[FoRK] On the unreality of bottom-up brain simulation

Russell Turpin russell.turpin at gmail.com
Wed Aug 18 15:08:51 PDT 2010


On Wed, Aug 18, 2010 at 3:59 PM, Jebadiah Moore <jebdm at jebdm.net> wrote:
> Isn't that one of the main points of computer modeling?  That by figuring
> out the underlying laws and running the thing, we can see what happens
> without having to know in advance? ...

Generally you have to know some cases in advance, empirically, as part
of model validation. What modeling does is then let you generate
answers for the other cases. Though there is the tricky question of
knowing the valid scope of the model.

On the "understanding" question, let me push it from another
direction. One meaning of "understanding" for a problem domain is that
the person who understands has a mental model of the domain, and knows
well how that model fits. For example, someone has a certain
understanding of orbital dynamics if they know Newtonian mechanics,
know where it applies, and know where it fails. Computer simulation
does two things. First, it provides faster calculation. Second, it
moves part of the model from the human mind into computer software.
Which creates a variety of relationships between the model and the
people who use it. For some the model will be a black box, that
generates an answer given a problem case. At the other extreme, the
way most modeling is done today, are those who create the model. They
generally understand the internal workings of the model. The model's
behavior might surprise. But they know how to trace that behavior back
to the internal workings, which often is important to determine
whether the surprise is a bug or an interesting aspect of the actual
domain. And it seems to me that that IS an understanding of the
problem domain.

Of course, more and more kinds of models are getting created
automatically. At which point, the understanding does lie in the
software.



More information about the FoRK mailing list