Digital Copyright and the Duties of Platforms to Prevent Third-Party Infringements

This post is not about actions that platforms take themselves (for example, Google in the Google Books case, whatever you think of how it came out, was just directly scanning books itself), but about their responsibilities with respect to user activities (for example, videos uploaded by users to Facebook or Youtube). Some copyright people might say that this is a distinction without a difference, as a file host might be a direct infringer—but we’re not talking about direct vs. secondary infringements, volitional action, and any of those fun doctrines, but the what I hope is fairly commonsense distinction between potential infringements that result from the platform acting on its own vs. those that directly result from user activity, while acknowledging that there may be grey areas where a platform takes affirmative steps to encourage third-party infringing use. (In those cases, under existing law, the platform may be liable at least as a secondary infringer, even if its behavior does not constitute direct infringement.)

This already hints at a pretty fundamental issue: to what extent should a platform that enables a user to infringe, also be held to infringe itself? Which conduct should actually constitute “infringement”? Does an internet platform only have an obligation to avoid infringing in and of itself, or does it have an obligation to prevent third-party infringements? If it fails to police third-party infringements should it become an infringer itself or is there some other penalty? (The current DMCA framework, which creates a "safe harbor" for conduct where there is sometimes not an underlying infringing act, does not contribute to clarity in this regard.)

More questions: Should an ISP be held to infringe if infringing material is transmitted over its wires? Should a file host be held to infringe if a user uploads infringing content to a private account? To a public one? Does it make a difference if the content is encrypted such that the file host or ISP can’t see it? Should services have an affirmative obligation to scan private content, and should this affect liability?

These questions are all about what the actual scope of copyright should be. It is very hard to disentangle views on enforcement mechanisms from views on the underlying scope of the law. I am personally more sympathetic, for example, to attempts to enforce rights that I also believe should actually exist. But the law does not distinguish between enforcement efforts relating to a 70-year-old out-of-print book that should be in the public domain by now, and a newly released movie. So let’s fix that.

Next, now that we’ve all agreed on what the scope of the law should be, what’s the best way to enforce the underlying rights? Who should bear the cost? The easy policy answer is that the “least cost avoider” should. But it’s not clear who that is, and anyway it might vary from case to case.

In general, I think that the cost of enforcement of many private rights ought to be borne by the rightsholder, because that’s the cleanest way to ensure that the cost of enforcement doesn’t exceed the benefit of the right. From a utilitarian perspective, creating a strict liability regime, with crippling statutory damages for platforms, could create a situation where people are spending $500 to enforce a $5 right. Not good! (This is basically one of the differences between private civil law and criminal law. If something is a crime, you might want the state to prosecute it, costs be damned. But if someone commits a tort against you, it’s on you to try to recover damages if you think it’s worthwhile. Most torts are just not worth suing over.)

But, there are some pretty clear equity problems with taking that approach too far. I might want platforms to have to spend a disproportionate amount of money to protect the rights of individuals, if not those of Disney. At the same time, some of the enforcement mechanisms long advocated by the content industry are easily affordable by say Facebook and Google—who already have voluntarily implemented some of them anyway—and not by smaller platforms. For example, “staydown” requirements require a fairly advanced filtering and scanning infrastructure, and people still find ways around Youtube’s Content ID pretty regularly.

Still, it seems that platforms are best-situated to see what happens using their facilities—what if it would cost $3 for a platform to protect a $5 right, and it would cost the rightsholder $4? In that case, don’t we want the platform to do it?

But there’s still another question: Which platform? If we’re talking about Youtube, it seems like a pretty easy question. But what about a third-party CDN, an internet transit provider, the user’s ISP, the operating system vendor, the browser maker, the end-user device manufacturer? There’s no reason to think that any of them might not be the least-cost-avoider. Remember that there was a pretty strong feeling among content types that Apple was either encouraging or at least profiting from piracy when it released the iPod, and that some countries charge a copyright levy on blank media. So there’s not some kind of “layers” principle that everyone already agrees with.

It seems that if the goal is to just empirically determine who can enforce a right in the most cost-effective way, everything should be on the table. People often say that you wouldn’t want to hold the manufacturer of the getaway car liable for a bank robbery, but that’s not really true; if holding the manufacturer liable was the best way to actually prevent robberies (note: it isn’t), we probably would. All this stuff about “proximate cause” you learn about in torts class would go right out the window. This is in part why payment processors are responsible for enforcing all kinds of junk.

These are all ridiculously hard questions. There are answers to many of them within the current framework of the law but we’re not talking about that right now.

If I were inclined to rework the basic legal framework in this area I’d probably:

  • Distinguish between the obligations of large/dominant/etc platforms, and other platforms. I think it’s fine to put extra burdens and duties of care on large platforms, though the mechanism by which to accomplish this should probably not be, “If you mess up you stand in the shoes of the uploader of the content.” There are other ways to create obligations, besides tinkering with substantive copyright law.

  • Not treat all “intermediaries” or platforms or whatever the same. Pure transmission companies should not be liable for what passes over their wires, for instance. Current law does not make these distinctions, at least not very clearly.

  • Some other stuff.

I’ll end with a note that is somewhat pro-content. I will admit to being surprised that The Pirate Bay is still up and running and accessible, after so many years. I mean, come on. I am not a fan of DNS/ISP-level site-blocking as was proposed in SOPA but it sure seems like we need a better way to actually take action against such a blatantly unlawful site. Unfortunately on the internet we quickly run into some really thorny jurisdiction issues that sadly show no signs of being resolved.

Device-to-device synchronization would be a cooler way to improve Siri, instead of just putting everything in the cloud

I agree with all of the criticisms of Siri in this post by Bryan Irace, but not in the proposed solution--that Siri should live more in the cloud, along the lines of Alexa and Google Assistant, instead of on-device. If we're going to be surrounded by tiny supercomputers all the time, we should use them as such and in a way that is inherently more protective of privacy, rather than as dumb terminals for cloud services.

First, of course, much of Siri already is in the cloud. However, and this is what's different about Siri vs. other voice assistants, generally personal data about you is stored just locally, on device. When personal information has to be transmitted to the cloud (for example, your current location, for weather or traffic results), it's done in a way that's not tied to a specific user. I'm sure I'm mangling some of this; Apple explains the setup here.

Calendars are a pretty clear way to see the difference. When you ask Google Assistant or Alexa what your calendar events are, it checks your cloud-connected calendar account, and then tells you. The response is formulated in the cloud and just read back to you on the phone or speaker or Alexa toaster. Siri does have to send your voice to the cloud for it to figure out what you want--personal devices aren't quite beefy enough yet--but then the cloud just sends back a command to the phone "Tell the user about upcoming calendar events" or whatever. What those events actually are depends on what's on your device. If you have different calendars set up on an iPhone vs. an iPad, you get different results. And if you have a device without a calendar, like an Apple TV, Siri just can't answer from that device. The drawbacks here are fragmentation, and devices without the ability to answer some kinds of questions; the benefits are customization and privacy. You can use Siri to ask about your calendar events even for a calendar on your phone that is not synced to the cloud at all, for instance, and it works the same whether you use iCloud calendars, Outlook, Google Calendars, or any CalDAV calendar service at all. A cloud-based system is more limited: Google Assistant only works with Google Calendars, and Alexa only works with iCloud and Google Calendars.

The drawback of Apple's approach that most troubles some critics including Bryan is extensibility. Siri can only be extended through local apps, as opposed to through cloud-based skills. This leads to another form of fragmentation as different devices might have different apps installed, and some devices (HomePod, Apple Watch, Apple TV) do not even support extending Siri through SiriKit apps at all.

That said, I still think it would be a mistake for Apple to move to a pure cloud-based approach. It would be simpler in the short-term, but the advantages of fast, local, native software would allow Apple to distinguish itself long-term. It allows developers to use existing methods of proving paid and subscription apps, and ensures that Siri is an alternate way of using a device, rather than a "platform" in and of itself--which some people want, but which I think adds too much complexity.

An example of the flaws with the "independent platform" approach can be seen with Alexa on Fire tablets, with is only loosely integrated with the device itself. Setting an alarm in Alexa does not have anything to do with any app on the device, for instance. You set an alarm with Alexa, and the only way to see or interact with it is through Alexa, as opposed to just opening the clock app, looking at a lockscreen notification, etc. By contrast asking Siri to create an alarm on your phone, has Siri create an alarm on your phone that is the same alarm you'd have created by poking and tapping around. This seems like a more solid approach. Long-term, also, local apps will allow Apple to have more and more Siri functionality (voice recognition, parsing etc) handled entirely on-device.

Rather that shifting more to the cloud, Apple should enable a user's various devices to synchronize with each other, either peer-to-peer or end-to-end encrypted through a dumb, storage-only cloud, whose contents Apple could not view even if it wanted to. This would allow a user to use Siri the same way on any device--for example to view and create calendar events on an Apple TV, even though the Apple TV would have no calendar app. (A user would still be able to turn off this synchronization for particular devices.) This approach would have several other benefits as well--for instance, a user would be able to keep a local, non-cloud calendar, but still have it synchronized to other devices. There was a bit of a stink when the inexorable move to the cloud took away people's ability to just sync a PC or Mac calendar to a phone without a cloud intermediary, you may recall.

As for apps, they'd sync, too, with irrelevant assets not installed on devices that don't need them. For instance a SiriKit app would be installed directly on a HomePod, but without any of the pretty pictures.

Apple, of course, doesn't have the best track record here. The aspects of many of its services that work the best, like CloudKit, are just regular cloud services like any other, and when some processing is done entirely on-device (e.g. face recognition in Photos) device-to-device synchronization is somewhat limited.

All that said, while I'm obviously not the right person to work through all the details here, it would be nice if Apple found ways to improve the functionality of Siri without turning into just another cloud-based service. The end goal should be more and more functionality being local and free of the cloud, not less.

Listen to music, the technology way

As part of my commitment to service journalism, here is how I organize my personal music collection and listen to it on all of the high-tech devices that so disastrously permeate my life, including smartphones (with offline playback) and Echo speakers.


I organize my music using the Mac program Swinsian. Swinsian is somewhat like iTunes used to be before it got crudded up with a bunch of extraneous features that have nothing to do with organizing local music files. It has many useful tools for editing and completing track metadata and, most importantly, it is very fast and reliable and does not choke on large music collections. A Windows equivalent would be MediaMonkey.

When I say I use Swinsian to organize my music, that’s really all I use it for. It sorts my music into folders based on criteria I specify (I use the standard iTunes-style Artist/Album folders) and resorts it if I change the metadata.

Swinsian‘s biggest drawback is that it does not support multiple libraries; you can work around this with alternate user accounts or a virtual machine.


This is sort of extraneous but, of course, I keep all my files, music included, backed up to the cloud with Backblaze. Backblaze is $5 per month and it is the best unlimited backup service for most people. It will back up everything on one computer, including external drives, but not network drives.

Music server

I use Plex media server to make my music available to myself wherever I go. Plex is its own separate universe; it doesn’t sync up with Swinsian other than periodically scanning the folder structure it creates. I’m not going to tell you how to get Plex set up (it's pretty easy), but to access it outside your local network you’ll need to configure port forwarding on your router to make sure Plex is available externally, if it doesn't happen automatically. You'll need a $5/month Plex Pass account to sync music onto your smartphone for offline listening and to get some other fancy music-related features.

Plex isn’t perfect: Its metadata scanning has a number of bugs (for instance, with older versions of the ID3 format it will insert a bunch of extraneous slashes into your artist name or whatever). Its apps can be unintuitive and weirdly designed, creating playlists is a pain, and there’s no way to actually import normal .m3u playlists. (It can display the contents of an iTunes library, but this is not very useful.)

The other disadvantage of this setup, obviously, is you’ve got to keep your home computer running all the time. However I still find it better than any cloud music service, which often do “scan and match” which often goes wrong, and which have a cap as to the number of tracks you can upload, and which do not stay in sync with your local music collection (for example, if you have edited metadata on some tracks locally, Google Play Music would not reflect this unless you deleted tracks from it and reuploaded them or somehow edited them on the service). (Of the current cloud music services, I think Apple does it best because cloud tracks can more easily stay in sync with changes made locally, but people seem to constantly get confused as to how it works, iTunes bogs down with larger libraries, iCloud Music library has a 100,000 track limit, and I fundamentally don’t like the concept of mixing my own personal music tracks with the tracks available in a streaming service as it just leads to needless complexity.)

Another drawback of this setup compared to the easy days of iTunes + iPods is that with the multifarious ways of listening to music the notion of keeping track of play counts and last-played is no longer really feasible. I used to have smart playlists that were composed just of music I hadn’t listened to in, say, the past three years, but I lost all that historical play data years ago and now no longer bother.

Plex Apps

Now that your Plex server is set up correctly you can access your music via various Plex apps and the web. There are apps for Android, iOS, Roku, Apple TV, PlayStation, etc etc, and you can access your Plex collection via the web, as well.

Some of these apps can be used as remote controls for other Plex apps; for example, you can use iOS Plex to control Plex on the Apple TV, but I’ve never found this to be very useful.

Plex also has a peculiar Mac and Windows music-only app called Plexamp. I like the idea but it’s a strange moon man app with its own non-native user interface concepts (no window chrome, annoyingly creates a menu bar icon, etc), so I don’t use it. However I do like its “Plex radio” feature that just plays selections from your music library and I hope it comes to other Plex apps. There is a desktop Plex app, too, that is just a wrapper around the web app, but it's pretty nice for listening to music since you can resize it without screwing up your web browser window size at all, and listen to some premium music features that don't show up in the mobile apps for some reason, such as filtering music by "mood."


The Plex Alexa skill was initially kind of useful for some things. Once a Plex app was running, for example, on an Apple TV you’d be able to say “Alexa, tell Plex to play Herb Alpert's Tijuana Brass” and it would start playing on the Apple TV. It still works this way for video, but audio will now play directly through the Echo.

That’s it

There are other solutions out there: more music-focused home server setups like Subsonic, Coppertino’s Vox apps and Loop cloud locker service, but none of them have the device support that I’m looking for. So Swinsian + Plex it is.

A basic problem with broadband measurement

If you're trying to figure out how much broadband is deployed, and how good it is, there's a threshold issue that has no one right answer.  Let's say that you want to measure the number of households with (or with access to, as will be discussed) broadband at 25 Mbps or higher.

The most "honest" way to do it would be to see how many households actually have 25 Mbps connections right now.  

But what if 25 Mbps is available, but lots of people choose a slower and cheaper connection instead? Then you might want to count "availability," not actual subscriptions.  But this approach has flaws as well:  What if the 25 Mbps price tier is absurdly overpriced?  It's "available," in the same sense that a private jet is "available" to me.  Also, if suddenly people do subscribe to 25 Mbps tiers, or the ISP upgrades everyone to 25 Mbps for free, that is certainly significant but should it count as progress in *infrastructure* deployment?

Also, there seems to be a big difference between, for example, new deployment of cable that is capable of 25 Mbps (actually trenching new wires), and upgrading an existing cable deployment to become capable of faster speeds through much-cheaper upgrades to electronics.  Again, both are significant, just differently so.  The difference between the two situations may inform how you would, for instance, direct subsidies.

Finally, there's a difference between a home that is "passed" by broadband and a home that is actually connected to a network.  A "passed" home might be easy to connect, but it isn't necessarily connected; by contrast I am currently a FiOS subscriber but I also have currently-unused Comcast coax attached to my house.

There are pluses and minuses of each kind of measurement, and some are easier than others, but in lots of reporting about broadband these distinctions aren't made.

weight loss tips

Since I lost 50 pounds last year, some people have asked me how I did it.  This is how.

I only eat between noon and 8 pm.  (At first I did this accidentally and not very precisely, just by skipping breakfast except for black coffee.  But now it's deliberate, as apparently it's a form of intermittent fasting which is good for you somehow.)

I have a calorie budget of 1,586 calories per day, and I try not to go over it.  I track what I eat with the app "Lose It!".  Part of calorie tracking is coming to grips with guesswork and imprecision.  You just hope the errors balance out any are not always in one direction or other (you don't want to sneakily overeat, nor pointlessly starve yourself).  The main benefit of using a calorie-tracking app is that it makes it harder for you to deceive yourself with small little snacks and so on.

Learn about what foods are filling / satisfying but don't have too many calories.  For me, it's plain greek yogurt.  If I want some sweetness, I like to add raisins or honey.  Contrariwise learn which foods have a lot of calories but leave you hungry.  For me that's mostly carbs.

All things being equal, I try to avoid eating too much crap (too many carbs, candy, beer, etc) but I don't have a rule against anything in particular.  Everything just gets tracked.

I drink a lot of Diet Dr. Pepper in the day and herbal tea at night.  I don't know, these help.  Celestial Seasonings' line of Zinger teas sort of taste like dessert.  

Exercise is great and everyone should do it all the time.  But I find that even if my various workout apps tell me that burned for example 300 calories running, and then I eat an additional 300 calories that day, I'll actually start to gain weight.  Maybe this is because when people exercise they sometimes fidget and so on less throughout the day, balancing out the increased activity, or maybe it's because I just don't burn as many extra calories exercising as I think.  Who knows. So I only allow myself to eat about half the calories I've theoretically burned through exercising, which is hard, because exercising makes you hungry.

If you fall off the wagon it's pretty easy to get back on again.  I gained some weight over the holidays but am losing a little bit each day in January.  Eating more makes you hungry more, so there's a bit of a negative feedback loop here. I've found that the best way to get back on the wagon is to "reset" by doing a 24-hour fast (which just means skipping lunch, since I already don't eat breakfast).

Guess what.  Your daily calorie budget for losing weight may likely, at some point, become your daily calorie budget for maintaining your current weight.  It's not like you can diet, lose weight, and then start eating more.  Sorry!

Finally, why did I gain so much weight such that I have to do these things?  Getting older.  Office job where one of the primary daily activities is lunch, and where people always bring in treats.  Snacking at the desk.  Bad eating habits formed by many years of not being able to gain weight no matter what I ate.  It happens.

Fastmail Recommendation

I recommend Fastmail to people a lot, and this is why. It is a great email service. It costs at least $3 per month or $30 per year, so if that’s a dealbreaker for you then stop reading. Plus you might have to deal with stuff like IMAP and SMTP server names and possibly DNS settings if you want to get the most of it. But none of that is really all that hard.

Before I get to the technical advantages of Fastmail, just philosophically, email is a decentralized service, but for their personal accounts most Americans tend to use email from Google, Microsoft, and Yahoo.1 Other services are big in other countries but the story of centralization is similar. Google and Microsoft are pretty big in the enterprise as well. This is a shame—we should hold on to the original decentralized model of internet services, where no one company has too much control, as much as possible. And it’s very possible with email at least, even if very few people are using Mastodon instead of Twitter.

Also, I think (at least if you’re tech savvy enough and can afford to pay for a domain and then pay for an email service, which not everyone is lucky enough to do) you should own your own email address, one that’s not tied to some commercial service like iCloud or Gmail or even Fastmail. Mine is my first name @ my last name dot net. That’s not necessarily to say that you should never use those services—you don’t want to actually host your own email servers unless you really know what you’re doing. But the address that you share with the world should be one that is yours and that you can move from place to place, just like you can port your telephone number from one service to another. Similarly if you do a lot of non-work writing, it’s better to have it on your own domain, instead of just on places like Medium or, if that floats your boat, Facebook.

With that out of the way, I like Fastmail for a few very specific reasons.

Works well with your own domains. Lots of email services allow you to use other email addresses as aliases, but it’s less common to see an email service that allow you to easily use your own domains the right way. Fastmail does. When you own your own domain, you can have Fastmail host your DNS for you and configure your DNS records for you (that is, you set Fastmail as your name server; it’s not a registrar), or you can just set your DNS MX records (and a few other verification records) to point to Fastmail. That makes it so that the “official” mail servers for your domain names are Fastmail’s. This is not a how-to so I won't tell you how to do any of this; Fastmail's documentation is really good.

This means that when you send an email from one of your personal email addresses from your own domain, that email is properly “from” that domain name—your mail headers don’t show its being sent “on behalf of,” clients like Gmail don’t say things like “from via,” and the message is less likely to be flagged as spam.

The other way—the wrong way—is for emails to just be sent as aliases, which is sort of like faking your caller ID. Like faking a caller ID, it’s an officially supported thing, and also like faking your caller ID, it’s often abused. This means that emails sent as aliases are more likely to be flagged as spam or displayed in some strange way on the recipient’s end. With aliases, typically people will buy a domain name, and set up a “forwarding” address with it (say, all emails to will be forwarded to, and then tell the service (like Gmail) to just put the alias in the “from” field. But all of the other email headers will show that the message is just an alias, being sent through email servers that are not officially associated with the domain, and this is the kind of thing that spammers often do. (Even when you have everything set up correctly if you are using the Fastmail web interface, that will be reflected in the X-Mailer header of your message, and Fastmail might also be part of the Message-ID header. But those aren't bad things.)

You can get proper domain support with lots of services, of course, like a paid Google Apps account, though I don’t know what limitations there might be with using multiple domains. Some support them, others don't, others make you pay extra. With Fastmail you can use as many of your own domains as you’d like.

(When you set up an IMAP client with a Fastmail account, you just tell it the various email addresses you want to use with it, after using your actual Fastmail address as the user name. This is a normal IMAP thing.)

Works well with outside email accounts. Sometime you don’t own a domain but you want to send from an outside email address, but still not as an “alias” for all the looks-weird-sometimes/might-get-flagged-as-spam reasons above. Fastmail can handle this as well, because you can connect it to outside email accounts. Say you have a Google Apps account, but you’d rather use Fastmail to send mail from that email address. You can just have Fastmail send email from the Google mail servers, instead of using its own. Doing this means that using Fastmail with that address, including using the Fastmail web app, is exactly the same as using or Outlook. (You can also have Fastmail poll the outside email service and pull in emails from it via IMAP at regular intervals—but it’s better to just tell that outside email service to forward email to your Fastmail address. That way it comes in in real time, at least if that’s what you want.)

A nice thing about this is that when you send an email from a third-party email app like Apple Mail on iOS from that address, Fastmail knows to use the correct mail servers, even if those mail servers aren’t specifically configured in your app. So once you’ve set up correctly via the Fastmail web client, sending from it will always work the way it should for any third-party app that is configured to work with Fastmail without your having to manually give each third-party app the mail servers for the third-party email address. You just add addresses you use with Fastmail in this was as email addresses for the account in your IMAP client the way you’d add any other.

Push support on iOS. Push email support on mobile is sort of a pain in the ass since the “official” push method for IMAP is too much of a battery hog to actually use. So getting push on mobile usually requires using something proprietary, like Exchange instead of IMAP, or using a service-specific app like the iOS Gmail apps that hook into standard push notifications to sort of fake it. I don't mind third-party, non-system email clients but I really hate the idea of things like a "Gmail" app or a "Yahoo Mail" app--they're at best a necessary evil to work around limitations on push email. (Fastmail does of course have its own apps.)

There is another way, though—Apple has worked with different email services over the years, notably Yahoo, to support push on normal IMAP email accounts. A while ago Fastmail and Apple worked together to make this happen for Fastmail, as well. So when an email arrives in your Fastmail inbox it’s immediately sent to your mobile device. (I have no idea if any of this works on Android or if Android has something better.) This means that, for instance, if you use Fastmail to send and receive email from a third-party account, you get push for that account even if doesn’t otherwise support it. This is pretty cool.

Fastmail itself has written about all of this.

Fancy Email Tools.. Fastmail also has all kinds of nifty things like the ability to quickly scan for and delete duplicate messages. Sometimes when migrating email around you end up with thousands of them, and this fixes it. And if you’re moving to Fastmail, it can also import email from other email services very efficiently.

Uses normal folders. Fastmail just works like a normal email service, not a weird space alien one, which means it uses folders and not labels (as well as a very normal IMAP implementation). One advantage of this is that if you ever decide to move away from Fastmail in the future it’s very easy, just using an email client, to move all your messages somewhere else by dragging them from one folder to another. Gmail-style labels make that very hard and could result in your creating lots of duplicate messages (since the same message would be in “All Mail” and “Sent” at the same time, for instance).

Works with outside calendars. It's not really a huge differntiating feature, but of course Fastmail supports calendars and contacts. But while Fastmail has its own calendars of course, you can also point it to outside services (iCloud, Google) and have it use those calendars as well. My work calendars are on Google, and my personal ones are in iCloud, but I see them all (and can edit/make new events) in Fastmail. I can also see/edit them all if I point an iOS device or a calendar app to Fastmail, using CalDAV--meaning that Fastmail is a good way to aggregate lots of services and prevents you from having to sign into lots of different services on different devices. Fastmail doesn’t support outside contacts services using CardDAV, but it does make its own contacts available to third-party apps and devices via CardDAV. So, using Fastmail, you can cover the basics of email/contacts/calendars all in one place. It also has some file/web hosting support and a few other things like that that I don’t use.

It’s fast. It’s right there in the name, and it’s true.


Ok, that’s it.

  1. The best counterargument to using Fastmail is, if you’re a person who’s super tied into Google services, using Gmail gives you some integrations that you may or may not want—it can automatically scan your email for tracking numbers and tell you when packages are going to arrive, do some similar stuff with flight times, and so on. Plus Gmail makes it easy to respond to Google calendar invites, if you use the web app. But nothing about Gmail that involves using the web app appeals to me since I don’t like using web apps (they suck, because the web is good for publishing information but is terrible at software), and I hate all that automatic scanning stuff. Plus third-party apps do it all better, and in less privacy-invasive ways. ↩︎

Why Google’s Search Monopoly is So Vexing

Google’s search monopoly1 is very real, and durable, precisely because it is easy for end-users to switch to other search engines (I mostly use DuckDuckGo, for instance.) This is not a paradox—because Google has to keep working on and improving search to keep end-users, the fact is that most of them don’t switch. Google is still largely better than its competitors.2

For a lot of people this would be the end of the story—Google has a monopoly, but through its own merits, and therefore there is no consumer harm, though maybe some additional regulation (e.g. on privacy practices) is necessary. But that’s not the end of the story at all, because the fact that Google has such a lock on users means that the content, companies, and edge services that rely on Google as a source of traffic exist at its mercy. If you are a restaurant that depends on Google for people to find out about you there is no “switching.” You can’t just decide one day “Why, I would like to start getting my business through Bing, instead.” If you’re trying to compete with Google in one of its “verticals” (say, image or map search) you’re going to, at least, have a very hard time.3

I’m not even going to address the advertising issues.

It’s not impossible for other internet business to thrive in this environment, and for many of them Google is a good source of traffic and probably will always be. But if there is a risk that that traffic might drop other entities need to find a way to not depend on Google. You can own popular computing platforms, like Apple, you can be completely app-based, like Snap, or you can own one of the handful of sites that people go to directly, like Facebook or Amazon. But these are the exceptions that prove the rule. (As an aside this is not a paradoxical or contradictory statement. Crack a Wikipedia some time.) Most business are screwed if they suffer a sudden drop in Google traffic.

A further question is whether Google is a “natural” monopoly or just the regular old kind—that is, is its current dominance mere happenstance or does the search market inevitably tend toward monopoly, meaning that if not Google some other company would be dominant? Here, I don’t know, but I think there are reasons to suspect it might be a little bit of both.

The typical story with natural monopolies is that there are such cost advantages to scale that having more than one competitor doesn’t make sense. For instance with utility companies, there are very high fixed costs, yet marginal (or even no) costs associated with serving a new customer. So a new entrant would have to spend the same amount in fixed costs as the incumbent, yet the incumbent could always undercut the new entrant when it comes to serving any particular customer. It can spread its fixed costs around the rest of its customer base while the new entrant has to find some way to recover its costs from its smaller customer base. This is hard to do. (But, nothing is impossible. With a dense-enough population, a rich-enough new entrant, or a lazy-enough incumbent anything is possible. The question is what is actually likely.)

With platform companies things are more complicated—there are network effects and virtuous (or vicious, depending on your point of view) cycles. For example social networks become more valuable the more users they have, meaning that new users will gravitate to them, and software platforms are valuable to users if they have applications, and to application developers if they have users. These dynamics can be very powerful.

So the question is whether Google search enjoys some of these dynamics—and the usual answer is that, yes it does. Namely, the more users use Google, the more data it collects on user behavior, which allows it to improve its offering, which attracts more users, and so on. So like a social network the “natural monopoly” here is control of the user base, not data centers or CDNs (though those things help).

But, there is a limiting principle to these network effects dynamics: diminishing returns. Many people thought that Android, with its much larger user base, would drive iOS out of the market, the way that Windows very nearly did to classic Mac OS. Until the web became more important than desktop software, Apple was on a death spiral since all the good applications were being written for Windows first, or only for Windows, so even long-time Mac users had no choice but to switch, further creating reasons for new applications to focus on Windows. But this hasn’t happened in mobile, in part because the iOS user base, though much smaller than Android’s, is still gigantic: iOS alone, for instance, has more users than the entire PC market ever did. Of course it makes sense to write apps for this platform.

Similarly, though it seems obvious that user data improves a search engine—how much data is enough? Could the market sustain two or three very large search engines, supplying them with all the user data they could ever need? Or will the search engine with a slightly bigger user base always be able to eke out a slight quality advantage, thus attracting more users, and starting the whole snowballing process over again?

Here, I just don’t know. I think it’s an empirical question, and the value of data is not fixed but could change as more deep learning-style techniques are developed. For example Google was recently able to make its photo recognition algorithms a few percentage points more accurate by vastly increasing the size of the training data used—is this a meaningful difference? Would something similar happen in search? At the same time all of Google’s smarts don’t seem quite up to the challenge of preventing its AI from spitting out “fake news” answers to questions—it’s only as smart as its input data and the internet is not very smart.

This matters because the business and policy response is different depending on the nature of the monopoly in question. If there’s a natural monopoly you can’t just wish it away—maybe you regulate it, or nationalize it, or turn it into a co-op—but if you just break it up it will either re-form, or you will create permanent inefficiencies and quality problems.4 But if it’s just a regular old monopoly you can break it up, or wait for normal competitive dynamics to either replace it (what if people stop using search engines altogether? what if something new and more important comes along?) or create a new competitor (like Facebook was to Myspace).

I don’t have any answers here. The title of this post was “Why Google’s Search Monopoly is So Vexing,” not “Here is an Answer to Google’s Search Monopoly.” To summarize: Google keeps its monopoly by offering a high-quality (and free) service to end-users, which in turn gives it tremendous power over those sites it links to. Those sites can’t “switch” anywhere. So typically those sites complain, not users, and it can be hard to distinguish between legitimate complaints and illegitimate ones (for example, if Google de-lists a scam site, this is not a problem but the scam site might make the same antitrust arguments as a legitimate site). At the same time, it’s not clear whether search engines are inherently a natural monopoly market where, left alone, cost or quality considerations will always drive towards monopoly, or whether it’s just one of those things that happens, and where competition or technological change, or antitrust action, can shake things up. Because search engines are at the center of internet, which is now at the center of commerce, culture, and politics, this is all quite vexing.

  1. Its market share is a bit over 60% in the US and a bit over 80% globally, and is over 90% in some markets. Monopoly is not either/or; the question is whether it has market power.

    However I don’t think search is Google’s even strongest monopoly, from an end-user perspective. Youtube is. Google’s search monopoly is probably more broadly important to the economy but (from an end-used perspective, again, and as discussed in this post) there are alternatives. But for content on Youtube there are often no alternatives. This is trivially true of many media sites—there are no alternatives to Netflix for Stranger Things, for instance—but for the kinds and sheer quantity of content Youtube hosts I believe it would be a much harder platform to displace.

  2. DuckDuckGo differentiates itself through a focus on privacy and power-user features but most people value good results much more. Even I turn to Google when DDG lets me down—and DDG supports searching Google via its “bang” command feature—one of those power-user features. ↩︎
  3. There’s a simple rejoinder here, that if Google doesn’t do a good job in presenting results in a fair way that it would risk losing customers—and to an extent that may be true. If Google de-listed Amazon or Facebook it might lose customers. But if Google decided to just flat-out de-list some smaller category of site and launch its own how many customers would it actually lose? And even if they quality of Google decreased somewhat, how long would it take end-users to even notice? ↩︎
  4. Think it through for a classic natural monopoly—how would you break up the water company? You’d just create a series of smaller natural monopolies (which is probably inefficient, though having just one nationwide water company is not ideal either) or you’d have to lay down a competing set of water pipes, which is inefficient. Or you’d create some kind of pseudo-competition like some states have with energy, which might be valuable but is just a form of regulation. ↩︎

bandwidth and electricity are the same but also different

Electricity is ultimately a scare resource (burning coal, collecting sunlight), with the complication that it is being produced all the time so there are good reasons to try to offload usage to non-peak hours, as opposed to just reducing usage entirely.

Bandwidth is beguilingly similar to electricity, in that people access the internet via this huge network of wires, and there are similar peak usage problems.

Another similarity between electricity and bandwith is that non-usage can be wasteful--there has to be slack to accommodate burst and peak usage, but for the most part if communications infrastructure isn't being used to its full capacity it is being wasted; it's like building a six-lane highway that no one drives on.  Similarly, particularly with fossil fuels, if power is generated but not used it just means that the fuel used to generate it was wasted.  But there is a difference, in that electricity generation can be ramped up and down, while the amount of bandwidth available on given infrastructure remains fixed. (That's not to say that an individual user can't buy more bandwidth on demand, meaning that increased usage could cost an individual ISP more in, for example, transit costs; it's just that the total global bandwidth available remains fixed until someone installs more stuff, in the same sense that road capacity remains the same until more roads are built.) Also I suppose it doesn't matter too much if more solar or wind power is generated at a given moment than can be used--especially if it can be usefully stored in batteries or by pushing water uphill.

However I think drawing too many conclusions from the similarities between electricity and bandwith can lead to problems, because ultimately electricity is a scarce resource that is used up, while bandwith is mostly just a congestion problem. The idea situation is that all communications infrastructure would be used to just-below its peak capacity at all times.  (Similarly it would be great, if impossible, if people's road trips were evenly distributed throughout the day--no rush hours or event congestion.) But there's no reason not to use as much bandwidth as is available--to the extent there can be waste the waste comes about from building unnecessary infrastructure.  This seems very different than electricity where it always makes sense to try to reduce usage.

So, a per-bit metering system for bandwidth would likely be a bad policy because it would discourage usage all the time, but it makes no sense to discourage usage unless there is actual congestion.  Thus while it might make sense to use pricing or other means to try to smooth out bandwidth usage or even to reduce it overall when infrastructure is nearing capacity, unlike with electricity encouraging "conservation" does not always make sense--instead you often want to encourage use of the infrastructure, and the best way to do this might be a metered price of zero.

Assorted Free Speech Takes

So I don’t have to try to tweet all this, here are my various takes on the free speech controversies that have been welling up lately.

These are my personal (and provisional and subject to change) opinions and I don’t speak for my employer or any organization I belong to.

  • I am almost but not quite a free speech absolutist when it comes to the government: prior restraint is bad. “Hate speech” is not an exception to the First Amendment. I think the ACLU’s suit against WMATA is correct. If the government is going to sell advertising space (which it doesn’t have to do and probably shouldn’t) in public places then under current law it needs to sell ads to NAMBLA and Nazis, and in any circumstance must accept “political” ads.



  • The settled law that viewpoint-neutral time, place, or manner restrictions on speech are legal is correct. This is true even though people will get tricky and try to characterize viewpoint discrimination as viewpoint-neutral. A rule that leads to fact-specific arguments in situations like and occasional abuses is better than doing away with time, place, or manner restrictions. Based on reading the briefs and looking at the facts as they were known and presented to the court I believe ACLU was wrong in challenging the right of Charlottesville to move the Nazi demonstration to a different public park, and the judge was wrong in his decision, and that the events bear that out.


  • Major online platforms should afford everyone due process. Whether something is a “major” online platform depends on the specifics of the market and the availability of alternatives. They should be required to have clear content policies in advance, they should be required to give notice for any content that they wish to take down (unless it is per se illegal, see above), they should permit people and opportunity to export their content (again, unless it is per se illegal) and they should give people an opportunity to challenge the “charges” against them before a neutral decision-maker and have their content reinstated if victorious. (Maybe something like Hal's idea would work, though I don't think it's the right approach for ISPs.)


  • If a platform is an actual monopoly (probably whether a natural monopoly or otherwise) or more "infrastructural" in nature, then it should be subject to common carrier rules of some kind. This is a heightened version of the above, and it does not require them to carry per se illegal speech, but it would put some restrictions on their ability to decline to carry content even for reasons that are delineated in advance, and limit their discretion in individual cases. Some kind of restrictions that go beyond what the government would be permitted to ban may be permitted for some, but not all common carriers of this kind—for example with infrastructure-type services common carriage should be strongest.


  • The best way to avoid the abuse of power is to prevent accumulations of power. To the greatest extent possible would should avoid having any single platform having the ability to remove someone from the internet, which might mean breaking up, or restricting the size or conduct of some companies, even if they've done nothing "wrong" to achieve their position. Although if a natural monopoly exists we should deal with that fact instead of idly wishing for more competition. (Why not transform them into customer-owned co-ops?)


  • This is not really a "platform" issue but one of to what extent businesses should have the right to decline to do business with someone for any reason.  We already prevent discrimination based on protected categories (mostly immutable characteristics) but should we go further?  Obviously I think in some cases the answer is yes.  The reason to talk about "platforms" is that they tend to be more closely speech-related so these issues are more salient than they would be, for instance, in the case of a gas station.


  • In general, unions are very good and the best way to protect workers from arbitrary management. But even non-unionized companies should afford due process rights to employees, who should be able to challenge terminations. Due process must always be available to avoid pretextual termination or ones depending on falsified or no evidence. Believing and advocating hateful things, even outside the workplace, should be grounds for termination, but those kinds of things need to be spelled out, in advance, though not to the same level of precision that First Amendment exceptions need to be spelled out. For example, it would be ok to fire someone for denying the Holocaust (or the Holodomor or similar events). It would be ok to fire someone for believing in racial supremacy or gender-related pseudoscience. It should not be ok to fire someone for being a libertarian or a socialist or a supporter of the divine right of kings. Believing in things that are just generally stupid should not be grounds for termination. However companies should have the ability to buy someone out in such a circumstance, again under some clear pre-set rules.
Source: my brain

Letter of Recommendation: Detroit Techno

The radical act of Detroit’s techno rebels was that they entered an inhuman network of machinery and found a voice within it — which aligns them with a different Detroit legacy. The city’s first independent black autoworkers’ organization was called the Dodge Revolutionary Union Movement — or DRUM. Members chanted the word while marching, as though keeping a beat.



But a rack of lamb is not the Roman Empire.

“And yet, I think it was a better description of the market than they had, because the least of the market is the food.”

"What is it then?" Hardesty asked, though he already knew.

"Buying and selling, faces, the color, the light, the stories that breed within it, its spirit. Where else would you find all these clear lights strung so high and gleaming in the cold?" she asked, indicating the chains of electric bulbs over the stalls. "Harry Penn got a telegram from Craig Binky, that said, ‘How can you cover the market and not mention food?' Imagine, they send telegrams between two offices on the same square. Harry Penn cabled back, 'Eating assassinates the spirit.' 

"I like to eat," she said. "In fact, I'm hungry right now. But a rack of lamb is not the Roman Empire."

Mark Helprin, Winter's Tale.

This pretty much sums up how I feel about food writing.

Comments in the mp3 available here.

Soviet Jokes.png
Embed Block
Add an embed URL or code. Learn more