Occam has a razor, and he is not afraid to use it

You open your eyes and see an apple, you reach out and grab it in your hand and test its apple-like texture, you take a bite off of it and taste its apple-like flavor, your nose is filled with an apple aroma, therefore an invisible trickster demon with ten limbs and six horns must be filling your senses with false apple-related data.

You are given a sequence of numbers: $1, 2, 3, 4, 5, 6$, and you are asked to predict the next one. You immediately recognize the famous sequence $n + (n-1) \cdot (n-2) \cdot (n-3) \cdot (n-4) \cdot (n-5) \cdot (n-6)$, and deduce that next one is obviously $727$.

Bacon's inductive reasoning leaves us with an infinite number of possible explanations to any given phenomenon, and the abductive reasoning tells us to favor those who are more plausible, but how can we tell which one is the most plausible one? Occam's razor is an epistemological principle that claims that simpler explanations to a given phenonmenon are more likely than complicated ones. Occam's razor is not an irrefutable principle of logic, but a hueristic, a technique to infer mostly good results, with a risk of some false ones as well. A strong heursitic is one that has a high true results to false results ratio.

It is important to note that "simpler" does not mean "shorter in the English language" nor does it mean "easier to explain", but something more along the lines of: "with fewer parameters". It was originally phrased in Latin: Entia non sunt multiplicanda praeter necessitatem, meaning: Entities should not be multiplied beyond necessity. While it is easy to explain every phenonmenon with an invisible imp that causes events to unfold they way they do, this explanation raises more questions than it answers: Why is the imp invisible? Where did they come from? What are their motives? Why do they do what they do? How are they even able to do all they do?

Occam's razor helps up with another issue: Suppose you have a hypothesis that explains all of your data so far, but when you make another observation something goes wrong and the data doesn't fit anymore. You can make a small modification to the hypothesis that states that everything works the same as you previously speculated, except on the first Tuesday of each month between 4pm and 5pm. You can do so, but of course this sort of modification incurs additional complexity. I will expand more on that idea in the next post.

Further Reading

Occam's Razor on LessWrong