Miscalculation Invalidates LHC Safety Assurances

KentuckyFC writes "In a truly frightening study, physicists at the University of Oxford have identified a massive miscalculation that makes the LHC safety assurances more or less invalid (abstract). The focus of their work is not the safety of particle accelerators per se but the chances of any particular scientific argument being wrong. 'If the probability estimate given by an argument is dwarfed by the chance that the argument itself is flawed, then the estimate is suspect,' say the team. That has serious implications for the LHC, which some people worry could generate black holes that will swallow the planet. Nobody at CERN has put a figure on the chances of the LHC destroying the planet. One study simply said: 'there is no risk of any significance whatsoever from such black holes.' The danger is that this thinking could be entirely flawed, but what are the chances of this? The Oxford team say that roughly one in a thousand scientific papers have to be withdrawn because of errors but generously suppose that in particle physics, the rate is one in 10,000."

Comments

Voodoo Science

[alain94040]

This is voodoo science. And I don't mean the LHC experiments.

I mean the TFA that in essence claims that because an expert may be wrong, any probability the expert assigns to a risk can be ignored and inflated by as much you feel like it. Talk about bias.

--
The 5 Steps to a Great Startup Idea

Re:Voodoo Science

[orclevegam]

Essentially their argument boils down to because people make mistakes and we can calculate the odds of them making a mistake, if they calculate the odds of something and it's greater than the odds of them having made a mistake then you have to use the odds of them making a mistake as the probability of the event happening. Of course this reasoning is total bullshit, and just the sort of abuse statistics gets a bad name for. By that sort of reasoning we should all go play the lotto as clearly the odds of someone miscalculating the chances of winning the lottery are much better than the calculated odds of winning, never mind the fact that even if they made a mistake in calculating the odds it wouldn't shift the calculation enough either way to get it anywhere near the odds of them having made a mistake.

Re:Voodoo Science

[bhagwad]

I agree. Just look at this statement: "The focus of their work is not the safety of particle accelerators per se but the chances of any particular scientific argument being wrong." Can you get any broader than that? What they're essentially saying is that anything can be wrong - Including their own paper.

Re:Voodoo Science

[KagatoLNX]

Actually, this isn't that much voodoo.

It's just saying that, if someone has a 1/10,000 chance of being wrong, their assurance that there is a 1/1,000,000,000 chance of something isn't that good of a bet. In other words, if you want the latter level of certainty, you don't really have it, because of the fallibility of the research itself.

This is actually rather obvious. If Jimbo tells you that there's a 1% chance that your tire will go flat if you don't fix it, that's not 1% if Jimbo is wrong 50% of the time. At best, it's 50.5%. Or something like that.

Assuming his brother Jethro is just as bad (but uncorrelated) with him, then their dual recommendation that it will go flat only gets you 25.25% certainty, not 1% (or 0.01%). The numbers may not be exactly right (my stats are rusty), but you get the point.

Basically, they're saying that the research provides a wider error bound than it may claim, assuming that scientists uniformly make logical mistakes--which they very probably do.

The implication, then, is that the LHC estimates should be independently done by other teams. This is, well, the basis of the scientific method, so essentially this study provides a statistical analysis of what we already know--after enough work, science gets results. Of course, the base theories assumed by all of the researchers could be wrong, which would be unfortunate, but the LHC is going to nail that one pretty quickly. :)

This is not surprising, but not voodoo either.

Re:Voodoo Science

[IorDMUX]

then you have to use the odds of them making a mistake as the probability of the event happening.

This isn't what the actual study states, though the summary seems to hint that way. To quote from one-of-the-FA's:

Which means we are left with the possibility that their argument is wrong which Ord reckons conservatively to be about 10^-4, meaning that out of a sample of 10,000 independent arguments of similar apparent merit, one would have a serious error.
Of course, this doesn't mean that the LHC is dangerous, only that there is no reasonable assurance of safety which, as Mark Buchanan writing in New Scientist this week says, is not the same thing at all.

To sum it up, they say that if a researcher predicts an occurrence rate for an event that is less than the researcher's own error rate, then the occurrence rate remains unknown ('cannot be assured')... not that it is equal to the researcher's error rate.

Re:Voodoo Science

[Anonymous Coward]

Actually the point the article makes is not that there is a 1/1000 chance that the LHC will destroy the world but rather it is meaningless to say that the odds are as small as they safety reports etc say because the chance of the reports being wrong is greater than their predictions.
 
  It basically boils down to saying that the scientists are saying there is a one in a billion chance that the LHC is dangerous then turning round and saying that there is a 1/1000 chance that that figure is wrong. Basically the point is that neither statistic is very helpful. Since the 2nd invalidates the first but tells you nothing about the actual probability of a dangerous event.

Re:Voodoo Science

[SEE]

But for the LHC, arguably there is no accurate prior because nothing in that energy range has ever been done before

How many natural events involving hadrons in LHC+ energy ranges do you need?

99% of cosmic rays are made of hadrons (mostly protons and helium nuclei, but heavier nuclei are known), and they regularly collide with other objects made of hadrons (most of the mass of the visible universe) at LHC-plus energies.

Want me to worry about the LHC? Tell me when a cosmic ray collision has turned the Sun into a black hole or strange matter or new Big Bang or whatever your LHC disaster scenario is.

This is dumb as shit.

[bluefoxlucid]

Opponent: Oh crap, you're whacking things together, it could destroy the earth, crazy scary technology we don't understand!

Proponent: That could never happen.

Opponent: OMG yes it could you don't know wtf you only have studied this shit your whole life you're not a sane normal rational person like the boys in Alabama!

Proponent: Look, we've done tons of calculations; we've compared this against real-world natural occurrences; we've considered the number of times the conditions we've come up with have occurred in our lifetimes, and it's huge. We're just scaling it down to a laboratory level so we can observe it in a controlled environment. It can't break anything.

Opponent: BUT YOU COULD BE WRONG!!!!

Re:Are they good for anything?

[aliquis]

I don't see the problem, facts:

1) We will all die some day.
2) The solar system will stop working some day.

So what's the problem? Sure it may kill us and all life on the planet, but does it really matter? We're screwed anyway.

Re:My first thought from reading this

[Thiez]

> I STILL don't think the LHC will kill us all but the fact we're debating it says something.

Yes, it says that people are easily scared by things they do not understand. See also: wireless, mobile phones, things that have a 'chemical' smell... Ask some random people what would happen if the sun were to be replaced instantaneously by a black hole with a mass equal to that of the sun (moving in the same direction as the sun with the same speed, etc). Most people will reply that the earth would get 'sucked' in the black hole... if you don't even understand gravity you have no place in a debate concerning the LHC.

Everyone is entitled to an _informed_ opinion.

awesome logic

[j0nb0y]

paranoid person: The LHC is going to cause a black hole!
scientist: No, the LHC is not going to cause a black hole.
paranoid person: The chances of a scientist being wrong is 10%, therefore there is a 10% chance that the LHC will cause a black hole!

Voodoo posting

[Burning1]

This is actually rather obvious. If Jimbo tells you that there's a 1% chance that your tire will go flat if you don't fix it, that's not 1% if Jimbo is wrong 50% of the time. At best, it's 50.5%. Or something like that.

Okay seriously?

The probability that Jimbo is wrong is unrelated to the probability of your tire failing. If jimbo says that you have a 1% chance of your tire failing, but there's a 50% chance that jimbo is wrong we can reach the following conclusion: There is a 50% chance that your tire has a 1% chance of failing. There is a 50% chance that your tire has some other probability of failing. Some other probability of failing includes values such as 0%, .5%, and 2%. It also includes a 100% probability of your tire failing.

However, we have to assume that Jim isn't pulling the 1% figure out of his ass. If your tire was 100% likely to fail, we can still assume that Jim based his statement on a reasonable analysis. Perhaps Jim didn't notice a nail in your tire, but without knowing the quality of Jim's inspection of your tire, or without having access information Jim doesn't have, it's hard to say that he has a 50% chance of being wrong.

Finally, in some cases a professional will include a certain amount of leeway in his figure. Chances are, Jim fully inspected the tire and doesn't see any reason why it would fail prematurely. Chances are, that 1% is left as wiggle room in case of invisible manufacturing defect or a mistake in his evaluation. In this case, Jim has already factored into his evaluation the chances that he's incorrect.

Frequency of outcome vs. degree of belief

[Estanislao Mart�nez]

In fact, you make the same mistake that the authors appear to in your logic.

No, it's not a mistake. It all comes down to the fact that there are two general types of interpretations of probability:

1. The frequency at which one of the possible outcomes happens in repeated instances of an event of a specified type. For example, the probability of heads in a coin toss.
2. The degree of belief that a cognitive agent assigns to a sentence. This degree of belief is related by the laws of probability to the degree of belief that an agent should assign to other sentences, in such a way that only some assignments are consistent (by a technical definition I won't go into here).

Basically, you're treating this as an argument about probability in the first sense, when it is really about probability in the second sense. The argument is that even if your formulas lead you to asssign a degree of confidence of .00000000000001 to the proposition that the LHC will not destroy the Earth, that means very little if we assign a degree of confidence of .000001 to the proposition that you are wrong.

The point now, which other posters in this thread have made in other ways, is that the frequency model for probability theory is not relevant here, because this situation is not like a coin toss. For the situation to be like a coin toss, we would have had to do something like run the LHC a gazillion times, and observe how many of those times it ended up destroying the Earth. Therefore, the probabilities must be interpreted as degree of belief, and the number produced by any formula must be tossed out if the probability of getting the formula wrong is bigger than that number.

It's this fallacious reasoning -- that if the theory is wrong, the probability of the event must be greater -- that make this article technically true, but useless.

The assumption you're making here is that the number is the "probability of the event." Again, it is not; it is the degree of belief warranted to a specific proposition, given some other information.

An excerise in stating the bloody obvious

[Roger W Moore]

It's both right and wrong. The conclusion that we can't trust the probability of disaster if we got it wrong is correct...bloody obvious, but correct. The part where they use the population of the Earth to determine whether the LHC "risk" is acceptable is frankly insane. This seems to suggest that if Bird flu wipes out half the population then the "risk" of running the LHC is suddenly now more acceptable?