[Osx-nutters] Denial

[Stefano Mori]

On 2007-Aug-16, at 07:23, David Cake wrote:

>> Anyway, having a hypothesis about mechanisms is of course very
>> valuable. But the thing is, having a complex model doesn't
>> necessarily mean that your predictions are more trustworthy.
>
> 	Really?

Yep.


> 	It seems to me that the model of the sunrise that says 'it
> comes up every day, just like it did before' is much less trustworthy
> than the one that includes celestial mechanics. The one that includes
> celestial mechanics works at the poles. And catches the occasional
> eclipse. Once you go to the poles, it immediately becomes more
> trustworthy than theories that don't predict this details.
>
> 	The point is, having a more complex model doesn't make it
> more trustworthy by itself - good old William of Occam - but having a
> more complex model *that explains more observed facts* IS
> intrinsically more trustworthy.

It's not more trustworthy for making predictions. Most people believe  
that it is, but it's one of those counter-intuitive things.

When the heliocentric model was proposed, the Church apparently had  
mathematicians working up ever more sophisticated models that could  
include all the inconsistencies into a geocentric model, thus  
allowing them to maintain that the Earth was the centre of the universe.

You can always add complexity to explain more freakish observed  
facts. So that's one problem.

The next problem, particularly for prediction, is that the  
complexity, even if correct, translates to greater uncertainty in  
your predictions, because small errors in the many relationships in  
your model compound themselves as you run the model forward through  
time.


Stefano