Comment by eigenrobot | Hacker News Reader

Comment by eigenrobot | Hacker News Reader

Parent Full thread

>science is the realm with novel data

I just dont think this is true in practice. One of my industry-linked professors really turned me on to the stark difference in data availability in the private sector vs what academics (in social science to be clear!) have access to and it's night and day.

Economists frequently just rehash the same tired quarterly data sets with marginally different models or methods; meanwhile Amazon generates petabytes of incredibled detailed and realtime sales data every--what, month? And indeed they're making use of it: https://www.amazon.science/publications/new-goods-productivi...

Your model of science seems highly idealized compared to the reality that I experienced; and this isn't an ontological question, it's straightforwardly empirical. Social media sites have billions of users and graphed networks; academic psychologists have groups of 30 undergraduates. Economists have surveys and Fed data while Amazon has . . . Amazon's entire dataset, along with what was when I last checked the largest concentration of economics PhDs outside the Federal Reserve System.

I agree that industry isn't universally preferable to the academy and I definitely wasn't trying to make that claim.

I think one fallacy I see when people critique academia vs industry is that they take an example like Amazon from industry and show how it is better than the average output of academia.

To do a proper comparison one should compare Amazon’s products with top products from science: eg ontogenetics, CRISPR, gravitational waves, large hadron collider, and solar voltaics.

how about Falcon Heavy? :D

Falcon Heavy is certain an impressive engineering feat, but it doesn’t compare to the Voyager 1 (40 years ago!) in terms of new science and new frontiers.

Industry is better than academia when it comes to getting to scale and reducing costs. But usually new frontiers and new science does not often have those constraints.

> meanwhile Amazon generates petabytes of incredibled detailed and realtime sales data

The problem is that you can't use Amazon's sales data to answer research questions about maternity leave policies, UBI, or the criminal justice system. All of that data is great but it benefits a relatively small part of social science.

this deserves its own post, but the other advantage working with tech datasets on the regular is developing perspective on the methodological shortcomings of what's done at universities.

what i mean is that once i started running product experiments at scale in really hyperoptimized environments, my sense of what's realistically possible with small data sets collected by humans describing meatspace circumstances contracted fairly dramatically. I don't just mean replication crisis issues, although those are real; im instead skeptical of nearly all of the empirical work being done in nearly every social science field.

that is, im not arguing that tech does a great job facilitating research into eg maternity leave policies. rather, im saying i dont think anyone is learning much about them at all.

im happy to acknowledge that this is a minority view.

Are you mainly concerned about the quality of the data that is collected, or the econometric methods that are applied to them, or the way in which respondents are sampled?

I think that the direction that is promising is combining data sets like those that are built up at Amazon with administrative data that come from other sources. It is a combination of such data sets that opens up answers to important research questions. But I am doubtful that Amazon wants to fund the pursuit of such questions. That is not what will increase shareholder value.

Reply on news.ycombinator.com