11-10, 09:00–09:45 (America/New_York), Central Park West (6th floor)
The Jupyter ecosystem offers powerful tools for authoring documents that run code. An obvious application coming out of the reproducible research movement would be for academics to start distributing their working papers as Jupyter notebooks instead of PDFs. But authors won't switch to notebooks until Jupyter is ubiquitous in the specific sense that every interested reader has access to an open-source reader/runner for notebooks that is (1) user-friendly, (2) robust, (3) complete, and (4) safe.
-
User-friendly means that neither installing nor using the reader/runner requires use of the command line.
-
Robust means that installing other software will not break it. (We'll have fun talking about how far the Python ecosystem is from this requirement.)
-
Complete means that authors can use the entire Python standard library and all the external libraries used by researchers.
-
Safe means that running notebooks should be safer than opening PDFs.
Safer than PDFs is not a high bar. PDFs are a dangerously effective vector for distributing malware. Nevertheless, safety is by far the most challenging of the four goals.
In my talk, I'll demonstrate an open-source solution that is user-friendly, robust, and complete. It follows in the footsteps of Jupyter Desktop instead of Jupyter Lite. Jupyter Lite, which can run some notebooks in the browser, is an exciting complementary effort that will support progress toward the ubiquity of Jupyter, but any solution based on browsers and web assembly will struggle to support all Python libraries and is likely to achieve the modest safety of the browser only by sharply limiting interaction with the local file system.
The crux of the challenge is safety. The usual strategies -- scanning for known malware and hobbling and patching the apps we use to interact with digital files -- are already failing. They don't stand a chance in a world where people exchange documents that are supposed to be capable of running arbitrary code.
The viable alternative -- a reputational equilibrium -- relies on two principles that humans have relied on since the Pleistocene:
-
Authentication: Know your colleagues.
-
Selection: Interact preferentially with people who have not demonstrated that they are unreliable.
These two principles also explain why science is such an amazingly effective system for cooperating at scale. Traditionally, we had to rely on face-to-face interaction for authentication. Asymmetric cryptography lets us extend a reputational equilibrium into cyberspace.
The software that runs notebooks can sign and verify documents. Crucially, it can also help users manage secret keys. To make Jupyter and digital signatures ubiquitous, the remaining hurdle is the "public key distribution problem." I could be wrong, but it seems obvious to me that we can solve this problem if we stop philosophizing and just do it.
And what about "Making Billions from Crypto"? That was click-bait. By building on the shared norms of science and open source software, the program I'll outline really could "make the world a better place"; but it will succeed only if we make sure that this time, insiders won't be able to use it to take billions from everyone else.
No previous knowledge expected
Paul Romer, a University Professor at NYU, was co-recipient of the 2018 Nobel Prize in Economics Sciences. His work lies in the intersection of economics, innovation, technology, and urbanization. The central conclusion is that there are many feasible ways to speed human progress. Before coming to NYU, Paul taught at Stanford, and while there, started Aplia, an education technology company he later sold to Thomson Learning. Prior to his current role at NYU, Paul taught at Stanford, UC Berkeley, the University of Chicago, and the University of Rochester.