Bayesian Word Salad

Matt Dillahunty posted a video as part of the Atheist Debates Patreon Project of a debate he participated in with Blake Giunta. You can see the debate here, and also take a look at the Q&A which follows here. 

Matt is fairly predictable in that he's making the same arguments he always does. But he has the advantage of his arguments making sense so he has far much less need to change it up. But the arguments coming from Blake are a bit novel.

Here he attempts to use a statistical tool known as Bayes' Theorem to infer a high probability that God exists. This is, at least in principle, the right approach. In real life when all of your conclusions are tentative - the best guess you can reasonably make after considering all the evidence available to you - if you wanted to actually quantify your uncertainty, you would want to use Bayes' Theorem. It essentially amounts to a deductive argument backing up inductive reasoning.

This stands in stark contrast to what you normally hear. Typically the attempt is to "deduce" the existence of God logically. Deduce is in quotation marks because the arguments usually amount to window dressing disguising that they are the argument from ignorance or a form of circular reasoning that sneaks in God as an unstated assumption.

He is on the right track and deserves props for trying to both explain and apply it within the space of a 30-minutes presentation. Richard Carrier gave a Skepticon talk explaining the principle as it related to his particular branch of Jesus studies. That took longer for just the explanation. His actually application of the method became a 700+ page tome released a few years after. And while he started on the right track, he veers off it fairly fast.

It begins by a length proof by intimidation (a bunch of philosophers have proved a bunch of stuff that, take my word for it, tell you God is necessary.) The entire exercise is a strange contradiction in terms. If you're starting with a (confusing and unconvincing) argument that God is a necessary being, which means She must exist, then why is that information being fed into a probabilistic calculation to increase the probability that She exists? And if that's the case then the prior probability he starts with should be the probability that something which necessarily exits in modal logic actually exists in the real world.

The problem arises when you try to map the properties (like maximally great) onto the real world. Given what we know about the laws of nature, it's not clear whether greatness is limited by them, or if greatness includes the ability to violate those laws. For example, could a maximally great being send information faster than the speed of light? And if it can't, then does our ability to imagine such a being make the term maximally great itself contradictory?

But he's not even providing evidence for the right starting probability. Because the compliment of his Hypothesis (not H) is not metaphysical naturalism; it's [every other conceivable god or metaphysical naturalism]. So that long winnowing down process he goes through where you multiply fractions together to get that tiny decimal, that also needs to happen as you winnow down monotheistic gods vs. polytheistic gods, good gods vs evil gods, deist gods vs. theist gods, gods who have an objection to shellfish vs. those that don't, gods who are omnipresent but not omniscient. He's smuggled in a boat load of assumptions without evidence which lower his starting probability.

It's interesting why he's able to do this so easily. (At least why it feels so easy if you found his talk compelling, and why, if you don't take the assumption that a god exists for granted, why so much of his talk sounds like word salad.) We've been tweaking and refining the definition of God for so long, we've forgotten that the entire concept of God is a bad ad hoc hypothesis. But the concept was assembled so long ago, that we take the long list of properties that are assumed to go with him as just part of the definition of god (just go listen to Stephen Law use perfectly similar arguments to defend the existence of a perfectly evil god just to realize how little we notice this sort of piggybacking in every day thought.)

Ultimately it turns out that the reason his presentation is so confusing is he's not even defending the God he claims to be defending:

God has a lot in common with an electron but is even simpler.

If the creator of the observable universe and personal master of our individual destines can be simpler than an electron then it's unclear how this discussion can even take place when words can spontaneously take on their opposite meaning whenever necessary to aid the discussion. And you can easily imagine a related discussion where the seemingly incomprehensible miracle which is consciousness would be brought forth as evidence for the existence of an all powerful god, but here he's just argued that these properties can exist "simply" in a disembodied mind that exists necessarily.

The truly funny thing he hasn't grasped is that he's not even asking the right question. It's a common misapplication of probability. Conditional probability (how probability changes when new information is introduced) is almost entirely counterintuitive. And here it's fairly obvious that Blake doesn't understand math, he's just following his intuition towards his pre-determined conclusion. The famous example of this mistake appeared in the OJ Simpson Trial (how's that for #tbt). It was explained in greater detail in a great, but now fairly old book What the Numbers Say. Here's the short version.

Jury is presented a statistic which is that it is relatively improbable that an abusive husband will kill his wife. The conclusion is then supposed to be, it's improbable that it happens, therefore it's likely he is innocent.

But the truth is the jury was handed the wrong statistic. It turns out the statistic they needed was a conditional probability. If a woman is in an abusive marriage and is killed, what is the probability that the husband did it? It turns out that probability is really, really high.

Blake makes the same fundamental mistake. He spends his entire time exploring the probability of god existing when that's the wrong question. The right question (which Matt approaches about 33:00 into the Q&A) is

A conscious observer looks out at the universe and wonders how she got there. What is the probability that observer is there because of the plan of a conscious disembodied mind with magic super powers?

The answer turns out to be surprisingly low because the universe we observe is completely different from what we would predict on theism.

It would be interesting to see someone actually take a stab at making the Bayesian argument for God properly to see what they come up with. A huge problem with most of these arguments is they contain the tacit assumption that the universe we observe is inevitable; that things had to turn out this way and couldn't be any other way. If you approach things with the mindset that the unstated purpose of the universe is to create questioning observers (like us!) The thought that a universe that doesn't contain philosophers asking why they're here is somehow not an acceptable universe. If you do that, all of the Bayesian estimates you make will be massively balanced.