Category: Web/Tech

Holden Karnofsky emails me on transformative AI

Here is Holden, our discussion started with this post of mine, for his words I will use quotation marks rather than dealing with double indentation:

“…debates about specifics between climate scientists get incredibly intricate (and are often very sensitive to parameters we just can’t reasonably estimate), and if you tried to get oriented to climate science by reading one it would be a nightmare, but this doesn’t mean the big-picture ways in which climatologists diverge from conventional wisdom should be discounted.

I think the broad-brush picture here is a better starting point than an exchange between Eliezer, Ajeya, me and Scott.

Even shorter version:

  • You can run the bio anchors analysis in a lot of different ways, but they all point to transformative AI this century;
  • As do the expert surveys, as does Metaculus;
  • Eliezer’s argument is that he thinks it will be sooner;
  • The most naive extrapolations of economic growth trends imply singularity (or at least “new growth mode”) this century;
  • Other angles of analysis (including the very-outside-view semi-informative priors) are basically about rebutting the idea that there’s a giant burden of proof here.
  • Specific arguments for “later than 2100,” including outside-view arguments, seem reasonably close to nonexistent; Robin Hanson has a (unconvincing IMO) case for synthetic AI taking longer, but Robin is also forecasting transformative AI of a sort (ems, which he says will lead to an explosion in economic growth and a relatively quick transition to something even stranger) this century.

So I ultimately don’t see how you get under P=1/3 or so for this century, and if you are way under P=1/3, I’d be interested if there were any more you could say about why (though recognize forecasts can’t always totally be explained).

P=1/3 would put “transformative AI this century” within 2x of “nuclear war this century,” and I think the average “nuclear war” is way less likely (like at least 10x) to have super-long-run impacts than the average “transformative AI is developed.”

That’s my basic thinking! It’s based on numerous angles and is not very sensitive to specific takes on the rate at which FLOPs get cheaper, although at some point I hope we can nail that parameter down better via prediction markets or something of the sort. Prediction markets on transformative AI itself are going to be harder, but I’m hopeful about that too. I think a very fast transition is plausible, so it could be very bad news if folks like you continue thinking it’s a remote possibility until it’s obviously upon us. (In my analogy, today might be like early January was for COVID. We don’t know enough to be sure, but we know enough to be highly alert, and we won’t necessarily be sure very long before it’s too late.)”

End of Holden, now back to TC.  And here is Holden’s “most important century” page.  That is our century, people!  This is all a bit of a follow-up on an in-person dialogue we had, but I will give him the last word (for now).

The U.S. is really good at destroying things

If a country is really good at X, and along comes social change Y, you ought to figure there is a pretty good chance Y will feed into X.

So for instance the United States is really good at retail.  So along comes Wokeism, and, lo and behold, Wokeism slots wonderfully into retail, whether you like that fact or not.  The Woke is marketed all the time, and so you can find “green” versions of so many products, even if organic food costs more energy, etc.  Or if you invent a new app, with anti-corporate purposes in mind, don’t be surprised if you wake up one day and the app has been co-opted by retail corporations.  And so on.

What we are now observing is that the recent innovation of “The Woke’ has been co-opted for the purposes of destroying things, in this case the economy and possibly the society and polity of Russia.  The lining up of European allies, the mobilization of sentiment on Twitter, the inducement of Visa and Mastercard to pull out, and you could go on and on and on.  The Woke campaign against Russia has turned out to be extremely powerful, well beyond what I had been expecting to happen.  Energy purchases might be next to go.

To be clear, sometimes we are good at producing or threatening destruction in very beneficial ways (WWII, 1973 airlift to Israel, ending the post-Yugoslavia wars, stopping Saddam from taking Saudi oil, etc.), and sometimes we are good at destruction in very harmful ways (you can supply your own list, but it is extensive).  In this post I am not going to try to assess the net expected value from the ongoing destruction of Russia, only to say that the final outcome is uncertain.

Nuclear power, airplanes, computers, GPS and much more all have been co-opted into destroying things, again noting these effects may be net positives on the whole.  They have been co-opted into retail purposes as well, nuclear power excepted.

Make no mistake about it, many of the most important “contributions” of Wokeism are to feed into, and enhance, those capabilities that America already is good at.

Which include retail and yes, destruction as well.

The Coase theorem, combined with the force of increasing returns to what a country is really good at, will see to this.

The same feeder tendencies may well be true for other innovations you might have in mind, be they practical or intellectual in nature.  Whether you like it or not, they will contribute to both American retail and to the American capacity to destroy.

Some people observe these trends and think the Woke is transcendent and all-powerful.  But under another reading, it is actually the Woke that is somewhat being pwned.

NFT markets in everything

Ukraine plans to become the first developed country to issue its own collection of non-fungible tokens, as it looks to capitalise on a flood of crypto donations to back its war against Russia.

Mykhailo Fedorov, Ukraine’s vice-prime minister, announced the plan in a tweet on Thursday and said Kyiv would reveal details of its NFTs soon.

The move is the latest sign of the Ukrainian government embracing digital assets as a way to fund its armed forces in their battle, and comes after it raised more than $270mn in “war bonds”.

One for each Russian tank destroyed?  Here is the full FT story, via Natasha.

Jesse Michels interviews me at Hereticon

Jesse’s description was “Wide ranging discussion with the brilliant @tylercowen. Topics include: Satoshi’s identity, Straussian Jesus, the Beatles and UFOs. Taped in early January but he presciently expresses concerns around Russia/Ukraine”

Great fun was had by all, and they added in nice visuals.

What should I ask Daniel Gross?

I will be doing a Conversation with him, noting that he is my co-author on Talent: How to Identify Energizers, Creators, and Winners Around the World.

Daniel is an entrepreneur and venture capitalist and here is his Wikipedia page.  Here is Daniel on Twitter.  Here is Daniel’s ideas page.  Here is Daniel on his work, including Pioneer.

Since we are co-authors, this won’t just be the standard interview format, how do you think we should do it?  And what should we ask each other?

An underdiscussed biomedical problem?

Including for longevity research, and perhaps most of all for longevity devices:

Neural implants—devices that interact with the human nervous system, either on its periphery or in the brain—are part of a rapidly growing category of medicine that’s sometimes called electroceuticals. Some technologies are well established, like deep-brain stimulators that reduce tremors in people with Parkinson’s disease. But recent advances in neuroscience and digital technology have sparked a gold rush in brain tech, with the outsized investments epitomized by Elon Musk’s buzzy brain-implant company, Neuralink. Some companies talk of reversing depression, treating Alzheimer’s disease, restoring mobility, or even dangle the promise of superhuman cognition.

Not all these companies will succeed, and Los Angeles–based Second Sight provides a cautionary tale for bold entrepreneurs interested in brain tech. What happens when cutting-edge implants fail, or simply fade away like yesterday’s flip phones and Betamax? Even worse, what if the companies behind them go bust?

Here is the full story, via Anecdotal.  Was Colonel Steve Austin built out of parts from start-ups?  How many of those companies still would be around?

TechCrunch interviews me about crypto

TC is the interviewer, TechCrunch (John Biggs), not I, here is one bit:

TC: Folks liken this tech to cargo cults. You build the trappings of an economic system in hopes that one magically appears. Is this accurate?

Cowen: I think the crypto people are super, super smart on average. They’re smarter than economists on average. And they have skin in the game, right?

TC: Does the profit motive color the experience?

Cowen: Well, people in crypto want to build systems that work. It’s fair for all of us to have uncertainty about how that will go. But the price of crypto assets have been pretty high for a while now and they’ve taken big hits and come back. So I don’t think you can now say it’s just a bubble. So what exactly it would be is still up for grabs, but I think the bubble view is increasingly hard to maintain.

TC: Are we assuming this stuff is here to stay? That bitcoin won’t disappear in a decade?

Cowen: That is strongly my belief. Now, there’s a lot of other cryptoassets and I think most of those will disappear. There were many social media companies fifteen years ago also and many are not around but obviously social media is very much a thing.

TC: What’s your take on decentralized, from an economic perspective?

Cowen: I think we will end up with both centralized crypto and decentralized crypto, and they will serve quite different functions. So, obviously, there were advantages to centralized systems. You can change the more quickly, more readily. There’s someone to manage them, someone to oversee them. But you also pile up costs. So I think both will prove robust. But again, I would readily admit that still up for grabs.

There is much more at the link.

My Conversation with the excellent Sebastian Mallaby

Venture capital most of all, hedge funds as well with the Fed tossed in.  Here is the audio, video, and transcript.  Here is one excerpt from the Conversation:

COWEN: What do you think of the view that in recent years, there’s been a huge consumer retail tech boom? Basically FAANG stocks, right? And when that is over — it might be over now — the excess returns to VC will go away. If you look at venture capital for biotech, which has been hammered lately, as we’re speaking here, late January 2022 — and maybe venture capital is a limited model for one period of time, and otherwise, it just does okay. True or false?

MALLABY: False. I say that because, in a cyclical sense, you might be right, but I think there’s a deep structural shift, which is really important. That is that intangible capital has become more and more important in our economy. The nature of intangible capital is that it’s hard to measure it in financial reports.

To understand whether a particular software investment, for example, is worth a huge amount or, really, nothing, you need to understand what that software development within the company is doing. You need to be hands-on. You need to have the technical skills to evaluate that software project. The more that intangible capital rises as a share of new GDP creation, the more this venture-style hands-on expert investing is going to be valuable.

COWEN: Your explanation — if I understand it — to me seems to suggest that venture capital for biotech won’t work very well. You’re portraying it as something that’s very, very hard to do, a very limited skill, so you’re going to be wrong a lot of times. That means the times you’re right, the product has to be scalable very rapidly.

But in biotech, there are regulators. You often need a sales force. It’s not scalable in the way that, say, LinkedIn or Netflix are scalable. Doesn’t that mean VC will just stay limited to a very small area of those things that are super rapidly scalable? Or if you think it’s pretty easy to pick winners, then you have to think the rents get exhausted.

MALLABY: [laughs] Yeah, this is a version of, actually, a wider debate which goes beyond biotech, which is the claim that venture capital is really only good for software projects, that software can be scaled very, very fast; there are network effects once you get product-market fit, and you don’t need much capital.

And this:

COWEN: I have some questions about other topics. You have some highly regarded books about hedge funds and about the Fed. In the late ’90s, the bailout of Long-Term Capital Management — was that a kind of original sin that just set us on a path of bailing more things out at higher and higher price tags? Should we have just let LTCM fall?

MALLABY: No, I think the original sin was Continental Illinois, much earlier in 1986, I believe, when the Fed bailed out this bank which it thought was too big to fail. I’m not sure it really was too big to fail, but it was a moment when the Latin American debt crisis was still casting a shadow, when the banking system was perceived to be fragile, and the Fed just wasn’t willing to let it go. That was the original sin because taxpayer money was used to bail it out.

There is much more at the link, and I am very happy to recommend Sebastian’s new and very good book The Power Law: Venture Capital and the Making of the New Future.

Netflix economics and the future of Netflix

Ted Gioia writes:

Netflix’s market share has been declining steadily, and has now fallen below 50%. One estimate claims that the company’s share of consumers fell more than 30% in a single year. Netflix’s recent quarterly report was a disaster, spurring a share sell-off. You could easily conclude that “Netflix’s long awaited funeral is finally here”—as Bloomberg hinted in its blunt assessment of the results.

Of course the company is still worth quite a bit, so my own view is no more or no less optimistic than what the market indicates.  Still, it is worth asking what the equilibrium here looks like.  There is also AppleTV, Disney, Showtime, HBOMax, Hulu, AmazonPrime, and more.  I don’t think it quite works to argue that we all end up subscribing to all of them, so where are matters headed?  I see a few options:

1. Netflix and its competitors keep on producing new shows until all the rents are exhausted and those companies simply earn the going rate of return on capital, with possible ongoing rents on longstanding properties of real value (e.g., older Disney content).  These scenarios could involve either additional entry, or more (and better?) shows from the incumbent producers.

2. Due to economies of scale, one or two of those companies will produce the best shows and buy up the best content.  We end up with a monopoly or duopoly in the TV streaming market, noting there still would be vigorous competition from other media sources.

3. The companies are allowed to collude in some manner.  One option is they form a consortium where you get “all access” for a common fee, divvied out in proper proportion.  Would the antitrust authorities allow this?  Or might the mere potential for antitrust intervention makes this a collusive solution but one without a strict monopolizing, profit-maximizing price?

4. The companies are allowed to collude in a more partial and less obvious manner.  Rather than a complete consortium, some of the smaller companies will evolve into “feeder” services for one or two of the larger companies.  Those smaller companies will rely increasingly more on the feeder contracts and increasingly less on subscription revenue.  This perhaps resembles the duopoly solution analytically, though a head count would show more than two firms in the market.

It seems to me that only the first scenario is very bad for Netflix.  That said, it seems that along all of these paths short-run rent exhaustion is going on, and that short-run rent exhaustion is costly for Netflix.  They keep on having to pump out “stuff” to keep viewer attention.  It doesn’t matter that new shows are cheap, because as long as the market profits are there the “bar” for retaining customers will continue to grow.  Very few of their shows are geared to produce long-term customer loyalty toward that show – in contrast, people are still talking about Columbo!

Putting the law aside, which economic factors determine which solution will hold?  My intuition is that there are marketing economies of scale, but production diseconomies of scale, as the media companies grow too large and sclerotic.  So maybe that militates in favor of scenario #4?  That to me also suggests an “at least OK” future for Netflix.  The company would continue its investments and marketing and an easy to use website, while increasingly going elsewhere for superior content.


What is the central political question of our day?

No, it is not about The Woke.  From my latest Bloomberg column, here is the core argument:

How to respond to climate change is often postulated as the central question of our time, and while that’s undeniably important, I have another nomination: How will we stop our new and often splendid technologies from being weaponized against us?

I use the term weaponization quite literally — drone attacks, cyberattacks, hostile uses of artificial intelligence, and attacks from space, bioweapons and more. It’s good that the world is emerging from a period of technological stagnation, but therein lies a danger: It is a general principle of world history that new technologies, even the most beneficial ones, are eventually used either as weapons themselves or as instruments of warfare. That was true of the horse, the railroad, the airplane and, of course, nuclear power. It likely will be true for these new developments, too…

Most current ideologies are unprepared for this coming new world. These problems do not have obvious solutions, nor do they offer any obvious way to confer political advantage. The U.S. hasn’t even made much progress on preparing for the next pandemic, and that is with more than 2,500 Americans dying a day from Covid-19.

Here is another point:

There are ideologies that address parts of the weaponization problem. Effective Altruist circles, especially those that focus on the dangers of artificial general intelligence (AGI), are afraid that super-smart AI will develop a mind of its own and impose its will on us, or otherwise engage in evil activities.

That may be a valid concern, but my fears are more general. If AGI is so powerful, then it stands to reason that intermediate products could, in conjunction with human efforts, cause a lot of military conflict. The problem isn’t necessarily Skynet going live. It’s that 40% of Skynet will be plenty dangerous.

The Luddites also have an ideology, namely that the development of new technologies should be stopped altogether. One could debate the benefit-cost ratio of that decision, but suffice to say that China, Russia, and many other rival nations have no such plans, and the U.S. has no real choice other than to try to stay ahead of them.

China is discussed as well, recommended.

The incidence of India’s crypto tax

The crypto tax is the first item listed in a section of the budget memo headed “Revenue Mobilization”. The document [PDF] explains that India wants to tax income from crypto-assets at a 30 per cent flat rate.

By comparison, India currently taxes short-term capital gains made by selling shares at 15 per cent. The budget memo also calls for a one per cent tax on sales of cryptographic assets, payable by parties to the transaction, to widen India’s tax base.

Here is the first article link.  As I understand it, the 30% is on net income from crypto, and there is no tax deductions for losses (see this explainer).  (Does the tax define gains “year by year,” or “for each bitcoin sold”?)

I am wondering what is the incidence of this tax.  Presumably India is a price-taker in the crypto market as a whole, so this initiative should not much affect the global price of crypto, unless you take the policy as a signal about other, future crypto taxes to come around the world.

Under one (unlikely) scenario, all Indians were marginal crypto buyers, and so with a 30% tax they just stop holding crypto.  The Coase theorem suggests that others are always willing to bid more, because in many other countries the crypto taxes are lower.

More realistically, many Indians are infra-marginal buyers, with sufficiently high expectations of price appreciation that some of them will stay in the market.  The “saner” marginal buyers will drop out, and sell their crypto to non-Indians, and the most optimistic Indian buyers will stay in.  Looking forward, crypto in India will be shaped by the giddiest and most bullish asset holders, compared to the status quo.  More crypto will be held by fewer, more enthusiastic hands.

The Indian government is also signaling that it will not ban crypto outright.  That ought to increase the demand of the “giddy” buyers all the more.  If you are going to stay in with the higher tax burden, at least you know that bitcoin and other markets will continue in India.

How does the tax affect the value of the rupee?  In the short run, some Indian taxpayers may sell their crypto for rupees, raising the value of the rupee, but probably very slightly.  Longer term, the rupee may be worth less because it is a less effective vehicle for investing in crypto, again with the effect here likely being small.

Otherwise, the demand for non-crypto risky assets in India will increase.  If those assets can be used for loss offsets, they will be relatively more valuable because crypto cannot be so used.

Insofar as India has a local, “India-only” crypto market, new issues there will have to be lower in price to attract buying interest.  That will serve as a tax on those Indians who supply inputs into crypto production.

Indians who have made a great deal from crypto may attempt to give up their Indian citizenship and Indian taxpaying liabilities (how easy is that?).

What else?

A simple theory of culture

The transistor radio/car radio was the internet of its time.  Content was free, and there were multiple radio stations, though not nearly as many as we have internet sites.

People tuned into the radio, in part, for ideas, not just tunes.  But the ideas that spread best were attached to songs.  Drug use spread, in part, because famous musicians sang about using drugs.  Anti-Vietnam War themes spread through songs, as did many other social movements.  Overall, ideas that could be bundled with songs had a big advantage.  And since new songs were largely the province of young people, this in turn favored ideas for young people.

Popular music was highly emotionally charged because so much of it was connected to ideas you really cared about.

Of course, by attaching an idea to a song you often ensured the idea wasn’t going to be really subtle, at least not along the standard intellectual dimensions.  But it might be correct nonetheless.

Today you can debate ideas directly on social media, without the intermediation of music.  Ideas become less simple and more baroque, while music loses its cultural centrality and becomes more boring.

We also don’t need to tie novels so much to ideas, although in countries such as Spain idea-carrying novels remain a pretty common practice (NYT).  A lot of painting and sculpture also seem increasingly disconnected from significant social ideas.

In this new world, celebrities decline in relative influence, because they too are no longer carriers of ideas in the way they used to be.  Think “John Wayne!”  Arguably “celebrity culture” peaked in the 1980s with Madonna and the like.

When I hear various complaints about the contemporary scene, sometimes I ask myself: “Is this really a complaint about the disintermediation of ideas”?

In this view, the overall modern “portfolio” may be better, but the best individual art works, and in turn the greatest artists, will come from the earlier era.

We need a better tax system for crypto

From N., an MR reader:

I own crypto in 3 different centralized exchanges, two hardware wallets, one software wallet (Metamask), have four cryptos staked in multiple different pools and I also have some cryptos I gained by mining them using my GPUs. I have made 600+ transactions between the exchanges, wallets, and staking pools. I hold 75% of my portfolio and trade the rest. So most of these transactions were for trading one coin for another from which I have profited handsomely in the 2021 bull cycle run.

But I am doing my Crypto taxes right now its an unbelievably complicated nightmare. Prior to 2020 I only held a Coinbase account and I downloaded the tax forms or the transaction list as a .csv file from it and submitted them to my tax advisor. But in 2021 I have gone deeper into crypto and I have purchased hardware wallets, held crypto in soft wallets, DeFi platforms like Aave, staked crypto, mined crypto, and traded crypto between exchanges for lower transaction fees, for coins that are available only in certain centralized and decentralized exchanges, etc etc. Many of these types of crypto transactions are taxed differently and are from different institutions.

So it’s impossible for me to do my crypto taxes easily with just a single tax form from Coinbase. I have to link all my exchanges (and expose all my crypto holdings and trades) to a crypto tax website, I have decided to use which charges $99 to do my taxes. I do not have any other realistic choice.

After I linked all 3 centralized exchanges where I hold crypto, the capital gains estimate gave seemed too large. I realized it was because it was counting the crypto I sent from centralized exchanges like Coinbase to my hardware wallets as a “Sell” so it was counting them as capital gains. I have too many transactions of this nature to manually go through them one by one and mark them as “Transfer” i.e. transfer between my own wallets. So if I want the tax software to do it automatically, I have to expose the public keys of my hardware wallets so Koinly can automatically mark them as transfers. (I haven’t done this step yet because I don’t want to expose my hardware wallet public address to anyone or anywhere and I am researching alternate ways to do this.)

But if there isn’t any other way either a) I have to spend hours going through each transaction manually and marking them as “Transfers” or b) expose the public keys of my hardware wallets to

Also, there is more manual work to be done for categorizing certain transactions as moved to staking pools, marking transactions from my mining pool to exchanges as income, etc.

I know that fiat currency debit and credit card purchases are absolutely not analogous to crypto but that’s the comparison many crypto maximalists make (“take down the traditional financial and banking system!”).

Imagine if TurboTax needs your complete transaction history from your banking institutions and it goes through all credit and debit card transactions to accurately do your taxes. Would anyone accept that?

Dominant Assurance Contracts and Quadratic Funding

Here is my keynote talk for the ACM Advances in Financial Technologies conference. I discuss Two Novel Mechanisms for Funding and Discovering Public Goods, namely dominant assurance contracts (with experimental support here) and the Buterin, Hitzig, Weyl work on quadratic funding.

Lots of other excellent talks on blockchains, AMMs, decentralized finance and so forth are at the link.