Monday, February 23, 2015

The Ensemble Model of Religion

Religious beliefs are a lot like medium-range weather forecasts. They're our best educated guesses about things about which we don't know for sure and don't have much direct, indisputable evidence. The evidence we do have can be interpreted in a variety of ways. Rather than expecting to get everything perfectly right, the main goal of medium-range forecasts is to minimize the error. I think that's a good goal for religious beliefs too. And I think a similar method can help.

The human mind is hopelessly biased, especially when it comes to religious beliefs. We have all kinds of [often subconscious] motivations that lead us believe what we want to believe. We think our beliefs are based on evidence and the beliefs we reject lack evidence. But our interpretations of the evidence are so infected by confirmation bias and other biases that we often come to opposite conclusions when evaluating the same evidence.

Weather forecast models also have errors and biases. One of the best ways to minimize them is to use ensembles. Ensembles are collections of different forecasts based on slightly different initial conditions and/or model physics. They rely on the principle that the ensemble mean (i.e., the average of all forecast solutions) has, over a sufficiently long period of time, less error than any single ensemble member (i.e., an individual forecast). Biases of individual members tend to cancel each other out and their unique errors tend to be somewhat corrected by other members with different solutions. On any given day, a few of the members might be more accurate than the ensemble mean, but we seldom know which one will be the best until it's too late.

So it is with religion. There's a wide variety of flawed, biased beliefs, many of which contradict each other. Some believe there is one God, some believe there are many Gods, and some believe there is no God. They can't all be right, but they all might have some insight that others don't have. Attempts to determine which beliefs are most accurate are inevitably contaminated by a plethora of cognitive biases. But we don't have to give up and adopt total agnosticism. Fortunately, in situations where the evidence is ambiguous and there are multiple conflicting answers, science gives us a reliable default solution: the ensemble mean.

The most straightforward way to define an ensemble model of religion is to consider the beliefs of each person on Earth as an ensemble member. Everyone gets one equal vote. The ensemble mean would be the “average” of everyone's beliefs. This generally would be similar to the world's most common beliefs, with moderate/centrist positions in areas of disagreement. Using this definition, here are a few beliefs that I think would represent the worldwide ensemble mean:
  • There probably is a God.
  • There probably is only one God.
  • That God probably is the God of Abraham, as originally described in the Torah.
  • God probably created animals and humans via the process of Evolution.
  • There probably is some kind of life after death.
  • There probably is something uniquely special about Jesus of Nazareth.
  • Unique doctrines taught only by particular sects within Christianity, Islam, etc. probably aren't true.
  • Extreme fundamentalism and extreme theological liberalism probably aren't the best interpretations of holy texts.
This “ensemble mean”, based on global religious statistics, is consistent with beliefs that are largely based on biblical Judaism, influenced by Christianity, Islam, and (to a lesser extent) smaller religions, and contain a relatively small but still significant dose of secularism and atheist skepticism.

To be clear, I'm NOT saying the majority is always right or that truth should be determined by popular vote. The majority has been wrong many times throughout history. The centrist position also has a long history of being wrong. As with ensemble forecast models, the mean tends to smooth out important details and minimize extremes that some members might be correct about. The point simply is that in the absence of compelling evidence, the ensemble mean is the best starting point.

It's natural for humans to think there's strong evidence when there isn't, or vice-versa. It's also natural to think we have insight that other people with different beliefs don't have, perhaps because we're more intelligent, had more-relevant life experiences, or are more educated in science, philosophy, or religion. Though these may indeed be useful in evaluating certain verifiable beliefs, they don't provide consistent non-circular answers to fundamental questions such as “Is there a God?”. Much smarter, more experienced, more educated people than you or I have come to opposite conclusions about such questions.

Another natural inclination is to believe other people's biases are stronger than our own – which itself is an especially pernicious bias known as the “bias blind spot”. It's easy to think of reasons why others' religious beliefs are biased – e.g., growing up in a particular religious environment, indoctrination, bad experiences with religious people, not wanting to accept that one's behavior is sinful, fear of death, etc. It's much harder to recognize biases in ourselves, some of which we're not even consciously aware of.

Though I believe the “ensemble mean” of religious beliefs is the best starting point in the absence of compelling evidence, I don't think we're stuck there. As I think I've shown in previous posts, evidence does exist, and it should shift our position away from the mean. My beliefs deviate quite a bit from the mean sometimes [as anyone who knows me can attest]. But it's something that I think we should be very careful about. Deviating far from the ensemble mean requires strong evidence. It also requires a lot of faith in one's own ability to overcome cognitive biases. That ability, at least in my case, is inconsistent at best.


  1. It's impossible to critique this without seeing the actual global statistics, but I sense strongly that the major religions would constitute a multimodal distribution, thus rendering the ensemble mean misleading at best.

  2. OK ... so we start with the ensemble mean, which you've already said doesn't mirror your own position perfectly. You've already said we all suffer from various cognitive biases, so just how do you think we should proceed from that starting point? Logic has certain requirements, of course, but you seem to be saying our biases render the process inevitably questionable. I grant that no one should assume they are the possessors of absolute truth, and everyone should be open to accepting contrary evidence, but it's quite natural to be dubious regarding that contrary evidence - elsewhere, I've tried to make clear what sort of evidence I would see as compelling for the hypotheses regarding your estimated ensemble mean. What you've offered as evidence has failed to be compelling, imho, and seems mostly to center around a "majority vote" argument.

  3. Josh, global religion statistics are available from Pew Research. Approximately 55% of the world's population is either Christian or Muslim, which explains most of my generalizations about the ensemble mean.

    Chuck, I didn't say it's impossible. I do believe there is strong evidence for reducing the weight of Islam, increasing the weight of Judaism, denying particular teachings of some branches of Christianity, etc. I haven't really discussed those yet but may get there eventually.

    If you believe the evidence isn't compelling, wouldn't that be a reason to stick close to the ensemble mean? Why should an outlier (i.e., atheism) be the default rather than the ensemble mean?

  4. Why? Because atheism is what I think matches the evidence better than your ensemble mean, or any of the deist religions. The absence of any compelling evidence in favor of the Abrahamic deity is, imho, strong evidence for the non-existence of said deity. Moreover, the Abrahamic deity, as it was explained to me, made absolutely no sense, so I never bought in, and that concept still doesn't make sense to me. I feel no compulsion to join the herd in their consensus regarding religions.

    I'm pretty sure you don't accept the ancient Egyptian, or Roman, or Greek, or Celtic, etc., gods with anything more than negligible probability. I share those strong doubts, but evidently, the difference between you and me is that I include the Abrahamic deity in that same league of discarded deities.

  5. Chuck, I understand that that's how you see the evidence or lack of evidence. The majority of the world disagrees, probably in large part because their biases are different from yours. The question is, why should Chuck's biased view be the default starting point rather than the mean of everyone's biased views?

    Your second paragraph reminds me of what I call "the Thor argument", which I've been planning to write about in my next blog post. In the context of a global ensemble model, the answer is very simple. People who believe in ancient Egyptian, Roman, Greek, Celtic, etc. gods are given exactly the same weight in my model as Christians and Muslims. There just happens to be far fewer of them, so they don't have a significant impact on the ensemble mean. If billions of people believed in those Gods, they'd have more weight and I'd pay much closer attention to what was written about them.

  6. This is repeated from the FB debate: Regardless of how you mash the potatoes, belief and non-belief are diametrically opposed world views...they share no common bond of belief and don't belong in an ensemble that purports to verify aspects of belief. If you run an ensemble to forecast weather probabilities, adding a single model that forecasts the likelihood of the spread of Ebola makes no physical sense at all.

    This ensemble is set up to verify the biases of believers while essentially erasing the opposing view of non-believers. Sorry, but this is an invalid ensemble at its very core.

  7. Josh, in the ensemble I described, believers and non-believers are both given exactly the same weight. The mean leans toward the believers' solution because there are more of them, just as a weather forecast ensemble with 7 members forecasting rain and 3 members forecasting no rain would lean a bit toward the "probably will rain" side.

  8. Sorry if that has been already answered, but how do you account for the fact that the "ensemble mean of religion" has probably varied wildly through human history? This is not a forecast in the traditional sense, but more of a survey from my understanding.
    So if the mean varied wildly since the dawn of human beings, why is it even relevant? It's not like the computing basis has improved over time as with the ensemble analogy to meteorology.....

    1. Good question. I think as human knowledge, experience, and access to information have increased with time, the ensemble mean has gotten more accurate, so I would pay more attention to the current consensus and perhaps learn something from the trend too, which, over the past 3000 years or so, has moved very much toward the God of the Bible. It's similar to looking at climate models from 20 years ago, which aren't as good as the ones we have today. The fact that their forecasts were so different back then isn't a reason to throw out the models we have today and call them "irrelevant". You can only go by the best information you have at the time. So if we were living 3000 years ago, we should've started with the mean from back then, which would've given us a different starting point that we'd have today. Keep in mind, the mean is only a starting point. If there is any evidence that isn't properly accounted for by the ensemble members, we should make appropriate adjustments away from the mean.

  9. Your blog contains lots of valuable data. It is a factual and beneficial article for us.pushpanjali mantra Thankful to you for sharing an article like this.