Tag Archives: slatestarcodex

The time-to-learn effect and the science slowdown

Posted on by

Scott Alexander writes,

There are eighteen times more people involved in transistor-related research today than in 1971. So if in 1971 it took 1000 scientists to increase transistor density 35% per year, today it takes 18,000 scientists to do the same task. So apparently the average transistor scientist is eighteen times less productive today than fifty years ago. That should be surprising and scary.

He is citing Bloom, Jones, Reenen & Webb (2018). This paper was discussed at a conference Alexander attended. He writes,

constant growth rates in response to exponentially increasing inputs is the null hypothesis. If it wasn’t, we should be expecting 50% year-on-year GDP growth, easily-discovered-immortality, and the like. Nobody expected that before reading BJRW, so we shouldn’t be surprised when BJRW provide a data-driven model showing it isn’t happening. I realize this in itself isn’t an explanation; it doesn’t tell us why researchers can’t maintain a constant level of output as measured in discoveries. It sounds a little like “God wouldn’t design the universe that way”

My favorite economics professor, Bernie Saffran, was wont to observe that learning takes calendar time as well as studying time. A student cannot master a concept merely by putting in a certain amount of hours studying it. It takes some amount of days or weeks or months for a concept to sink in. You could write L = f(T,t) where L is learning, T is the amount of time you spend studying, and t is the passage of calendar time. Throwing more T at a subject brings diminishing returns, unless you also increase t. We can speculate that some of the brain rewiring that takes place is unconscious, and you cannot artificially speed up this process.

Suppose that there is an analogous factor at work at the level of society. That is, scientific discovery depends on calendar time as well as the time that scientists spend working on a problem. It takes a while for X to sink in, and only after X has sunk in can we go on and discover Y.

Alexander sees no reason to expect that we can speed up scientific progress with simple policy changes or institutional tweaks. I am inclined to agree.

But having said that, I can think of institutional habits that may be holding progress back. I probably will write an essay on those. UPDATE: The essay offers two modest reforms.

Scott Alexander on the IDW

Posted on by

He writes,

Silencing is when even though a movement has lots of supporters, none of them will admit to it publicly under their real name. Even though a movement is widely discussed, its ideas never penetrate to anywhere they might actually have power. Even though it has charismatic leaders, they have to resort to low-prestige decentralized people-power to get their message across, while their opponents preach against them from the airwaves and pulpits and universities.

As usual with his posts, I recommend reading the whole long thing. My thoughts:

1. Commenters on this blog and elsewhere have said that Scott Alexander should be counted as a member of the Intellectual Dark Web.

2. “Scott Alexander” is not his real name, which suggests that the issue of what one would “publicly admit under their real name” is salient to him.

I would recommend trying to move away from “silenced” as a binary concept. The word itself invites a binary connotation–you are either silenced or you or not. But it may help instead to think of a continuum.

Instead, I would talk about something more like a filter ratio. For any given proposition, what percentage of time is it filtered out because of social pressure?

To take an actual example, consier the proposition that the variance of genetic mathematical ability is higher in males than females. I believe that proposition. But it seems that Larry Summers lost his job as President of Harvard because he affirmed that proposition. Since many people are aware of that story, I can imagine that not everyone would be willing to affirm this proposition publicly.

For any proposition, let the numerator be the total number of times a proposition that is relevant to a discussion is NOT affirmed by someone who believes it. Let the denominator be the total number of times that the proposition would have been affirmed in the absence of social pressure. The ratio of the numerator to denominator is the filter ratio.

When the filter ratio is zero, there is no silencing going on. When the ratio is 1, there is total suppression. “Silencing” is somewhere in between. If you want to stick to a binary view of the world, then you can say that any time the ratio is greater than 0, there is silencing. But I think a world of absolutely no filters is unrealistic. What we can reasonably argue about is how strong the filters should be for various propositions.

For example, back in the 1960s, if you had asked me, I would have been on the side of those trying to get rid of the filter that suppressed people’s use of four-letter words. But I have since come around to the view that suppressing cursing was a good thing, and getting rid of the filter was a mistake. People gave each other more respect when they acknowledged speech boundaries with one another.

In general, I see the IDW as battling the left over the issue of filters on topics related to race and gender. The left wants to implement certain filters, and the IDW sees these filters causing problems. In theory, we could get beyond name-calling and argue about what makes the filters good and what makes them bad. But the discussion rarely takes place at that level. Instead, it tends to become personal.

Scott Alexander on Eliezer Yudkowsky

Posted on by

Scott writes,

Everyone hates Facebook. It records all your private data, it screws with the order of your timeline, it works to be as addictive and time-wasting as possible. So why don’t we just stop using Facebook? More to the point, why doesn’t some entrepreneur create a much better social network which doesn’t do any of those things, and then we all switch to her site, and she becomes really rich, and we’re all happy?

The obvious answer: all our friends are on Facebook. We want to be where our friends are. None of us expect our friends to leave, so we all stay. Even if every single one of our friends hated Facebook, none of us would have common knowledge that we would all leave at once; it’s hard to organize a mass exodus.

This is Scott’s example of what Yudkowsky calls in his new book Inadequate Equilibria. Another excerpt from Scott’s review:

The Inside View is when you weigh the evidence around something, and go with whatever side’s evidence seems most compelling. The Outside View is when you notice that you feel like you’re right, but most people in the same situation as you are wrong. So you reject your intuitive feelings of rightness and assume you are probably wrong too.

…Eliezer warns that overuse of the Outside View can prevent you from having any kind of meaningful opinion at all.

My thoughts:

1. By the time this post goes up, I will have finished the book (recall that I typically schedule posts two or more days in advance). When I finish it, I am likely to write a long review.

2. The book is worth your time and your money.

3. I believe that Yudkowsky describes a real problem. Rather than call it “inadequate equilibria,” I would use a term popular in mathematical economics, “local optimum.” A group can find itself at a local optimum that is not the global optimum. It remains stuck at the local optimum because it resists going downhill to eventually go uphill.

Yudkowsky is focused on what I would call an intellectual local optimum. That is, it is possible for people to be stuck in a set of beliefs (leading to actions) that are difficult to discard but far from the global optimum. This is the way David Colander and Roland Kupers describe the state of economic thinking in their book Complexity Economics, which I described as

highly ambitious, always stimulating, and often frustrating.

I expect to say the same thing about Inadequate Equilibria. It is even more frustrating.

[UPDATE: I did indeed finish the book. I am glad that it stimulated me to think about the topic and to write an essay. But I think that you will find my essay on the topic will be more concise and more helpful than the book itself. I expect to have the essay up later this week on Medium.]

Three Axes as Tribal Rallying Flags

Posted on by

Scott Alexander has an essay post on tribalism. Read the whole thing. An excerpt:

in order to talk about tribes coherently, we need to talk about rallying flags. And that involves admitting that a lot of rallying flags are based on ideologies (which are sometimes wrong), holy books (which are always wrong), nationality (which we can’t define), race (which is racist), and works of art (which some people inconveniently want to enjoy just as normal art without any connotations).

What I call three axes are three rallying flags. Progressives rally around oppressor-oppressed, conservatives rally around civilization-barbarism, and libertarians rally around freedom-coercion. It is important to recognize that the actual belief systems are much more complex than that.

Great Minds and Hive Minds

Posted on by

Scott Alexander on Garett Jones’ book:

Hive Mind‘s “central paradox” is why IQ has very little predictive power among individuals, but very high predictive power among nations. Jones’ answer is [long complicated theory of social cooperation]. Why not just “signal-to-noise ratio gets higher as sample size increases”?

Me:

Can we rule out statistical artifact? Put it this way. Suppose we chose 1000 people at random. Then we create 50 groups of them. Group 1 has the 20 lowest IQ scores. Group 2 had the next 20 lowest IQ scores, etc. Then we run a regression of group average income on group average IQ for this sample of 50 groups. My prediction is that the correlation would be much higher than you would get if you just took the original sample of 1000 and did a correlation of IQ and income. I think that this is because grouped data will filter out noise well. Perhaps the stronger correlation among national averages is just a result of using (crudely) grouped data.