Emergent Ventures winners, fifth cohort

James Gallagher

16-year-old programmer from Northern England, grant for career development and his interest in income-sharing agreements.  Here is James on Twitter.

Namrata Narain

Incoming Harvard Ph.D student in economics, for work on “What happens to the ability of firms to write contracts when courts are dysfunctional? [in India]” and related ideas.  Twitter here.

Tejas Subramaniam

17-year-old from Chennai, Twitter here.  

Andy Matuschak

San Francisco, to support his project to reexamine and fundamentally improve the book as a method for learning and absorbing ideas, Twitter here.  Here is his essay on why books do not work.

Nicholas Donahue and Austin Kahn

Has a start-up, open source VR headset focused towards makers and web developers, based on the notion that the web is the proper platform for VR.

Clementine Jacoby and Recidiviz

To start a non-profit to collect and spread data on recidivism and penal reform for state-level policy, Fast Company article on Recidiviz here.

Mehdi Nayebpour

GMU, Schar School, “How can we explain a specific AI outcome? What if the law mandates it?”, with an eye toward an eventual start-up.

Colin Mortimer

Washington, D.C., for career development and to explore the marketing of neoliberal ideas through social media.

Shruti Rajagopalan, for Indian political economy and improving Indian economic policy, in residence at Mercatus.  Twitter here.

Jasmine Wang

20-year-old infovore, career development grant, Twitter here.


A non-profit working with survivors of sexual abuse in the Catholic Church, article here about their work.

If you have received an award lately, but are not listed, don’t worry — you’ll be in the sixth cohort.  Here are the earlier cohorts of winners.


'based on the notion that the web is the proper platform for VR'

First there was apparently no awareness that text and music have had a mixed media aspect on the web for a decade and a half, and now this.

This arose two and a half decades ago - https://en.wikipedia.org/wiki/VRML. The original name, Virtual Reality Markup Language, is self-explanatory. The headset aspect was not considered important, admittedly. Unless one is trying to return to Gibson's vision of cyberspace, which is approaching its fourth decade.

'with an eye toward an eventual start-up'

Always nice to see a non-profit fronting for a non-profit that is fronting for a man noted for VC success admit up front that it has an eye towards using the tax code to lower the cost of developing business ideas for others.

Why the obsession with "explainable" AI outcomes?

A seeing-eye dog can't explain how it makes the decision to cross a road or not, and yet we entrust human lives to it. A cadaver-sniffing dog can't explain how it sniffs out the fact that some specific missing person's body was transported in some specific car some time in the recent past, but if it does, that "testimony" is even admissible in court in some jurisdictions, if I'm not mistaken.

The only thing that should matter is if AI makes reliable decisions, not how those decisions are arrived at.

Most likely though, the West will stifle AI with a million restrictions while China and other countries move full speed ahead. The disparity "between those who believe in aspiring to something greater and those who do not" will not be a culture war, but a technological war, and the "wrong" side will win.

...If those aspiring to something greater win, may be it is not the "wrong" side. Yes, unfortunately the other side .

Very simple answer: liability.

Imagine a coffee machine. If one day fails at making delicious coffee safely, it's either 1) no maintenance or expired warranty, 2) bad operation : putting sand instead of coffee into it, 3) design problem : the vessel cannot hold hold water and it leaks. Scenarios 1 and 2 are my problem and therefore I pay to solve them. Scenario 3 implies the manufacturer must pay the cost of the machine and even liable to damages if my hand got burnt with hot water.

So, simple for coffee machine but not possible yet for some AI algorithms. If you cannot establish if the malfunction cause is bad operation (bad input) or bad design. Who's liable for the the millions in damages? It's millions since it makes no sense to put AI to work in simple problems for humans.

One of the issues in AI will be sex, gender and racial discrimination.

Let's first get definitions straight: AI that I am talking about is predictive analytics based on prior behavior that teaches the machine.

If we have a segment of the population that discriminates, and then we have an AI program that serves up based on the bias--eg, predicts criminal behavior, parole decisions, availability of loans, etc.--we run the risk that we will be served back a brew that escaberates current problems because we will never see what went into the AI decision.

What this means is that the learning algorithms might have to be debiased by excluding certain categories...such as religion, race, gender, etc. or adjusted to avoid bias. Explainability might help in those circumstances.

Bill and Axa are both right. There is also a more philosophical reason with some practical implications. A black box AI is a bit like a data table. It can give you outputs based upon inputs, but it is a very data intense way to do so rather than say, a nice neat formula. Equally important, if an AI can predict something, it still isn't really discovered scientifically if we don't understand how it does it any more than we do by just logging observations and looking for similar cases. One reason that this is important is that if an AI is trained on a set of data it isn't easy to know what the domain of applicability of its approach is in a way that is possibly, often easily, in a "transparent box" approach. For example, without knowing why it reaches the results it does, you won't know if data developed from a population in Minneapolis will be accurate for a population in Alabama or Lyons, France.

What about pharmaceutical research, though? Most of your objections would apply there too.

We are nowhere close to explaining the exact mechanisms of complex drug interaction with any specific human body. Every human body is a unique black box. Some people may experience very rare side effects.

If you had to meet the threshold of "explaining" every possible drug outcome, that would halt the use of all past and future drugs.

I don't think Tejas and Jasmine have fully fleshed out bylines. Good job to every one who made it into the round.

Andy Matuschak. I read his wordy essay on why books don't "work".

The foundations of the essay are a bit weak. Non-fiction books are Romance for older men, entertainment comes before learning. He looked a bit into textbooks but completely missed scientific articles. They're quite good for basic science. Accountants or engineers have conference papers which are also great summaries of knowledge and experience, not "science" but really useful.

Then, he relies a bit too much on his own experience to conclude books don't work. Humans are neurodiverse, what fails for one individual for learning is great for another. He honestly explains his experience of realizing later that he never really understood the idea while believing the opposite at reading time. He supports this insight with a nice anecdote instead of research. What if there's enough people out there really good at learning from texts?

Then comes a long digression on why lectures don't work while the essay is titled why books don't work.

The final issue is the focus on learning. You don't change the world by learning. You change the world by developing new things. Learning is just a few steps in the long path of discovery. When you specialize on something, you read all what has been written and then the only way forward is to do research, build things, do experiments, simulate, try, fail, try until it works.

My anecdote: during the PhD few people complained about not having time to learn/read, most people complained about not having enough time for experiments either in the lab or coding, or writting. Now in engineering consulting, no one complains about not having time to learn, we just want more time to run another set of simulations or account for more physical phenomena (less simplifications). I'd really like to see some research on why learning is the limiting factor on development.

Excellent analysis of the article. As you suggest the value someone can get from a book is ... YMMV.

As a Senior Citizen I find my biggest limitation on learning new things is that my memory is not as sharp as it was at 45, let alone 25.

So his recommendation of ‘spaced repetition’ is spot on for me.

Narain: Many of our judges, especially but not solely state court judges, come from backgrounds that removed from business and economics, such as criminal law, personal injury, and family law. Even basics are foreign to them. One can imagine the difficulty when the subject is a complex health care regulatory case or other complex case. Judges in those cases are in a fog. The tendency isn’t to let the jury decide when it’s not really a factual case, or to dismiss the case for failure to state a claim, in both cases for the same reason: the judge doesn’t understand.

There’s been an effort to move such complex cases to specialty courts, but that is resisted because of the insularity problem it creates. The better answer is judges from a broader background, but hoe does one convince a lawyer with a thriving practice in a complex area of the law to become a judge, in many state courts by having to run a political campaign. And many of the efforts to recruit better judges are driven by ideology, the Federalist Society for example.

My point is that court dysfunction is not limited to places such as India.

Agreed. Also, you should include judicial clerks as well, because they fill a gap for judges, and if they are selected by the Federalist society, we have further problem. Also, clerks become future judges.

My experience with judicial clerk hires is that on complex subjects they are a mile wide and an inch deep, and frankly, so were there judges. But, worse than being a mild wide, is to be very narrow and unaware of other ways to look at things.

"Many of our judges, especially but not solely state court judges, come from backgrounds that removed from business and economics, such as criminal law, personal injury, and family law. Even basics are foreign to them."

True story. I was in a divorce court hearing and was asking for additional discovery (pre-trial gathering of information) regarding shares that the other party owned in a closely held family corporation. I mentioned that he was an owner of that corporation. The judge asked me, clearly appalled, "do you mean to tell me that someone who owns shares in a corporation is an owner of the corporation?" The judge was not pleased when I answered "yes" even though a refrained from saying "duh!". The same judge was also incredulous at my assertion that the value of a closely held company wasn't necessarily very close to the book value of the corporation when its assets consisted mostly of shares in other companies purchased for a pittance decades earlier.

The judge in question had been a prosecutor specializing in child abuse cases for the judge's entire career before being appointed as a judge.

Cool list! I was checking out some of the links.

I notice Jasmine Wang aims to read a book a day. Top of her list was Parfit's "Reasons and Persons." Couldn't help but think she'd be much better off aiming for a chapter if not a page a day... Though perhaps she is in the mold of TC. :)

A book a day is definitely aspirational! I hope to post a retrospective on how it went, after. I expect to read something like 1 book / 3 days.

I was thinking more about this. Spending one day or three with a heavy philosophical test surely is better than spending no time with it at all!

Whatever the optimal reading strategy is, I wish you all the best in your endeavor. Have subscribed to your newsletter for updates

Short, easy books, with large type and wide margins will be easier to read at that rate.
Just a tip. Keeping it real.

I wish there were a faster war to immolate Peter Thiel's fortune, but I guess this will do.

Comments for this post are closed