Proof of Immortality III

[JoeMorgue]

Jabba,

We can't make any clearer. You can't use "probability" (even if you were using it correctly and you just aren't) to override direct evidence after the thing in question already has or hasn't happened.

You can't sit down and calculate the probability of a coin landing heads or tails after the coin has landed to try and prove it landed the other way and that's all you're doing.

People die. You can't prove they live forever by probability after the fact.

Human mortality isn't some future event we're trying to predict. It's happened billions of times already with a 100% positive rate.

 

[Loss Leader]

You can't sit down and calculate the probability of a coin landing heads or tails after the coin has landed to try and prove it landed the other way and that's all you're doing.

... after you know how the coin has landed ...

 

[JayUtah]

They are, in fact, estimates -- that's how Bayes works. Each of us is free to project our own estimates, and to judge the plausibility of the other estimates.

No, that is most certainly not how Bayes works. Again, your own experts have told you that's not how Bayes works. Why would you go to the trouble of finding and consulting experts you insinuate that you trust, and then disregard their advice?

Bayesian inference is not a magical machine that turns belief into fact. It can test belief according to fact and refine the belief accordingly. But you don't get to just make numbers up for both your belief and your likelihood ratio and pretend that Bayes somehow make them rigorous.

No, your numbers are in no way estimates. In the context of numerical modeling, an estimate must still have some rational basis and fact behind it. If I take you down to the docks and show you a ship and ask you how long you think it is, that's an exercise in estimation. You may suck at estimating sizes, or you may be very good at it. But the point is that you're looking at the object and mentally pacing off the length in feet or meters, and that number would be an estimate. What you're doing instead is akin to telling your friend, "I'm picturing a ship in my mind. How long is it?" Your friend could only guess because he can't see the ship you're picturing.

So no, you don't get to equivocate your completely made-up numbers as "estimates."

This is the part where you tell us that you aren't just making numbers up, that you've carefully weighed all the factors that went into producing you as you are. And you say that sum shows that you're so very unique, as a product of those factors, that the scientific model can't have produced you as you are except by the slimmest luck. And it's been explained at length how that's the Texas sharpshooter's fallacy -- it produces a number, but the number it produces is not a probability. Why? Because you pre-identified you-as-you-are as some target to be hit, and that's cheating in the same way drawing the circle around the pre-existing bullet hole is cheating.

That means the number is still a guess. Creatively attributing a line of reasoning to your guess doesn't turn it into an estimate if that line of reasoning is irrational or fallacious.

 

[Mojo]

- That's where Bayes comes in. In our case, OOFLam only almost excludes the event -- the zero is the result of rounding off to three digits.

If you rounded off to one digit you could make it look as if absolutely everything is either certain or impossible. But you wouldn't have any effect on probabilities in the real universe.

 

[Loss Leader]

Taylor Swift's song "22" was almost certainly written by aliens.

The chance that Taylor Swift would come into being is ridiculously small, the same as Jabba. The chance that she would choose those exact notes with those exact words is itself ridiculously small. The two events together are ridiculously small squared.

If there's only a 0.00000001% chance the song was written by aliens, that is dozens of orders of magnitude greater.

Thus, alliens wrote the song "22."


n.b. You'll never believe who wrote " You Belong With Me."

 

[JayUtah]

People die. You can't prove they live forever by probability after the fact.

This goes back to the grand equivocation. Jabba said, "I think I can essentially prove immortality mathematically." That can mean he thinks he can prove immortality by mathematical means, or that he can prove immortality to a mathematical level of rigor. The latter, generally speaking, would be considered deductively strong. But Jabba means the former: he thinks he can prove, using the tools of mathematics, that immortality is a fact.

And yes, statistical modeling is one of many tools mathematics has to offer. But it can't prove anything according to most of the ways we think about proof. Properly used, it can establish the probability of some outcome, and the probability can, in some cases, approach what we would consider enough certainty to bank on that outcome with low risk. But no amount of statistically derived assurance overturns a contrary observation. If everything we can observe about consciousness doesn't start until birth and doesn't persist after death, no amount of statistical mumbo-jumbo can change those observations.

One of the classic examples when teaching Bayes is cancer testing. Bayes' Theorem applies because we have a good prior and good estimates for the reliability of testing. H is "I have cancer." The prior P(H) is the measured cancer rate among middle-age males. A test for cancer has a known probability P(E|~H) of false positive results and a known probability P(~E|H) of false negative results.

So I go get a cancer test and it comes up negative. What is the probability that I have cancer? Isn't it just straight-up the false-negative rate? No, it's the combination of the prior and the properly normalized conditional rates. That's what Bayes is about -- the proper way to combine those values. The prior is the cancer rate for my demographic, which is one of several reasonable ways to estimate the probability of me having cancer without measuring anything about me except for the most broad-strokes characteristics. We could actually just use the cancer rate for humans as a whole, and the method would still be valid albeit slower to converge.

I don't remember the real-world values for these off the top of my head, but the gist is that if the test comes up positive, your probability of having cancer is still only something like fifty-fifty. That's why you get the second test, using the first posterior as the prior for the second test. If that also comes up positive, then the probability you have cancer -- the second posterior -- rises to over ninety percent. It's that intermediate posterior/prior that defies common sense. This is why we invented statistics.

But going back to my test above, if it comes up negative then the small false-negative rate combines with the small prior cancer rate to produce a very small posterior probability that I have cancer. But that's still only a probability. It may be one that I can reliably use to plan the rest of my life, but it's qualitatively different from a biopsy that shows cancer cells under the microscope. The observable fact of having cancer is not the least concerned with the extremely low probability suggested by mathematics. I cannot use the mathematics to disprove the biopsy.

Jabba has been invited to identify observations that identify a soul and that extend beyond the lifespan of the organism. Not only has he failed to do so, he's gone so far as to say it could never be done. Why? Obviously so that he doesn't have to keep contending with the qualitative difference between probability and fact.

 

[jt512]

You can't sit down and calculate the probability of a coin landing heads or tails after the coin has landed to try and prove it landed the other way and that's all you're doing.

But you can calculate the probability of an observed outcome (say heads) under competing models of the coin to determine which model is better supported by the observation. That's the correct analogy to what Jabba is doing. There are innumerable problems with his methodology, but the idea of using the probability of an observation under competing models to help discriminate between the models isn't one of them.

 

[Loss Leader]

But you can calculate the probability of an observed outcome (say heads) under competing models of the coin to determine which model is better supported by the observation. That's the correct analogy to what Jabba is doing. There are innumerable problems with his methodology, but the idea of using the probability of an observation under competing models to help discriminate between the models isn't one of them.

It really doesn't matter because he has only 1 data point. You cannot discern the probability of anything from 1 data point. You can discern its possibility, but not its probability.

 

[jt512]

It really doesn't matter because he has only 1 data point. You cannot discern the probability of anything from 1 data point. You can discern its possibility, but not its probability.

In deciding between two competing hypotheses one data point can be informative, indeed decisive. Consider a classic hypothetical problem. One of two coins has been chosen at random, but you don't know which. One coin is a fair coin; the other has been bent so that it has only a 1% chance of landing heads. The coin is flipped once and it lands heads. Starting from a prior probability of 50% that the chosen coin was the fair coin, the single observation that the coin landed heads makes the posterior probability that the coin was the fair coin 98%.

As I have stated in previous posts, I disagree with every number in Jabba's argument along with the myriad untenable assumptions that led to those numbers (not to mention the one gross mechanical error I noticed: failing to normalize weights in calculating a weighted average); but, in a very broad sense, the form of his argument is valid. The machinery—Bayes' Theorem—is correct. But his inputs are nonsensical; thus, so are his conclusions.

 

[MRC_Hans]

In deciding between two competing hypotheses one data point can point can be informative, indeed decisive. Consider a classic hypothetical problem. One of two coins has been chosen at random, but you don't know which. One coin is a fair coin; the other has been bent so that it has only a 1% chance of landing heads. The coin is flipped once and it lands heads. Starting from a prior probability of 50% that the chosen coin was the fair coin, the single observation that the coin landed heads makes the posterior probability that the coin was the fair coin 98%.

As I have state in previous posts, I disagree with every number in Jabba's argument and the myriad untenable assumptions that led to those numbers (not to mention the gross mechanical error of failing to normalize weights in calculating a weighted average); but, in a very broad sense, the form of his argument is valid. The machinery—Bayes' Theorem—is correct, but his inputs are nonsensical, and thus so is his output.

I quite agree. Jabba's problem is that he only at best has the probability from the "fair coin", and even that is a wild guess.

Unless some evidence can be presented on how an immortal soul would affect the outcome, no calculation can be made.

Hans

 

[caveman1917]

No. The normalization I'm talking about does not cancel out. P(E|~H) should be .62, not .0052.

Fair enough, I'm not even sure exactly what Jabba is calculating there. I thought you were talking about normalizing the terms in the standard Bayesian formula for calculating relative posterior probabilities, in which case you don't need to normalize.

 

[caveman1917]

It really doesn't matter because he has only 1 data point. You cannot discern the probability of anything from 1 data point. You can discern its possibility, but not its probability.

You say you have 10 datapoints of coin tosses (H or T), I say you have 1 datapoint of binary strings of length 10. Or differently:

There is no inherent different between a single observation and a sequence of observations, for any given p and n there exists another p such that when taking n=1 we'd get the same result.

 

[caveman1917]

You'd say that because you don't know anything about quantum mechanics.

He got the sign of the derivative of the importance of consciousness in quantum mechanics wrong (he says it's increasing rather than decreasing), his critics are claiming that radioactive decay has "nothing to do with anything in quantum mechanics." Hence I'm still going to count him as slightly ahead in the quantum mechanics department, not that it is a really relevant department here either way.

 

[jsfisher]

Ok, so Jabba is the coin. So what experiment was then performed on Jabba comparable to a coin flip to get our one observation?

 

[MRC_Hans]

You say you have 10 datapoints of coin tosses (H or T), I say you have 1 datapoint of binary strings of length 10. Or differently: There is no inherent different between a single observation and a sequence of observations, for any given p and n there exists another p such that when taking n=1 we'd get the same result.

Mmm, each coin toss is a datapoint. Once we have tossed it a sufficient number of times, we have data points enough to make a statistical inference of the distribution of heads and tails. We cannot infer that from a single toss.

That we may make the same inference by studying the physics of a coin is an irrelevant coincidence.

ETA: It is true that they are binary data points, but that only means we need more of them.

Hans

 

[The Sparrow]

I'm going to repeat a question that Jabba is ignoring because I think its relevant.
If I truly am immortal, but have no memory of my past lives, of what 'benefit' is it to be immortal/reincarnated. If next time I am born, into a being called Fred, but Fred has no memory and hence no sense that he is a continuation of me now, how is that any different or better than Fred being a brand new clean slate. What part of me continues into Fred??

 

[caveman1917]

Mmm, each coin toss is a datapoint. Once we have tossed it a sufficient number of times, we have data points enough to make a statistical inference of the distribution of heads and tails. We cannot infer that from a single toss.

I'm not comparing it to a single toss. Suppose you have 10 datapoints in the space of binary sequences of length 1 (ie H or T in this case), you call this 10 datapoints yet it is the same as 1 datapoint in the space of binary sequences of length 10.

Suppose you have a special coin that can only be tossed in batches of 10. Your datapoints are hence binary sequences of length 10 and suppose you have 1 datapoint. You make an inference on the distribution of H and T in those sequences of length 10. Suppose someone then asks "but what if we could toss it only a single time, what would we expect?" Then you can calculate that and the answer will be the same as if you had thought of it in terms of 10 datapoints of single tosses from the start.

Hence my claim that there's no inherent difference between a single datapoint and multiple datapoints.

 

[Loss Leader]

I'm going to repeat a question that Jabba is ignoring because I think its relevant.
If I truly am immortal, but have no memory of my past lives, of what 'benefit' is it to be immortal/reincarnated. If next time I am born, into a being called Fred, but Fred has no memory and hence no sense that he is a continuation of me now, how is that any different or better than Fred being a brand new clean slate. What part of me continues into Fred??

This is a restatement, in part, of the essential question Jabba cannot answer: What is the functional definition of a soul? If we don't know what it's supposed to be doing, not only can't we test for it but we can't even say it's necessary to the system.

 

[Loss Leader]

Hence my claim that there's no inherent difference between a single datapoint and multiple datapoints.

So, if you wander over to a coin that is already facing heads up, you can extract the same information as if you flipped a coin 10 times?

 

[caveman1917]

So, if you wander over to a coin that is already facing heads up, you can extract the same information as if you flipped a coin 10 times?

Yeah that must be what I just said