4 ms·
A better indicator of quality of a model is to ask, "is this model useful?". At the end of the day, models are representations of another system. Like tools, so
by briantakita 10y ago
A better indicator of quality of a model is to ask, "is this model useful?". At the end of the day, models are representations of another system. Like tools, some are more useful than others.
- haberman 10y agoOk, so I propose (for the sake of argument) the following model: any message posted on an electronic message board that casts aspersions on "The Establishment" was actually written by an AI running on a computer in Siberia. Let's suppose that many people find this model useful, because it allows them to dismiss such posts without having to critically engage them. They downvote/flag all such messages, and they feel good for having done so because they believe they are resisting an attempt to manipulate their opinions. Do you have any problem with this?
- briantakita 10y agoIt depends ;-) Useful to what systems & what aims? What's important to an Entity to engage what systems?
- paulddraper 10y agoAI needs tuning. Doesn't seem to be reading the posts it's relying to. ("useful, because it allows them to dismiss such posts")
- slavik81 10y agoUseful in that context is typically used to refer to being able to make accurate predictions about the future.
- wrsh07 10y agoDoes it need to make accurate predictions or just inform good decisions? I think a model can be bad at the former but good at the latter.
- logicallee 10y agoIt's useful to think that God will reward you for hard work (or this model gives statistically useful, actionable, and accurate results and makes useful, actionable, statistically accurate predictions) but it is not an accurate model for the mechanism by which external changes in your life occur. I vote against "is this model useful" as an indicator of the quality of a model.