I speculate there’s at least two problems with the creationism odds calculation. First, it looks like the person doing the calculation was working with maybe 60,000 protein molecules rather than zillions of protein molecules.
The second problem I’m having trouble putting precisely in words, concerning the use of the uniform distribution as a prior. Sometimes the use of the uniform distribution as a prior seems to me to be entirely justified. An example of this is where there is a well-constructed model as to subsequent outcomes.
Other times, when the model for subsequent outcomes is sketchy, the uniform distribution is used as a prior simply as a default. Or, as in this case, it’s clearly not an appropriate prior. In this case, the person is probably assuming that all combinations of proteins are equally likely (I suspect this assumption is false.)
Consider that
1) There is more than one possible arrangement of proteins which qualifies as a living cell, and that
2) the materials of which proteins are made had quite a long time to shuffle around and try out different configurations between when the earth cooled and the present day, to say nothing of other planets elsewhere in the universe, and that
3) once a living, self-replicating, self-repairing cell has come to exist in an area with appropriate raw materials and a steady energy source it will create more such cells, so it only has to happen once.
So, we’re looking at a sample size equal to, by my back-of-the-envelope estimation, the number of cell-sized volumes in Earth’s atmosphere and oceans, times the number of planck instants in a little over four billion years, times the number of earth-like planets in the universe. The actual universe, not just the part we can see.
For intelligent design to be the most reasonable explanation, the probability of life emerging spontaneously would have to be low enough that, in a sample of that size, we wouldn’t expect to see it happen even once, and, furthermore, the designer’s own origin would need to be explained in such a way as to be less improbable.
If the temperature is high enough that there’s molecular movement at all, you could observe a collection of proteins every Planck-instant and see a (slightly) different arrangement each time. You might be stuck with similar ones, especially stable configurations, for a long time… but that’s exactly the sort of bias that makes life possible.
Isn’t the problem more like: they are ignoring the huge number of bits of evidence that say that cells in fact exist. They aren’t comparing between hypotheses that say cells exist. They are comparing the uniform prior for cells existing to a the prior for only random proteins existing. They sound more like they are trying to argue that all our experiences cannot be enough evidence that there are cells, which seems weird.
This is a misinterpretation. The argument goes like this:
True statement: There is lots of evidence or cells. P(Evidence|Cells)/P(Evidence|~Cells)>>1.
False statement: Without intelligent design, cells could only be produced by random chance. P(Cells|~God) is very very small.
Debatable statement: P(Cells|God) is large.
Conclusion: We update massively in favor of God and against ~God, because of, not in opposition to, the massive evidence in favor of the existence of cells.
This is valid Bayesian updating, it’s just that the false statement is false.
False statement: Without intelligent design, cells could only be produced by random chance. P(Cells|~God) is very very small.
You’re absolutely right! This is one of the key mistaken beliefs that creationists hold. I’ve had the most success in convincing them otherwise (or at least making them doubt) using the argument given by Dawkins in The God Delusion:
Our likelihood heuristic is strongly tied to both our lifespans and the subjective rate at which we experience time passing. Example: if we lived hundreds of times longer, current probabilities of, say, dying in a car accident, would appear totally unacceptable, because the expected number of car accidents in our lifetime would corresponding be hundreds of times higher.
The hundreds of millions of years between the formation of the Earth and the appearance of life are simply much too large of a time-span for our likelihood heuristic to apply, and doing some simple math [omitted; if someone wants to give some approximate numbers that’d be nice] shows that the probability of replicators arising in that time-span is far from negligible.
I speculate there’s at least two problems with the creationism odds calculation. First, it looks like the person doing the calculation was working with maybe 60,000 protein molecules rather than zillions of protein molecules.
The second problem I’m having trouble putting precisely in words, concerning the use of the uniform distribution as a prior. Sometimes the use of the uniform distribution as a prior seems to me to be entirely justified. An example of this is where there is a well-constructed model as to subsequent outcomes.
Other times, when the model for subsequent outcomes is sketchy, the uniform distribution is used as a prior simply as a default. Or, as in this case, it’s clearly not an appropriate prior. In this case, the person is probably assuming that all combinations of proteins are equally likely (I suspect this assumption is false.)
Consider that 1) There is more than one possible arrangement of proteins which qualifies as a living cell, and that 2) the materials of which proteins are made had quite a long time to shuffle around and try out different configurations between when the earth cooled and the present day, to say nothing of other planets elsewhere in the universe, and that 3) once a living, self-replicating, self-repairing cell has come to exist in an area with appropriate raw materials and a steady energy source it will create more such cells, so it only has to happen once.
So, we’re looking at a sample size equal to, by my back-of-the-envelope estimation, the number of cell-sized volumes in Earth’s atmosphere and oceans, times the number of planck instants in a little over four billion years, times the number of earth-like planets in the universe. The actual universe, not just the part we can see.
For intelligent design to be the most reasonable explanation, the probability of life emerging spontaneously would have to be low enough that, in a sample of that size, we wouldn’t expect to see it happen even once, and, furthermore, the designer’s own origin would need to be explained in such a way as to be less improbable.
You shouldn’t use Planck times unless the protean can rearrange themselves that quickly.
If the temperature is high enough that there’s molecular movement at all, you could observe a collection of proteins every Planck-instant and see a (slightly) different arrangement each time. You might be stuck with similar ones, especially stable configurations, for a long time… but that’s exactly the sort of bias that makes life possible.
Isn’t the problem more like: they are ignoring the huge number of bits of evidence that say that cells in fact exist. They aren’t comparing between hypotheses that say cells exist. They are comparing the uniform prior for cells existing to a the prior for only random proteins existing. They sound more like they are trying to argue that all our experiences cannot be enough evidence that there are cells, which seems weird.
This is a misinterpretation. The argument goes like this:
True statement: There is lots of evidence or cells. P(Evidence|Cells)/P(Evidence|~Cells)>>1.
False statement: Without intelligent design, cells could only be produced by random chance. P(Cells|~God) is very very small.
Debatable statement: P(Cells|God) is large.
Conclusion: We update massively in favor of God and against ~God, because of, not in opposition to, the massive evidence in favor of the existence of cells.
This is valid Bayesian updating, it’s just that the false statement is false.
You’re absolutely right! This is one of the key mistaken beliefs that creationists hold. I’ve had the most success in convincing them otherwise (or at least making them doubt) using the argument given by Dawkins in The God Delusion:
Our likelihood heuristic is strongly tied to both our lifespans and the subjective rate at which we experience time passing. Example: if we lived hundreds of times longer, current probabilities of, say, dying in a car accident, would appear totally unacceptable, because the expected number of car accidents in our lifetime would corresponding be hundreds of times higher.
The hundreds of millions of years between the formation of the Earth and the appearance of life are simply much too large of a time-span for our likelihood heuristic to apply, and doing some simple math [omitted; if someone wants to give some approximate numbers that’d be nice] shows that the probability of replicators arising in that time-span is far from negligible.
Upvoted for successfully correcting my confusion about this example and helping me get updating a little better.
Edit: wow, this was a really old comment reply. How did I just notice it...