To get in touch with me, I can be reached by email at: jd@jdpressman.com
If you would like to send me an encrypted message, my PGP key is available here.
Of the pages I have added to this site since it went live in 2014, the about page has probably been the most frustrating to write. It is a frustration that I have been met with each time I go to revamp the site and each time I have at least somewhat deferred the task to a later self. Part of the frustration is structural: There is a sense in which a well designed personal website should make an about page redundant. The entire site is already about me. Part of the frustration is depersonalization: Someone in the habit of thinking about projects, philosophy, history, fundamental questions and society probably finds themselves boring and infrequently considered. That leaves the about page in an awkward position, it either becomes superficial or too personal. At risk of the latter here is one of several narratives that could compress my life up to this point into a unified theme:
I think the first time I interacted with 'open source' or 'hacker culter' was when I was 12 and saw a forum post describing the RepRap, a low cost 3D printer with the explicit goal of being able to print all of its own parts. While I was excited about the concept of 3D printers for years afterward, I've never actually owned one. A few years later when my Windows XP install started failing I didn't have a disk to reinstall it. I recalled another member of that same forum's advocacy of Ubuntu, a free Linux based operating system that was supposed to be an upgrade over XP. It was the idea of a live CD that made me take the plunge. Being able to boot the computer from an external disk was a mind blowing concept to me. Sure enough Ubuntu was in fact much faster and smoother than Windows XP (not exactly difficult) and after some initial difficulties I've been a Linux user ever since.
But the part that gives insight into me is the consistent thread of interest in projects like Linux, RepRap, OpenStreetMaps, Wikipedia, OpenPandora, RockBox and PinePhone. I keep coming back to the RepRap, to that first interaction in particular. What I find difficult to articulate in retrospect is exactly what it was that I was so excited about. Besides the cool factor, it's not like I or anyone else I knew had a desperate need for more plastic junk. It can't be that I was interested in free stuff, since many of these projects cost money that was precious to me as a teenager. Nor was it that they were easy, since things like RockBox took more effort than just using the stock iPod/Sanza firmware.
I think what excited me was an economic intuition I have that you may or may not share. All of these projects are unobfuscated systems, they are products that have a different lifecycle than things like Windows or an iPhone. Proprietary technologies are essentially a subscription service. You're not expected to understand or repair them, and you are fundamentally dependent on the manufacturer for updates and improvements. This relationship has a basic principal-agent problem where all development paths are the ones most beneficial to the makers rather than you. The convergent outcome is we develop an adversarial relationship with our own tools. When Richard Stallman first started talking about this it was an easy conclusion to dismiss from a strange man with offputting manners. But when farmers are in an adversarial relationship with their own tractors, reliant on flaws in the manufacturers security to repair the essential machinery of agriculture that undergirds our civilization we have to take a step back and admit he was basically right. His predictions in the 1997 short story "The Right to Read" have already come to pass in embryonic form in higher education. When I went to college there were digital textbooks you needed to pay subscription access to for the quarter, and if you didn't pay for this access from the publisher you couldn't finish your classes. Even if you agree with Bryan Caplan that contemporary college is mostly a wasteful costly signaling mechanism, it should deeply concern us that this practice has been normalized for an entire generation of college students. Despite his essential correctness Stallman is an ineffectual advocate for his ideas. His overfocus on the concept of liberty is rhetorically obnoxious and fails to engage with the moral language of most people. As Jonathon Haidt says, this is a typical libertarian failure mode.
Liberty may just be a word but principal-agent problems are very real. I have spent my entire career so far working on them in one form or another. When I was a teenage Linux advocate I was trying to get people out of accepting worse and worse terms of service from Microsoft. When I worked on cryptocurrency I was most excited about the potential for smart contracts to provide neutral 3rd parties to smooth many otherwise tricky kinds of trade. And now working on AI and the AI alignment problem I am once again thinking about princpal-agent dynamics. Deep learning presents an interesting variant in that in principle every computation in these networks occurs in program memory. We can look at every thought these networks have but the format is opaque to us. Unlike in humans where our only option is to create clever economic incentives that align peoples interests, with AI we can make the principal fundamentally trustworthy. With the right development trajectory I think that deep learning latent spaces, BCI, and interpretable models will begin to really let us work together again as a whole society. We will be able to trust and collaborate the way we did before postmodernism, but without the premises resting on rotten foundations as they previously did.
The following books significantly changed the way I think. I recommend them often.
Silence On The Wire by Michel Zalewski (ISBN-13: 978-1593270469)
This book more than anything else I've read shows the fundamental habits of thought that go into defeating computer security. Zalewski takes the reader through a tour of the technologies that underlie computer networking and communication. Along the way, he points out design-level vulnerabilities in systems that make them susceptible to attacks and information leaks. These usually aren't the kind of exploits you can patch, they dig deep into the fundamental structure of the system itself. He shows that by carefully analyzing whole systems in as much detail as is possible, you can find the places where false assumptions get made, coordination issues wreak havoc and useful abstractions break apart into thorny security issues. Highly recommended to anyone trying to cultivate the "evil bit", or whose work involves designing secure systems.
Silicon Dreams: Information, Man, Machine by Robert W. Lucky (ISBN-13: 978-0312055172)
Silicon Dreams is the sort of book you dearly wish were available in an updated edition. Using information available to him in 1989 Robert Lucky attempts to sketch the fundamental limits of human computer interaction. To ground this analysis Lucky uses Shannon's information theory, which lets him clearly examine the information processing abilities of humans and computers. While the computer side of this equation has changed significantly in the last 30 years human capabilities have not, and it's human abilities that create the bottleneck. Lucky's wise choice of information theory for his analysis has kept his thoughts relevant and useful through 30 years of transformative change in the computer industry. I suspect this book will not be obsolete until humanity has become a true cyborg, and begins moving past its traditional information handling limits. Even then, the perspective it uses is fascinating and seemingly applicable to more kinds of systems than just human computer interaction. I would love to see a similar book that looks at the information processing capabilities of human organizations, for example.
Call Center Management On Fast Forward by Brad Cleveland (ISBN-13: 978-0985461119)
Don't let the title fool you, this book is relevant to much more than just call centers. Brad Cleveland understands the subject deeply enough to see a call center for what it is: A resource management problem that arises whenever a system has to deal with randomly distributed queues. In this stunningly accessible distillation of best practices for running call centers Cleveland takes the reader through every aspect of designing a system that smoothly handles fluctuating resource demands. Beginning by defining a problem that seems too complex for rigorous analysis, Cleveland shows the reader tools they can use to define the 'physics' of call center management. There is little fuzziness here, this book lets you answer questions like "how many resources are required to get this level of performance" with confidence. Outside of the subject itself, the book serves as a beautiful example of a worked problem in logistics. It lays out its subject so clearly that you'll begin seeing places to apply it everywhere. I would recommend it to anyone who is working with complex systems.
Rationality: From AI to Zombies by Eliezer Yudkowsky (ISBN-13: 978-1939311238)
In this ambitious text Eliezer Yudkowsky attempts to put together accumulated research in psychology, epistemology, game theory, morality, statistics, and philosophy to create a self-repairing adaptable worldview that guides the user towards an accurate map of the world and effective action. As any such attempt would, it leaves some things to be desired. But it delivers on the basic premise far more effectively than any other attempt I've seen to date. Yudkowsky calls his philosophy 'rationality', or the art of winning. Rationality in this sense has very little to do with Spock's inflexible logic, instead referring to the game theoretic rationality discussed by economists focusing on making choices that maximize an agents utility. While certain ideas he focuses on have become much more popular over the last decade, the real value is not in any one idea or concept but the hints towards a deeper underlying attitude of effectiveness that the work leans towards but has trouble precisely defining. By the authors own admission it's something of a rough draft, but it's a rough draft that does a great deal of what it's intended to.
Superforecasting: The Art and Science of Prediction by Philip E. Tetlock & Dan Gardner (ISBN-13: 978-0804136693)
If empiricism is the king of science, then measurement is the queen. The thesis of Superforecasting is that expert predictions fail because we don't use standard measures of performance. Most predictions can't even be clearly evaluated as true or false, lacking objective criteria for resolution. Tetlock takes the reader inside his Good Judgment Project, an alternate reality where predictions have clearly satisfiable conditions and what matters is your track record not your credentials. In the process he introduces the concept of the Brier Score, a way to measure prediction accuracy over time by aggregating a series of probability estimates for events with boolean true or false outcomes. He also takes the reader through the techniques that his best performers use. These "Superforecasters" outperform even professional intelligence agents with access to classified information. The book is in a similar genre to Yudkowsky's, but less weird and more narrowly focused. If you're more interested in being correct than feeling correct, you'll like this one.
I cofounded a computer repair lab with five other students in my Freshman year of high school. What started out as an interesting summer project bloomed into a multi-year program that's still making headlines in my local community:
In 2016 I ran the LessWrong survey, which has had some interesting 3rd party analysis.