I've used this example when teaching students about models used for predictive power vs. models used to better understand mechanisms:
You have a model that predicts with great accuracy that rowdy teens will TP your house this Friday night, so you sit up late waiting to scare them off.
You have a model with less predictive power, but more discernible parameters. It tells you the parameter for whether or not houses have their front lights turned on has a high impact on likelihood of TP. You turn your front lights on and go to bed early.
Sometimes we want models that produce highly accurate predictions, sometimes we want models that provide mechanistic insights that allow for other types of action. They're different simplifications/abstractions of reality that have their time and place, and can lead you astray in their own ways.
Can't you run the model in reverse? Brute force through various random parameters to the model to figure out which ones make a difference? Sure, it could have absurd dimensionality, but then it would be unlikely one could even grasp how to begin. After all, AlphaGo couldn't write a book for Humans about how to play go as well as it can.
That's what model interpretability research is. You can train an interpretable model from the uninterpretable teacher, you can look at layer activations and how they correspond to certain features, or apply a hundred other domain-specific methods depending on your architecture. [0]
Sadly, insight is always lost. In a noisy world where even with the best regularization, some fitting on it, or higher order features that describe it, is inevitable for maximizing prediction accuracy, especially if you don't have the right tools to model it (like transformers adapting to lacking registers [1]) and yet a lot of parameters within chosen architecture.
What's worse, bad expectations are often much worse than none. If your loan had been denied by a fully opaque black box, you may be offered recourse to get an actual human on the case. If they've trained an interpretable student [2], either by intentional manipulation or by pure luck, it may have obscured the effect of some meta-feature likely corresponding to something like race, thus whitewashing the stochastically racist black box. [3]
This reminds me of another thing I use when teaching: a perfect model of the entire world would be just as inscrutable of the world itself.
I think having multiple layers of abstraction can be really useful and have done it myself for some agent-based models with high levels of complexity. In some sense, these approaches can also be thought of as "in-silica experiments".
You have a model that is complex and relatively inscrutable, just like the real world, but unlike the real world, you can run lots of "experiments" quite cheaply!
"Tuition reimbursement" is definitely an accounting gimmick, at least once you're done taking classes. After qualifying exams, the idea that a PhD candidate is receiving anything that you would pay tuition for is laughable. You're doing research and teaching, both of which are things a professor does, albeit at a smaller scale and with some oversight/guidance, but that is far more akin to having a manager than having a course instructor. I'm all for including tuition as a concept when you're still taking courses, but after that point it makes no sense.
I think the typical argument about tuition being an "accounting gimmick" is that it's the univerity paying itself for something -- does it actually come out of a budget, paid for by other students? (Unclear.) If there were no "tuition", would that money instead go to the student? (Unlikely.)
The question of whether senior doctoral students who aren't taking classes anymore should be paying tuition is a good one too! In the case described in the original article, though the student was in a masters program and quite likely taking classes—so not quite this scenario.
Anecdotal data from the grad programs in my area is that at least for PhD students your supervisor pays their tuition from whatever funding source they use to pay the stipend.
Yes, and the supervisor also pays a fraction (often 50%+!) of any incoming grant money to “overhead” — to the institution, for lab space, staff, operations, etc.
Why some of this money is categorized as “tuition” and other money as “overhead” is at the root of the question I think.
Pretty skeptical of a floating "3.5%" figure without any additional context. NYT estimates that 15-26 million people participated in the protests following George Floyd's murder. That breaks the 3.5% threshold, and I don't think we have seen a whole lot of serious police reform in the US since then.
There are tons of other insect groups that could be very easily described as similarly "minimum viable" that don't have nearly the diversity. Abundance of some group doesn't necessarily correlate with the speciation within that group. Ants are an exceptionally successful type of insect with orders of magnitude fewer described species.
I think this is much closer than the "they're a very good blank slate". There are plenty of exceptionally successful groups of organisms with far less diversity. The point is not how successful beetles are, it's how differentiated they are. Something about their ability to occupy niches that promote isolation and therefore speciation has to be involved.
The article mentioned that they diversified early due to the diversification of the first flowering plants, so re-radiating into each others' niches over the following hundred million years could certainly help that while keeping species distinct.
> The point is not how successful beetles are, it's how differentiated they are. Something about their ability to occupy niches that promote isolation and therefore speciation
It's also how differentiated (and isolated) the habitats themselves became over 100s of millions of years of climate change and plate tectonics.
Lots of beetles, but almost certainly even more wasps! Parasitoid wasps attack pretty much every known insect species, even other parasitoid wasps. If there's not a known parasitoid for a given insect species, you usually just haven't looked hard enough. Given that parasitoids tend to be specialists, attacking one or only a few other species, the math works out to there being more parasitoids than anything else around. Great paper on the topic here: https://bmcecol.biomedcentral.com/articles/10.1186/s12898-01...
You have a model that predicts with great accuracy that rowdy teens will TP your house this Friday night, so you sit up late waiting to scare them off.
You have a model with less predictive power, but more discernible parameters. It tells you the parameter for whether or not houses have their front lights turned on has a high impact on likelihood of TP. You turn your front lights on and go to bed early.
Sometimes we want models that produce highly accurate predictions, sometimes we want models that provide mechanistic insights that allow for other types of action. They're different simplifications/abstractions of reality that have their time and place, and can lead you astray in their own ways.
reply