Thread by @Meaningness

A meandering 🧵 on how to do science, creativity and execution, political censure, research universities, Silicon Valley, and stuff, prompted by @paulg's latest post: paulgraham.com/conformism.html

@paulg .@paulg's 2004 essay "What you can't say" is a classic. The points he made may be obvious to the independent-minded, but his exposition is exceptionally excellent. paulgraham.com/say.html 2/

@paulg "What you can't say" is exceptionally relevant now, as increasingly you might be punished for saying almost anything. @paulg's new post is about that. But this meandering 🧵 isn't! 3/

Instead I want to take this one paragraph out of context, object to it, and suggest some possible implications: "To be a successful scientist... you have to be right when everyone else is wrong." 4/ paulgraham.com/conformism.html

A 2x2 on of ways of doing science: unusually vs conventionally; carefully vs badly. 5/

Popular mythology emphasizes a handful of conceptual breakthroughs by Science Heroes—mostly just guys making new theories of gravity. This is highly misleading, because it is not at all how most science works. 6/

[There's a broader point here: our understanding of science is massively distorted by taking theories of gravity as the paradigm. Nearly nothing else in science has similar dynamics of forward motion. But let's not get side-tracked into that attractive diversion…] 7/

Science progresses mainly by doing fairly obvious experiments carefully and correctly. Then we know something new, although maybe not exciting. But it's an actual truth! This should be obvious? But maybe it's not, nowadays. 8/

The Romantic Rebellion myth that science proceeds mainly by Manly Heroes Bravely Challenging The Establishment distorts incentives toward producing unexpected, false results from sloppy work. A main driver of the replication crisis. meaningness.com/romantic-rebel… 9/

Maybe we've taken for granted that science automatically works reliably, because the Scientific Method is guaranteed to be Cosmically Correct by some rational proof that no one can quite remember. (Because there isn't one.) 10/

If there were a Scientific Method, just mindlessly following the rules would be enough to guarantee that you found a truth. Anyone could do that, in which case the axis of creativity vs conventionality would be the only meaningful one. But there is no such guarantee… 11/

In medical research, where it matters most, nearly all science is done so badly that the outputs are meaningless. Creativity vs conventionality is irrelevant. journals.plos.org/plosmedicine/a… 12/

In parts of psychology that valorized the unexpected over the careful, progress now may require suppressing some sorts of creativity for a decade or so, until baseline truths can be established. See this from @siminevazire: journals.sagepub.com/doi/10.1177/17…

@siminevazire Back to OP. This *should* be false: ”To be a successful scientist... you have to be right when everyone else is wrong."

To be a successful scientist, it should be sufficient to work carefully enough that your outputs are meaningful. Too few meet that bar now. 14/

"Everyone else was wrong" is irrelevant even in most cases of conceptual breakthroughs. Usually they weren't actively wrong; they just weren't considering a factor that turned out to be important. They were looking in the wrong place, perhaps... not the same thing. 15/

It's not obvious to me that unconventionality is critical in Silicon Valley either. It's an SV truism that ideas are less important than execution. Most successful startups do something obvious and get the details right. Stripe, Airtable, Whatsapp... 16/

There is an important truth here, though. Most science *is* conventional to a fault, as well as bad. Looking in a different place does often drive rapid progress. The best science is both careful and eccentric. There's no recipe for that. (The Eggplant tries to help...) 17/

Universities are no longer a good place to do science. Corporate research labs are also no longer good.

Silicon Valley recognizes it has had no significant new ideas for too long.

"We need a new Xerox PARC!”

"We'll have to invent replacements for research universities." 18/

SV is the only live scene left with enough power (perhaps?) to get anything significant done, so the idea that it could create a new environment for sciencing is exciting! 19/

Creating better, new sciencing environments requires accurate understanding of what makes science go. That’s impeded both by pop misunderstandings of science (“defying conventional wisdom”) and SV’s own heroic mythology (“move fast and break things”). 20/

My personality is such that I couldn’t do conventional work, so I welcome SV’s impulse to fund eccentrics!

OTOH, I observe a lack of appreciation for the amount of time it takes to do research carefully enough to get meaningful results—especially, valid unexpected results. 21/

I find little awareness in SV of what research even is.

I suspect it’s a serious problem that nearly no one in the SV power scene has a PhD.

The understanding of research you get from a good PhD program is qualitatively utterly different from undergrad/MS STEM. 22/

If you have an undergrad STEM degree, you don’t know what you don’t know. You implicitly imagine that a PhD is lots more of the same.

It is—or should be—utterly different. It requires a new way of being; a personal transformation; a radically altered attitude to knowledge. 23/

[Hint: undergrad STEM is all rationality. A genuine PhD forces you to confront ontological nebulosity—and therefore to begin to think, feel, and act meta-rationally. There are other ways to undergo that transformation… but not in science.] meaningness.com/eggplant/nebul…

Silicon Valley was founded by people with PhDs. I suspect that matters?

SV was about silicon in those days. Maybe you need a PhD for ICs, but not for software?

Computer “science” is mostly useless (unlike semiconductor physics), so maybe a CS PhD is irrelevant. 25/

OTOH, many of the researchers involved in PARC’s breakthrough software developments did have PhDs (often not in CS). Maybe that’s significant? 26/

Maybe what was important was not that they’d learned stuff about software, but that they’d undergone the meta-rational transformation… which is what enables you to conjure new or improved rational systems out of nebulosity. 27/

Better understanding of what makes for good and bad research, including institutional factors, is a prerequisite to creating better research environments. We don’t know nearly enough about that.

Some intuitions from SV culture probably can help, but many would misfire. 28/

Some of academia’s best are leaving because universities have become intolerably awful. They could be a critical resource for understanding what would be better.

But, they (we) are the ones who find institutions most oppressive, and may be constitutional misfits. 29/

Collecting a Salon des Refusés might be a good start—but it’s probably not the way to build a durable replacement for the research university. en.wikipedia.org/wiki/Salon_des… 30/

Any academic researcher can tell you some of what’s needed for a better environment. Less administrative bullshit, more time to think, less pressure for short-term results.

SV will immediately grok the first of those, less the second, & may be incapable of allowing the last. 31/

I believe ethnography could uncover significant new insights into the environmental factors that make for productive research scenes.

Interview the participants. None of us know the whole story, but many who’ve done breakthrough research have anecdotes that would add up. 32/

I say “scenes” because they drive most important research. That’s why a Salon des Refusés, and funding individual eccentric researchers, is inadequate (though worthwhile). meaningness.com/metablog/upgra… 33/

A scene, no matter how productive, peters out after a few years. Research needs a balance of disruptive innovation and institutional stability. The Rad Lab, Bell Labs, PARC, and the others—they had that. 34/

The personality types & capacities required diverge. You need straight-ahead ops people for competent administration and occasional injections of sanity. You need weirdos with repellent haircuts & sometimes repellent ideas. You need stick-in-the-muds who say “but is it true?” 35/

In productive research environments, those personality types constantly clash. Everyone has to reluctantly tolerate the conflict, because you can’t do without any of them. 36/

There has to be sufficient grudging respect between the admins, creative weirdos, and careful conventionals.

There has to be mutual acknowledgement of excellence in their differing spheres.

That requires actual excellence—but also leadership. 37/

Someone who everyone involved basically respects has to model collaboration, and has to tactfully coerce people into to working together who would rather not. That person has to have research credentials, operational competence, and realism about what’s feasible. 38/

In retrospect, I’m awed by Patrick Henry Winston’s ability to do that at the MIT AI Lab, and John Seely Brown’s at PARC—the two clearest examples I’ve experienced personally, although at the time I was too callow to appreciate that sort of work at all. 39/

If SV wants to create a new research institution, finding that sort of leader may be a prerequisite.

But maybe such leaders are themselves products of an informal scene. They rise to the occasion because they see the necessity, and transform themselves into that. 40/

I’ll stop thinking out loud now. I did warn that this would be meandering! But I didn’t know it would go on for hours.

This stuff is important, though. We’re eating the seed corn.

Progress matters, and requires discovery and invention. We’re decreasingly capable of them. 41/41