
A question to better understand your opinion: if all alignment community would try to realize Political Plan with all efforts they do now to align an AI directly, what do you think is the probability of success of alignment?
To summarize our discussion:
There may be a way to get the right government action and greatly improve our chances of alignment. But it requires a number of actions, some of which may have never been done by our society before. They may be impossible.
These actions include: 1: learning how to effectively change people's minds by videos (maybe something bordering on dark epistemology); 2: convincing tens of percent of the population of the right memes about alignment by social media (primarily youtube); 3: changing the minds of interlocutors in political debates (telling epistemological principles in the introduction to the debate??); 4: Using on broad public support to lobby for adequate laws helps alignment.
So, we need to allocate a few people to think through this option to see if we can accomplish each step. If we can, then we should communicate this plan to as many rationalists as possible so that as many talented video makers as possible can try to implement this plan.
I agree that there are pitfalls, and it will take several attempts for the laws to start working.
If the US government allocates a significant amount of money for (good) AI alignment research in combination with the ban, then our chances will increase from 0% to 25% in a scenario without black swans.
The problem is that we don't know what regulations we need to actually achieve the goal.
Will it work to ban all research to increase AI capabilities except those that bring us closer to alignment? Also ban the creation of AI systems with a capacity greater than X, with a gradual decrease in X.
There are many ways to increase the number of AI alignment researchers that then lead to those focusing on questions like algorithmic gender and race bias without actually making progress on the key problem.
The idea is to create videos fully describing the goals of AGI alignment, so viewers would understand the context.
I don't understand the specific mechanism that makes us need rest days. I don't see gears.
So even if politicians make regulation we need and increase number of AI alignment researchers it doesn't increase our chances a lot?
Why?
If videos convince random people, then they will convince a certain number of politicians and AI developers.
If enough people are convinced of the need for AGI alignment, politicians will start promoting AGI alignment in order to get votes.
If we do videos well, the regulations of AI development will be introduced. If we do videos really well, the government can directly allocate money for research on alignment.
Spreading this idea will increase the number of our resources (more peoples will work on it).
It doesn't work that way for me.
For example, when I repeat litany of Tarski, I think "I really really really want to know the truth about this whatever it is, and, I hope, biases will not stop me". When I try get to know a person, I 1. create a question (feeling active curiosity about person in general); 2. ask it (feeling active curiosity about this question); 3. Go back to the 1-st point;
Even if I haven't a concrete question, I often have a lot of desire to improve my map. It's so for me, because one time I read "truth let us achieve our objective and make more powerful", and I thought "I love achieve my objective and became more powerful! So... I really love truth now!".
So when I try to understand my psychology or something else that is really important, something on periphery of mind say me: "You are on the good way! You're getting closer to Omega! Continue whatever it takes!"
Thanks I would note it
If someone is working on this, they are probably not going to reply here. But, ignoring the difficulty of the task, it is not sure whether doing so would actually improve our chances. On one hand, yeah, humanity could get a few extra years to figure out alignment. On the other hand, I am afraid that the debate around alignment would be utterly poisoned; for most people, the word "alignment" would start to mean "a dangerous terrorist". So during those extra years there probably wouldn't be a lot of alignment research done.
OK, it was too radical. But what's about "coordinate action of 200 peoples, that obtain a Taiwan visa, start to work as guardians on all chips fabrics and research laboratories. And, if all other's our plans fall and humanity are about to extinct, this 200 peoples synthesize a lot of nitroglycerin in a garage..."
That's I mean by organisation we need. If MIRI did it, I wouldn't know, but my intuition say MIRI did not do it.
My factual disagreement:
I suppose people are already doing this?
So do it more instead of writing articles "How Spend Last 5 Years Of LIfe".
That was kinda the original plan of Less Wrong, which in hindsight probably seems too optimistic. (Even Putin expected three days to take over Ukraine.)
Continuing this plan is better than nothing (than accept defeat). And... good joke.
Something like MIRI?
MIRI is working on the direct alignment only, isn't?
Different tasks require different levels of talent. Compared to saving the world, creating a successful startup is trivial
Taboo "saving the world". I don't want someone to "save the world", I just want someone to "create the best youtube channel ever in using a boring theme"... ok, maybe that's impossible. But maybe that's possible, who knows.
- I mostly agree with you.
- Thanks for information about rational youtube channels and other. I have update myself.
- In fact, even if someone already do this, I wrote this article to say "it is too early to capitulate". Even if we have small chances to surviving, we should to do something, not write articles like MIRI announces new "Death With Dignity" strategy and accept the defeat. Because if you accept defeat, you would do nothing after and you won't increase our chances to surviving (and you would if you don't accept).
Basically, the answer to "why aren't people trying harder?" is that many are already trying harder, for years, some of them for decades, and... well, the predictions are not very optimistic.
I see, it's a interesting point of view I didn't think about. But it's a bias. Even if fight have little sense, considering importance of space colonisation, everything else have even less sens. How can you think about "death with dignity", if your actions can increase probability of human Milky Way on 0.001%?
That's why I was so impressed to see cousin_it propose what I think is an even better solution on the Less Wrong thread on the matter:
Or you can write a cheque to your opponent for half of the winning amount in exchange for the fact that he will cooperate, and you will defect. It won't make sense for him to defect.
I understand. My question is, can I publish an article about this so that only MIRI guys can read it, or send in Eliezer e-mail, or something.
I realized something important about psychology that is not yet publicly available, or that is very little known compared to its importance (60%). I don't want to publish this as a regular post, because it may greatly help in the development of GAI (40% that it helps and 15% that it's greatly helps), and I would like to help only those who are trying to create an alligned GAI. What should I do?
For a joke to be funny, you need a "wow effect" where the reader quickly connect together few evidences. But- go on! I'm sure you can do it!
This is a good philosophical exercise- can you define "humor" to make a good joke
The probability of the existence of the whole universe is much less than the existence of a single brain, so most likely we are an Eliezer dream.
Guessing the Teacher's Password: Eliezer?
To modulate the actions of the evil genius in the book, Eliezer imagines that he is evil.
ok thanks
I realized something important about psychology that is not yet publicly available, or that is very little known compared to its importance (60%). I don't want to publish this as a regular post, because it may greatly help in the development of GAI (40% that it helps and 15% that it's greatly helps), and I would like to help only those who are trying to create an alligned GAI. What should I do?
Moloch’s Army
We worship what brings success. Therefore, crime bosses worship power, philosophers from lesswrong worship intelligence, and middle managers worship Moloch. And just as we are ready to be curious, even if it is not optimal in this case, and to persuade others to be curious, middle managers will spread the "cult of Moloch". The same psychological mechanism.
The fascist project was an attempt to turn national politics into a maze. Fascism consists in creating the most competitive state possible, and so that individual parts of the nation do not fight with each other, taking away resources that can be directed to fight with other nations. This is literally everything that fascism has. And indeed, at first, the fascist states, which directed most of their economy to the army (competitiveness), won, and began to form alliances with each other in order to fight together against those who did not worship Moloch. Many countries even began to become fascist only because the fascist states in that epoch were strong and an alliance with them was beneficial. But, thanks to coordination, we were able to reverse this process. This happened because we had all the information about politics and wars (in corporations the boss may not even notice the conflicts between subordinates and win of a maze on some level).
Moral: to resist the mazes, we need to have as much information as possible about the psychology of other members of the company, and coordinate with other opponents of the mazes against the Moloch cultists.
Professional sport is a maze in the sense that there is a huge competition there, and if you want to reach the level of professional sports, you will have to sacrifice all the health and personal time that will be required.
Let me reformulate this essay in one paragraph:
Glomarization is good, but sometimes we can't use it because others don't understand the principle of Glomarization, or because you have too many counterfactual selves, and some of them won't like just telling the truth. Therefore, when you are asked about Jews in the attic, it is acceptable to lie, but when you are asked if you would lie about Jews in the attic, you must ALWAYS tell the truth. So meta honesty is just a way to use glomarization as often as you want.
So it shouldn’t be surprising if acting like you have more status than I assign to you triggers a negative emotion, a slapdown response.
I think there's a different mechanism here. I don't like it if Mr. A can't do X, but doesn't know about it, publicly announces that he's going to do X, and gets a lot of prestige upfront. At the same time, I understand that he will not succeed, and he should not get prestige. And after that, A fails, and it makes me feel worse about those who claim that they can do X if they have no experience.
Imagine that some philosopher announces that he is going to create an aligned AGI in a month, after which everyone begins to admire him. That's exactly the feeling.
In other words, the problem is not that Mr. A doesn't have enough prestige, but that he doesn't have enough chances to succeed.
... but even if Mr. A decides to create an aligned AGI in a month without announcing it publicly, then you will wisely say, "This is impossible. Once I also thought that I could do it in a month, but it's not like that.". Wait - this is the reaction "juggling 3 balls is impossible"!
What did I understand: most of the exclamations "you don't have enough experience / look at yourself from the outside / it's not possible" from experts in this domainare true. I mean, if you decide to do X, but all the experts in the domain say that you will not succeed, this is quite strong Bayesian evidence in favor of the fact that you will not succeed. You can't dismiss it by deciding that they're just afraid to share their status.
But otherwise I agree with Eliezer.
Sometimes, maybe you don't have time for friends to let you know. You're living an hour away from a wildfire that's spreading fast. And the difference between escaping alive and asphyxiating is having trained to notice and act on the small note of discord as the thoughts flicker by:
"Huh, weird."
Our civilization lives an hour away from a dozen metaphorical fires, some of which no living person has seriously thought about
We have a lot of people showing up, saying "I want to help." And the problem is, the thing we most need help with is figuring out what to do. We need people with breadth and depth of understanding, who can look at the big picture and figure out what needs doing
Figure out how best to spread rationality, or at least ideas about X-risks. This is quite possible with our resources equal to zero, but if we can spread these ideas to, for example, 20% of the population, it will greatly help us with the fight against X-risks. In addition, we will have more people who will help us... to think about what we should to do, lol
1) "I think we call this "taxes"."
So I invented taxes for charitable donations.
2) The second option is better for most participants, but not for everyone, you are right
This is a very useful article that helped me understand many things about myself and society. Thanks!
This is a very useful article that helped me understand many things about myself and society. Thanks!
Okay I'll rewrite the post. Thanks for your answers
That's true, but Program B will still be worse than a human-written program, so we aim to avoid spaghetti towers.
Spaghetti towers work especially poorly in changing environments: if evolution were reasonable, it would force us to try to maximize the number of our genes in the next generation. But instead, she created several heuristics like hunger and desire for groin friction. So when people came up with civilization, we started eating fast food and having sex with condoms.
People with the simulacra level 4th can praise their political allies.
I'm talking about doing an audit of your whole life regularly, desperately trying to find the most effective things. Also, this technique is about highlighting potentially the most effective actions that you didn't spend a lot of time thinking about, but put them down as "stupid" because, for example, you need to get out of your comfort zone.
Does it clear?
I think that makes you what you pretend to be to protect yourself from occlumency. That's why Harry fell into a coma - he pretended to be a stone.
Do you like the article?
I don't see the difference. The theory of relativity and Newton's theory also have different philosophies: Newton's theory states that gravity is a force, that the universe is constant and eternal, etc.
Newton's theory is not exactly a special case of the theory of relativity, because it is less accurate.
Edit: I have a lot of disagreements on my commentary. Can you explain, why are you disagree?
Jeff Bezos may announce that he will pay 5,000,000,000 to whoever invents a cure for cancer. Or, rather, to give out a monetary reward for every step towards curing cancer. Thus, if you have an idea how to cure such and such a type of cancer, you take out a loan at high interest rates (because it is risky), and conduct research.
He can form a fund that will determine which research brings us closer to cancer treatment: after all, Nobel prizes work well.
Before reading this chain, I had an intuitive sense of the "bottlenecks" of production, but this chain allowed me to understand it much better. Thank you!
""I think America has better values than Pakistan does, but that doesn’t mean I want us invading them, let alone razing their culture to the ground and replacing it with our own" - why not? No, seriously. America invaded several Muslim (fundamentalist Muslim, not we-kinda-like-Quran-stop-accusing-us-of-ISIS Muslim) countries already anyway. Why not raze the fundamentalist culture to the ground and replace it with universal?"
Preserving their culture is part of their utility function. Destroying their culture just like that is not ethical for the same reason why it is unethical to torture people just like that: both reduce their utility function, and the utility functions of other minds are included in our utility function. Therefore, only the most harmful elements of culture (religion) should be destroyed, and very gradually.
In addition, very often actions that are unethical from a dientological point of view simply will not work, because people will start to resist, and you will not reap the fruits of your unethical sacrifices. If you invade Pakistan and destroy all the mosques, a week later the country will be on fire in a general uprising, and you will not get any economic growth associated with an improvement in the political situation.
I'm not a radical dientologicalist, and I can imagine situations where something really unethical should be done, like taking over the country for the common good. But this particular plan is stupid.
Ok I agree
It seems to me that people who represent the "naive graph" by intelligence mean "the possibility of achieving goals", and the "Eliezer graph" means by intelligence the total computing power or something like that. Thus, a function that takes a value from the "Eliezer graph" and returns where this point stands on the "naive graph" is hyperexponential.
I mean, we're simplifying reality down to Bayesian networks and scenario trees. And it works. It seems that we can say that the universe is Bayesian.
what exactly do users lose and receive karma for?
Comparing Bayesian theory with frequentism is like comparing general relativity with Newton's theory. Both explain reality, and it is even easier to explain Newton's theory to children, although the general relativity is more true.
Most people intuitively learn frequentism, but some learn Bayesian methods for cases that they constantly encounter and in which frequency methods are not accurate enough.
But, in any case, frequency methods are built on Bayesian one way or another. An ordinary person, observing the Bayesian world, comes up with a simplified version of it - frequency methods. Newton, observing the world of general relativity, comes up with his own theory, where gravity is a force.
To look at the successes of the person taking into account his initial conditions. If he is a Nobel laureate, he have success, if he do science for humanity. If he is an egoist we should to look his happyness.
That we couldn't be sure that there was no error in the papers which showed from multiple angles that the LHC couldn't possibly destroy the world. And moreover, the theory used in the papers might be wrong. And in either case, there was still a chance the LHC could destroy the world. And therefore, it ought not to be turned on.
We should be less confident in the results of complex reasoning
Or, to summarize this essay:
Dientological rules are almost directly based on experimental experience, and utilitarian statements are very complex arguments.
"If you've truly understood the reason and the rhythm behind ethics, then one major sign is that, augmented by this newfound knowledge, you don't do those things that previously seemed like ethical transgressions. Only now you know why."
In other words, your theory should describe the facts well. Let's say we know that 90% of the people who decided to do X with the best intentions ended up being villains. But if in such a situation it seems to you that if YOU had done X without moral preparation, then you definitely would not have gone over to the dark side of the force... this means that your theory does not explain the data well. But if your usual consequentialist morality produces 90% of the results similar to the consequences of the dientological rules, then I solemnly declare that your consequentialist morality is practically perfect, and those 10% of discrepancies are errors of dientology, and I advise you to trust your consequentialist morality.
"They don't mention the problem of running on corrupted hardware. They don't mention the idea that lies have to be recursively protected from all the truths and all the truthfinding techniques that threaten them. They don't mention that honest ways have a simplicity that dishonest ways often lack. They don't talk about black-swan bets. They don't talk about the terrible nakedness of discarding the last defense you have against yourself, and trying to survive on raw calculation."
In this paragraph, in fact, Eliezer says: "the world is more complicated than it seems, and we do not fully understand it, so complex theories work worse than they seem, so trust the dientological rules (simple theories)"
And one more thing: it seems that when you break a dientological rule, it would be wise to remember it as: "yeah, I broke a dientological rule. I don't see exactly where I went wrong, but in any case, this is Bayesian evidence that I was wrong." And this evidence may be decisive and change the result of reflection, or it may not be.