Hacker News Reader

Comment by jerf | Hacker News Reader

The article is not about picking parameters. The article is one meta layer up from that; they examine the distribution of parameters.

I can't guarantee this is novel paper in general, but it's not something I've seen treated this formally before. I mean, it roughly conforms to what I've been saying for a while, but I only worked the math very intuitively and in a manner that could have been flawed, so I get no credit. They've worked it formally and put it on a solid basis. Plus as we adjust our probability ranges in the future, the analysis can be re-run with new numbers easily and produce concrete, or at least, concretely-vague answers, instead of vaguely-vague intuitions.

View on HN

> I can't guarantee this is novel paper in general, but it's not something I've seen treated this formally before

Yea, I'd be very happy if they just said "here we make formal the possibility, considered since the paradox was first proposed, that there is enough uncertainty in each parameter that seeing zero ETI is perfectly compatible with expectations". That would have been a modest but useful contribution to the lit that would have informed further research. The issue I have is with claims that they are "dissolving" anything.

OK, picking distributions. I can't imagine that there's enough data to even do that.

I think you (and some others) are being dismissive of a good-faith attempt to formalize thinking of a problem with many unknowns. I can't access the paper itself, but commenter carry_bit posted a link to the slides, and the conclusion slide says:

The Fermi question is not a paradox: it just looks like one if one is overconfident in how well we know the Drake equation parameters.

• Doing a distribution model shows that even existing literature allows for a substantial probability of very little life, and a more cautious prior gives a significant probability for rare life.

• The Fermi observation makes the most uncertain priors move strongly, reinforcing the rare life guess and an early great filter.

• Getting even a little bit more information can update our belief state a lot!

I think their arguments deserve more consideration than what they are getting in this thread.

You should be able to access the paper. Arxiv doesn't require any logins or credentials or anything; just check the "download" section in the upper right. This should go straight to the PDF: https://arxiv.org/pdf/1806.02404 (Unless, you just mean you're on a device that can't render PDFs or something. It's a simple PDF, though, even a phone should be able to handle it.)

I'm on a large corporate network for a research organization, and I think they're rate-limiting us. All of the links give me an "Access Denied" message, with a referral to https://arxiv.org/denied.html. I can't even access https://arxiv.org/.

"OK, picking distributions. I can't imagine that there's enough data to even do that."

Of course there is enough data to pick the distributions of estimates given by various people. We have all the estimates given by people in hand.

You seem to be being dismissive without even remotely engaging with what is being discussed here.

Let's skip over the part where you fall back to the claim that the distribution of probabilities given by people isn't relevant to anything, because that's not important to the paper. The entire point is to examine the distribution of such probabilities, and compare it to the observed universe, and that's it. Nothing more.

There is quite a bit of literature on picking distributions in the absence of observations. "Probability Theory: The Logic Of Science" by E. T. Jaynes has a whole chapter on "ignorance priors", where he shows how to pick prior distributions in the absence of quantitative information.

Generally, these ignorance priors are improper (unnormalizable) distributions. One example is scale parameters, where the approriate prior is 'ds/s', which makes any order of magnitude equally probable. It doesn't have a mean, and it favors small values. A tiny bit of evidence will immediately turn it into a proper posterior distribution; I think this prior has zero information content.

Of course, if one applies that approach to the Drake equation, one ends up with an equally uninformative posterior distribution, unless one can add evidence for each single variable. But that evidence just isn't there for some of them, so the result remains the same: It's impossible to estimate N, but it's probably very small. As it should be, given the total lack of data.

Reply on news.ycombinator.com