On Mon, Dec 22, 2025 at 1:38 PM Matt Mahoney <[email protected]>
wrote:

> We establish causation by controlled experiments.
>

We establish causality with lossless compression of those observations
which we consider worthy of lossless compression.  Data selection precedes
not only model selection but modeling itself.

In AIXI terms, there is a particular point in the agent's online learning
where we establish causality:

Solomonoff Induction

However, Solomonoff Induction is subject itself to subjective
decision-theoretic constraints:

1) Data selection
2) Resource allocation to the lossless compression of #1

This is true whether data selection draws from controlled experiments or
not.  Anyone who has actually done controlled experiments knows that
resulting data must be viewed within forensic epistemology:

What was the process that produced the data?

That's when measurement instruments come into phenomenology including not
just devices like thermometers, but the institutions that create the
environment for research and, of course, the scientists themselves.

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T6cf3be509c7cd2f2-Mdc970393b426aa82e16f2ff6
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to