post interesting tweets
Why this matters: I’ve been thinking a lot about the future of research in biology and am unsure, as complexity of networks rises, whether specialized knowledge in the form of understanding each node in a network makes sense. Can humans really keep vast, complex, non-linear networks in their head in a meaningful way? Or, as perhaps this Google team suggests, we should build models without consideration of what we “know” as it’s woefully and misleadingly incomplete.
You could have built a network that is guided by some rule-based component representing individual nodes of information from the ‘Protein Folding’ ontology. E.g. “When you see X sequence -> contort in this way”. But it seems very likely to me that those rules would be wrong to meaningful degree.
Rather, just throw the network in a black box. Get comfortable with not knowing discrete bits of information, but strive for actually solving problems. A neural net representing a solution is knowledge, just not in a way that we’re used to and comfortable with. I don’t think that makes it any less real.
Giant wall of tweets incoming warning /
Great tweets but I’m not sure the mobility one means much. Maybe people were much more uniformly distributed across the US but they moved to the centers of the highest opportunity (which are static) in the last 50 years so they naturally are moving less. In fact, in the perfect world, we expect mobility to be very low.
100% agreed that the interpretation of those statistics is not immediate. I do think it’s an interesting stylized fact that it seems moving within the same county is what is driving the aggregate trend, in an accounting sense.
(Though I do wonder, if you looked at the same graph in percentage terms, if maybe it wouldn’t look like within-county moving is driving things. In other words: maybe the percentage decline in within-county moving is the same as the percentage decline in within-state moving)
Thanks for the tweets. Very interesting thread by Baez and this reply especially: