Tuesday, 30 December 2014

Applying Occam's Razor

Occam’s Razor is an important principle of scientific epistemology. The Encyclopedia Brittanica explains:

Pluralitas non est ponenda sine necessitate, ‘Plurality should not be posited without necessity.’ The principle gives precedence to simplicity; of two competing theories, the simpler explanation of an entity is to be preferred. [1]

In this article, we will focus on a particular class of applications for Occam’s Razor: the class of applications where it is used to select the null hypothesis. The null hypothesis is the hypothesis on a question that we assume to be true, by default, unless an alternate hypothesis on the question is satisfactorily proven. To clarify this, let us proceed immediately to an illustrative example.

Illustration: The Veracity of Jack and the Beanstalk
You finish telling a child the story of Jack and the Beanstalk, a story involving giants, immense climbable beanstalks, geese that lay golden eggs, and other bizarre phenomena. The child asks: ‘Is that story true?’ How do you respond?

 Well, the veracity of the story is dependent on the source of the story. Did the story come from someone’s imagination? Or was it actually based on an eyewitness account of actual events? Thus, our question is: What is the origin of the Jack and the Beanstalk story?

Let us say that we entertain two competing hypotheses as possible answers to this question. They are:

1. The story came from the active imagination of a storyteller.

2. The events in the story actually occurred on another planet and the eyewitness account of those events was communicated to the originator of the story on Earth via interstellar, telepathic brain waves.

It is unlikely that we could satisfactorily prove either one of these hypotheses. If we treated both of these hypotheses equally, than our response to the child’s query would have to be: ‘I have no idea’. 

This is a highly unsatisfactory answer though because we do in fact have an idea: even in the absence of satisfactory proof, one of the hypotheses just seems a lot simpler and more plausible than the other, which is pretty outlandish. Applying Occam’s Razor, we select this simpler and more plausible hypothesis as the null, or default, hypothesis, and the other hypothesis as the alternate hypothesis. 

Our setup now looks like this:

Question: What is the origin of the Jack and the Beanstalk story?

H0 (null): The story came from the active imagination of a storyteller.

H1 (alternate): The events in the story actually occurred on another planet and the eyewitness account of those events was communicated to the originator of the story on Earth via interstellar, telepathic brain waves.

Because we cannot satisfactorily prove H1, H0 remains standing, and we can tell the child: ‘I strongly suspect that the story came from the active imagination of a storyteller, and hence isn’t a true description of actual events’. Much more useful and satisfying than just proclaiming utter ignorance!

Anthropogenic Climate Change
The proper selection of the null hypothesis is an important part of answering many scientific questions. One such question is that of whether human-caused greenhouse gas emissions could cause disastrous global climate change in the future if continued unabated. I would set up this question as follows:

Question: What was the primary cause of the global warming that the Earth has gone through since the 1970s?

H0: The primary cause was not human-caused emissions of greenhouse gases, but rather was a ‘natural’ cause or set of causes, that would have happened even without any human interference.

H1: The primary cause was human-caused emissions of greenhouse gases.

Why do I choose the skeptical position as the null hypothesis? Isn’t this just indicative of my particular bias? I do so because the ‘natural’ explanation of global climate change is in fact the simpler explanation of the phenomenon. This is because we know that the Earth’s climate has changed significantly and relatively frequently in the past, long before humans were in a position to emit any significant quantities of greenhouse gases. We also know that humans have been spuriously blaming themselves (or their sins) for negative changes in the weather since the beginning of human history. Thus, the simplest explanation is that this recent bout of global warming has a natural cause, and that humans are once again blaming themselves erroneously for changes in the weather, as has happened multiple times throughout the Earth’s and humanity’s history.

The selection of the null hypothesis for this question is very important, given how complicated the Earth’s climatic system is. Due to this complexity, I don’t think that any well-informed person can honestly say that either of these hypotheses has been satisfactorily proven. Treating both hypotheses equally is not a good option, because then we would be forced to say ‘I have no idea’, which is not a very useful response when deciding whether implementing anti-emissions policies would be a good idea or not. Setting up the hypotheses as I have allows us to say: ‘I suspect that most of the observed global warming has been due to natural causes, not human-caused greenhouse gas emissions, and hence we should not implement any anti-emissions policies for the moment. However, if more convincing evidence favouring the emissions hypothesis came to light, I would be willing to change my opinion accordingly.’

The Existence of God
Occam’s Razor can help us answer this age-old and much-discussed question. I would set it up as follows:

Question: Does God exist?

H0:  No, there is no omnipotent sky ghost (God) in existence.

H1: Yes, there is an omnipotent sky ghost (God) in existence.

Why do I set the atheistic position as the null hypothesis? The first reason is because, given our current knowledge about how the world works, it is the simpler explanation. We can already satisfactorily explain many seemingly mysterious phenomena by employing the principles and findings of physics, chemistry, biology, and other sciences. Given this, the simplest hypothesis is that there is an as-yet-undiscovered scientific, non-omnipotent-sky-ghost-dependent explanation for the mysterious phenomena that we still cannot explain satisfactorily.

The second reason is because it is impossible to prove a negative. If religious people had their way and the existence of God became the null hypothesis, it would be impossible to prove them wrong, no matter what evidence came to light. For this reason, it is infinitely fairer, for these kinds of ‘existence’ questions, to put non-existence as the null hypothesis.

In addition, if religious people insisted that the existence of God should be the null hypothesis, for consistency’s sake the same should apply to the question of the existence of other magical creatures. We would have to accept the existence of the tooth fairy, unicorns, talking garden-gnomes, leprechauns, and Santa Claus, for the simple reason that is impossible to satisfactorily prove that any of these things don’t exist.

A group of people called Agnostics think that the two hypotheses regarding the existence of God should be treated equally; neither should be the null hypothesis. The problem here is that a consistent Agnostic can never say that anything doesn’t exist, because it is impossible to prove a negative. To be Agnostic just with regards to God is arbitrary and illogical; to be logically consistent you have to be Agnostic with regards to any being whose existence cannot be satisfactorily proven, including the aforementioned other magical creatures, such as Leprechauns.

If you have no idea which, if any, of the scores of magical creatures that people have dreamed up over the centuries exists, how do you know which actions to take in regards to these creatures? Do you pray to God? Sacrifice to Poseidon? Pay homage to Vishnu? Follow rainbows in search of the Leprechaun’s pot of gold? In practice, most so-called Agnostics will take no action related to any magical creature, including God, and thus seem to assume, like Atheists, that these things don’t actually exist, unless proven otherwise. Agnosticism, in practice, usually proves just to be a more socially-acceptable form of Atheism, with little difference in actual content. This is for the best, as a consistent Agnostic, not sure which of the thousands of potential gods to pray to, or whether to chase rainbows and hunt Unicorns or not, would probably quickly go insane.

Government Intervention
We can apply Occam’s Razor to the desirability of political action as well. I would set up such questions as follows:

Question: Should the government engage in coercive intervention x?

H0: The government should not engage in coercive intervention x.

H1: The government should engage in coercive intervention x.

Why set non-intervention, the libertarian position, as the null hypothesis here? In a theoretically pure free-market society, every individual would be free from violent coercion or harassment when it came to taking the actions they deemed most conducive to their material and mental well-being. They would similarly be free to engage in voluntary exchange involving each others’ services or material possession, as long as the transaction was deemed beneficial and worth engaging in by both parties.

The simpler and more plausible general hypothesis is that this regime of individuals being allowed to pursue their own well-being in the best way they know how and of mutually beneficial exchanges with others will in fact be the regime most conducive to most individuals’ well-being. It is certainly simpler and more plausible than the other possible general hypothesis, which is that a regime where all individuals’ lives are completely controlled by a centralized, coercive power will be the one most conducive to most individuals’ well-being. This is especially true given the extensive and pretty one-sided empirical evidence we now have with regards to the negative effect of totalitarian regimes on individuals’ well-being.

Given this setup, if we cannot satisfactorily prove that the government intervention in question will increase the relevant individuals’ well-being more than it decreases it, we retain the null hypothesis and opt for laissez-faire in this instance.

But again; why the need for a null hypothesis at all? Why not just treat both hypotheses equally. Because either the government intervenes or it does not; there is no ‘neutral’, or ‘I don’t know’, option available. If you don’t take a position on the question at hand at all, then you’re basically just supporting whatever the majority of the more active people have decided.

The Standard of Proof
I have used the term ‘satisfactorily proven’ a number of times throughout this article, referring to what must be done in order to reject the null hypothesis and accept the alternate hypothesis in its place. The level of certainty of a proof for it to be ‘satisfactory’, or the Standard of Proof, is a subjective determination, and will vary person by person and instance by instance. For example, in criminal trials the null hypothesis is innocent, and in order for the court to accept the alternate hypothesis of guilt, it must be ‘proven beyond a reasonable doubt’. This strong standard of proof makes the null hypothesis difficult to reject. By contrast, in civil suits, the standard of proof is ‘balance of probabilities’, meaning that it doesn’t really matter what the null hypothesis is, because if there is even a shred more evidence for the alternate hypothesis than for the null hypothesis, the alternate is accepted.

How do we decide what standard of proof to set? This will depend on how much simpler and more plausible the null hypothesis is over the alternate hypothesis. It will also depend on how bad the consequences of erroneously rejecting the null hypothesis, balanced against the consequences of erroneously retaining it, would be.  

Conclusion
Null hypothesis selection using the principle of Occam’s Razor is an important part of the scientific method. For many questions, it is important to select the default hypothesis properly because we do not have the luxury of waiting for all possible evidence to be collected before making a determination. ‘I have no idea’ is a pretty useless position to take when it comes to deciding upon concrete actions. Often we must decide one way or the other without delay, and in the absence of sufficiently definitive proof one way or the other, this means relying on the default hypothesis.
  



[1] http://www.britannica.com/EBchecked/topic/424706/Occams-razor

No comments:

Post a Comment