The Boundaries
of the Network

Session 3: Architectures of Innovation

Thursday, September 6, 2001

David Isenberg: Today I'm going to violate my own rule of no presentations because we have some absolutely extraordinary presenters. Well, maybe in this crowd maybe I should turn the expectation knob down.

The topic this morning is 'architectures of innovation.' Before the break we have Clay Shirky of Accelerator Group. After the break, we'll have David Reed. We will give Clay half an hour and then give Larry Lessig a shot at commenting or rebutting. After the break it will be David Reed, and Bob Lucky will get a well-practiced shot at commenting and or rebutting.

Clay Shirky: I'm going to try to do 4 things in my allotted time.

First I want to compress the last 20 years of computing history into one drawing. I will skip some things when I do this, so if I miss your favorite turning point, we'll get back to it later. I want to do this to stipulate what I think the background was in which peer-to-peer happened.

Second I want to define peer-to-peer and talk about what that definition means.

Third I want to sketch out some of the architectures that fall under rubric of peer-to-peer and what the ramifications of what those different architectures are.

Then, fourth and most important, I want to take on two current issues in peer to peer -- one of them being private name spaces pointing to the public internet and the other one being a potential hardware control point for preventing users at the edge from participating fully in the Internet.

We got the modern computing world, and everything we recognize as being based on it, in about 18 months starting on January 1 1983. On that day the IP network launched, although badly. That was the moment the whole network switched over to the familiar dot-delimited 32-bit address space that gave everyone their numbers. Anyone time-transported from 1983 to now would look around and see that we are still using that address space. They might be confused as to why, but it's still recognizably the same beast. About a year and a half later we got the domain name system which gave us the familiar human-readable addresses so people didn't have to go around remembering the 32 bit name space -- [in other words, e.g.,] IBM.com automatically mapped to its IP address.

At the same time as this was happening -- as we got the modern Internet -- we also got the modern personal computer. In 1984 Apple famously launched the Macintosh, which brought a GUI [graphical user interface] to the text-only operating system. Suddenly we had a machine where, if someone were to come from 1984 and sit down at any one of these machines, they would see windows icons, and [menu] pull downs and they would say, "Obviously a lot has changed but this is recognizably the same thing that I saw in the beginning of 1984."

So that 18-month period was a period of incredible foment. Then, astonishingly, nothing happened -- for a really, really long time. These two revolutions did not intersect in any significant way. [It was] a highly connected revolution for the rabbis in the center, a highly disconnected revolution for the Goyim at the edges. No overlap at all.

The philosophical characteristics of the PC and the Internet were so different --that's a whole side conversation, but it's essentially protocols which are both group defined and group owned versus APIs [Application Programmer Interfaces] which, although they are used by groups are defined by individual efforts, meant that it wasn't easy to cause these two things to overlap.

Then about 10 years later came the event that that changed all our lives and fundamentally altered the global economy, which was, of course, the image tag. Tim Berners-Lee did the heavy lifting for the Web, but until the image tag came along the Web was really an awesome replacement for gopher.

Isenberg: What's an image tag?

Shirky: The Web prior to the image tag was was a much better but still text-oriented way to find information. The image tag is a tag that embeded clickable images into the mosaic browser. It was the thing that brought the GUI to the previously text-only Internet in the way that, 10 years before, Apple brought a GUI to the previously text-only OS.

For the first time, and all of a sudden, after 10 years of this side-by-side development we had a reason to connect these two worlds [the worlds of the internet and the PC], and we made a total hash of it. We did that because these PCs at the edges had none of the characteristics of the machines at the center -- they weren't reliable, they weren't always on and they weren't always connected. The OS's were flaky and crash prone, the hardware was poor, and worst of all they had to connect using modems, so that even when they could connect to the network, the connections were extremely intermittent for a wide variety of reasons.

Now the regime of connectivity in the Internet proper assumed that a machine was always on, always connected, and always had the same IP address. The IP address was the core attribute of a machine. You can still see the effects of that today, where adding a new host name to an existing IP address is trivial, but changing an IP address under a host name can take days to propagate through world's routing tables. Permanence of IP address was the core attribute of connectivity in the center.

You could not assign permanently connected, permanent IP addresses to the edge of the network for two reasons: One, the connectivity was so poor. Modems were just a bad way to connect. But two, we suddenly ran into the ramifications of a 32-bit address space, particularly one that was assigned in large bulk -- Class B and and Class C addresses and so forth. We were connecting these machines, once we had a reason to, much faster than we had address space, so people began to dynamically assign IP addresses, and so at any given moment you could never count on any given computer being on, being connected, and having the same IP address.

In this period, 1994 to 1999, the first Great Wiring, we didn't so much connect the PCs to the Internet as we hung them off the edges. Now we had two zones, a first-class zone and a second-class zone. The zone of second-class connectivity [for PCs] meant that these machines could always consume but never produce or provide resources. So we had gone from a peer-to-peer core network, to a client-server network.

Now people noticed this early on.

Man: One observation. This is all true but before any of that happened PCs were used as dumb terminals in corporate networks. This perpetuated client-server a lot more than it should have. It was probably the most expansive use of address space at that time.

Shirky: I consider terminals not to be connected to the network, rather they are connected to something that is connected to the network.

Man: Universities were using computers as terminals to get to supercomputer research centers. That was thought to be the primary use of the Internet, as basic IP transport to connect some dumb terminal in Rochester to one in Ithaca.

Shirky: As it was through the browser revolution when your PC was considered to be a life support system for you browser; this was the public update of that mode.

People noticed that we were not connecting PCs to the network as first class citizens, and the answer came back, "Yeah, yeah, it's going to be fine. Fiber-to-the-Curb, IPV6, any day now. Set your watch by that, and everything will go back to normal."

So we set our watches, and calendars, and then we got tired of waiting. Finally one company broke ranks and said if this system is not letting us use the resources or connect the PCs, we are going to have a different system. And if the bottleneck is that we can't give the PC a permanent IP address, and therefore we can't give it a domain name, we need to rethink the whole thing from the beginning. The company was Mirabilis and the product was ICQ.

What Mirabilis did was they turned the domain name system inside out. They said we are going to make the human readable part of the address permanent and assign temporary IP addresses to it. So if I want to have chat session with David, what they did was they set up a central database at the center of the network, and when I would connect, it would take my name and assign and IP address to it -- "I am currently at this address" -- and when David connected, it did the same thing. When David wanted to chat with me, the ICQ server would say, "You guys want to chat, you're both in the cloud and you both have IP addresses, you figure it out." It would offload all connections 2 through N, to the cloud. Then David and I would establish a connection for the duration of our chat, both using our current IP addresses for an ordinary through-the-cloud connection that did not go through the central server. So it used the central server to bootstrap an ordinary through-the-cloud connection.

ICQ, just before AOL bought it, was running half a million simultaneous users on 13 standard Pentium PCs running Linux.

One of the big deals about this stuff [that goes to what Roxane was talking about yesterday] is that this will affect the hardware manufacturers in the same way that we talk about bandwidth providers yesterday, because super-efficient use of resources at the edge of the network is bad news for people that sell storage area networks and big fat servers in the center.

This is the beginning of the routing-around-the-domain-name-system. ICQ chat was a wildly successful application. I suggest we not use "killer app" -- it seems unhelpful. ICQ has wildly successful adoption characteristics, but it used this as just an inverted domain name system.

The real action came with Napster. Napster figured out that not only might David and I like to talk to one another, so we have this two row database of our names and IP addresses, but David is a big Metallica fan and he has all of their MP3s stored on his drive. What Napster realized is that if you contact the Central Server and upload information, it doesn't have to be just name and IP address, it can be anything. This is a database --you can hang any number of columns off of it. So what they figured out was that, in addition to David's current the IP address, I will get David's current music collection.

Man: When you say those columns what are they?

Shirky: [Refers to drawing.] Here is "name", here is "IP address", and here is "music". So the rest of the rows are, e.g., you have copy of this Britney Spears song, this Frank Sinatra song and this Metallica song. I can hang any amount of information off of a user name and IP address combination. Once I can do that I can have a look-up in the center. The Napster server says these are all the songs that David has. And when I go to get a song this is the round trip I make. I go through the cloud directly to David's machine. I don't go through Napster.

(Napster at one point thought it would get out of legal trouble because it was not storing any copyright infringing material, nor was it passing any copyrighted material between users, even tangentially, it was all going through the cloud. That was their idea.)

So that innovation --saying that if I can bypass DNS and make the human readable address the permanent part, and update all of the other transient information -- that is the thing that lets me break this second class bottleneck and bring the edge-connected resources directly into contact with one another without forcing them to go through the center of the network.

So my working definition of peer-to-peer is if it falls into both of the following two categories:

First, if it aggregates unused or underused resources at the edge of the network, bandwidth, computation, storage, human presence -- and really I want to emphasize that this can be anything. I have not yet seen a peer to peer application that needs 30,000 sound cards to run, but someone out there is working on it and I'm going to be awfully interested when it launches.

Second, if it bypasses the DNS system to aggregate those resources because it has to operate in a regime of unpredictable or intermittent connectivity

Scott Bradner: I don't see why that's a determining characteristic.

Shirky: My argument is that it's a historical characteristic.

Bradner: But I don't see why a determining characteristic. Many of them do that, but that doesn't mean someone with a fixed address can't be participating in Napster.

Shirky: That's right. It has to provide a way for things that don't have fixed addresses to participate in the system, so you can have both fixed and non-fixed IP addresses, but it has to make provisions for non-fixed IP address machines to work on equal footing with fixed IP address machines.

Napster bootstraps with DNS because server.napster.com is the pointer, but the nodes in system have to be able to operate outside the DNS system. If you use Napster even if you have a fixed IP address, you do not have a DNS address, you have 'Green_Day_fan'.

Bradner: I don't think that a determining characteristic, it has to be a historical accident because dynamic DNS didn't get done early enough.

Shirky: Right, I'm going to argue is that it is a frozen accident, that this accident, having happened, is not going to unhappen, because there is a generation of people for whom Napster is just information.

Bradner: That may be true, but I don't think that should be core to your discussion. I think it's totally tangential. The ability to have someone who is at a non-fixed address participate is a perfectly legitimate requirement. But what the technology to do that is seems to be irrelevant.

Shirky: I'm going to get to the DNS system again at the end because I do think this is going to have some significant ramification. My argument is going to be that it works in practice, even if it doesn't work in theory, and having happened, a generation of engineers is going to go after this whether or not it's a necessary characteristic of the network.

Man: Couldn't you fix Scott's observation by saying that no necessary central name service or no DNS required?

Shirky: Napster actually has a centralized name space -- it is more centralized than DNS.

Let me hold this until we get to the DNS discussion, because that is actually one of the two issues I want to bring up.

It is important to remember that "peer-to-peer" is a label and not a definition. The phrase came out long after the applications it was meant to describe. When you now hear people going around saying Napster is not peer-to-peer, we only try to find a word like peer-to-peer because of Napster. So there's no point in defining it out of the club after the fact. Many peer-to-peer applications, in fact most, use some aspect of client-server to bootstrap the connectivity. Pure distributed peer-y-ness is not actually a necessary condition for something to be considered peer-to-peer.

There are a bunch of architectures that fit under this model. I want to draw three of them to make the point that there are lots and lots of ways to aggregate edge resources.

The first one is the model used by SETI@home. There is a central server dispatching [pieces of computation]-- they look for a needle in a haystack by dispatching the hay one handful at a time to PCs at the edge of the network, which are coming back and saying "I have the needle" or "I don't have the needle" in this handful of hay. SETI looks a lot like client-server. It's a star topology, but in client-server the computational heavy lifting is done at the center and the results are distributed to the edges. SETI@home reverses that by doing computational lifting at the edges of the network and redistributing the results back to the center. That is one way to use the resources at the edges of the network outside the bounds of the client-server architecture.

The second [architecture], which is the commonest [in P2P systems], is the sort of triangle that I drew when I was talking about ICQ and Napster, where a central server bootstraps the connectivity. The first connection that anyone in the system mixes with the central server, but all conversations and file transfers and whatever else goes on in the system happen in the cloud away from the central server.

Model three dispenses with central server and says, all these other things are half measures, but what we really need to do is move the idea of the cloud all the way out to the edges of the network. We need to make a way for nodes to be able to address one another without even having the bootstrap problem.

So what Freenet attempts to do is restore Internet style connectivity to all nodes irrespective of whether they have hard coded IP addresses or not. Gnutella also tries to do this. It uses local bootstrapping, which is to say when you connect with any popular Gnutella server like LimeWire or BearShare or ToadNode, it goes to a well known Gnutella host and then distributes it.

Freenet is really an attempt to create an almost unbreakable and untrackable network where no node has a bootstrap. Freenet is an interesting idea, but it seems to be the Xanadu of our generation, which is to say the software that is so ideologically perfect it will never launch. Because its such a large attempt to create this sort of fabric built-in, that many things they could have done pragmatically they didn't do, and the result is that not nearly as large as Gnutella.

(Has anyone has read Dick Gabriel's paper, 'Worse is Better: the Lisp vs. C Debate?' Gnutella is objectively worse than Freenet, but there are many more people who want to make Gnutella better than want to make Freenet at this juncture.)

Man: I'm told that Napster could not have been invented by anybody at Microsoft because the older guys would have said it can't be done. So when you refer to the Xanadu of our generation and say this ideologically pure notion is non-achievable, would that be susceptible to things you couldn't know that were obstructing your view...

Woman: You many need to define Xanadu...

Shirky: Xanadu is Ted Nelson's hypertext system, which had many characteristics of a perfect hypertext system. Every piece of information existed in only one place, all links were implicitly two-way, all links were self-updating when information moved. And Berners-Lee threw out most of that work when he invented the Web. What he said was, when information moves we are just going to provide a mechanism for everything breaking hopelessly, and we are going to call it the 404. And that lack of coordination -- the fact that they didn't require global coordination -- meant that the system could scale much better because things could break asymmetrically and be updated asymmetrically.

So I'm not saying that Gnutella is the only example of this, while Freenet won't work. What I'm instead trying to do is say that there are two systems that occupy this zone -- Gnutella and Freenet. Freenet has the superior theoretical characteristics to Gnutella, but my bet is that because more people are using Gnutella and trying to make it better, it will improve over time, albeit in small increments. It may always be worse in practice than Freenet is in theory, but it has better survival characteristics.

Man: Are we talking VHS vs. Beta?

Shirky: Right, were Freenet to be implemented, it's survival characteristics would be superior. But it's not used with anything like the breadth of Gnutella. So, like the Lisp vs. C debate, where Lisp and C were both implemented, my bet is that Gnutella is going to improve because of its user base

Man: The only problem here is that you have Time Warner and a whole bunch of other new companies where Gnutella is the next target in their sites. Freenet is going to survive that.

Shirky: Freenet in this case is the cockroach.

Man: What happens when someone throws their shoe at it?

Shirky: This was one of the two last issues that I want to bring up. The last thing I want to say about this architecture is that although in the popular press you see "Napster was great, it was great, but then they came along and they killed it", there's a whole bunch of 19-year-old Computer Science students for whom this is information.

[These folks see Napster and think] "Aha! If I have an application that needs 6000 gigs of space, I don't need a 6k gig hard drive. I just need 6,000 people to give me one gig each." These are just architectures that a lot of people take for granted and they are going to start reappearing in all sorts of places that are very far from mp3 file sharing.

There are still two issues I'd like to bring up:

The first is name spaces. Here's the thought experiment. You have a PC under your desk, and let's say I gave you a file and said, "Here, serve this file from your PC." What would you have to do? You'd have to go to register.com, you'd have to pay them your money, you'd have to wait for it to come, you'd have to mess around with NIC handles, then you have to get somebody to give you a fixed IP address. You have to map the two and wait for the world's routing tables to update, then you'd have to download Apache and screw around with...

Bradner: There's nothing to do with routing tables in that -- the routing tables have the address whether you are using it or not.

Shirky: I'm sorry, I meant the DNS entry, in the DNS server. It would take days, it would cost you money, and you would have to involve a lot of other people. If the file I gave you was an MP3 file, you could download Napster and be serving it in five minutes. So what Napster does is it brings usability to name space configuration. With Napster, you can create a routable, human-readable address in five minutes without asking for help or permission. And this is a huge change.

The number of people in world who can create a routable, human-readable address has gone from an order 10 thousand to order 10 million in the last two years. So one of the big issues in this world where address spaces owned by Napster and ICQ are, what do we do if anything about privately owned address spaces that are pointing back out to the public network? This is something that people didn't really think possible; DNS seemed to be the only game in town for a long time.

We've just seen two companies build up hugely valuable assets by owning private pointers to the public network. The other interesting characteristic about those name spaces is that they violate almost everything we know in engineering terms about name spaces. They are user generated, they are flat, they are not well characterized. They encode information in the name, and yet, they have been wildly popular.

There were more Napster addresses created in Napster's short span than there were in the entire existence of the domain name system. So we have seen two examples where people have created tremendously valuable (and fragile in many engineering ways) name spaces that nevertheless worked well. So, what do we have to say about that?

My second concluding issue, which I hope will lead into what Larry Lessig is going to talk about, is control points. Something Larry said yesterday that I thought was very smart is that we've had this idea that the network itself is going to necessarily resist these attacks. But as we have grown with the network, we can see that there are some places where there's a bottleneck. Where people who want to control the ability for people at the edges to use their PCs to publish can get at it. The [ability] to be a global publisher has fallen below the level of the requirements of being a corporation, it's fallen to level of the individual. With Napster I can publish material to the world without even having a lawyer.

So who do they [e.g. the RIAA] go after? They don't go after me, who is infringing the copyright, they go after someone upstream who does have a lawyer. So where are those technological chokeholds? I'd like to propose that there are three and then I'd like to open it to discussion.

One is the hardware manufacturers. People who make disk drives are talking about installing digital rights management software into the drive, building DRM into the hardware directly. DRM is an attempt to take all the inconvenience of the physical world and map it into the digital world and then tell users you have to pay the same amount of money for content online as offline because it's costing us so much to make things so inefficient. If that works, the place that will happen is at the hardware manufacturer.

Jerry Michalski: You should point to John Gilmore's essay, "What's Wrong With Copy Protection" at toad.com, I think. He sorts through the various scary ways that the media and copyright industries have already crippled the next generation of PCs.

Shirky: The second place there is a choke point is the Operating System for PCs. We have about 1.08 vendors of OSs for PCs, and if any large subset of that group were to alter their OS, that's obviously a major chokepoint.

The third place it can happen is ISPs, who have two problems, one is that they don't want to be sued -- Pac Bell just announced that it was dropping 80% of the alt.binary groups because they are afraid for be sued for copyright infringement. The other problem is they sold you your bandwidth on the assumption that you won't use it. So if you start using your upstream bandwidth to publish, their under-provisioned network suddenly looks problematic.

So Cox Communications is the first one to go, and they banned people from using Napster on the grounds that it's a server and many of the ISPs will be following suit.

Now let's open up for questions.

Man: What about companies like Real Networks?

Shirky: Here's the interesting thing to me. I don't think that software is a choke point. The software market is open enough so that if any one software company builds this in, it can be routed around.

So of the four possible places [to control content, i.e. including software], the place where we have a functioning market is not the place where we have a choke point.

Michalski: Here's one thing I want to throw in -- Napster is not a great application if you love music. It's not about music appreciation. Napster's implicit message is hording files. I better get all the MP3s I can on my machine because they are about to shut this thing down. Now, if everyone were able to share, cache and move files around, you've separated whatever application you are going to run to enjoy music, which gives you context about music, which lets you tell stories about it, link music to other music -- separate that from where the files are actually stored.

I'll paint two alternate options -- one is the celestial jukebox model that the media companies want to sell you -- and there are several of them already in progress and the media seems to think that that's where the only battle is. In this case, the songs kind of get stored on some big celestial server, the celestial jukebox in the sky. You then pay a subscription and listen to all you want, but you are usually fetching the songs from the server.

In the opposite world, look at Freenet (the most distributed version). Instead of thinking about, "I'm going to have copies of all this music, I'm going to use up 30 gigs of hard drive storing all the songs here," think about the frequency of use and the patterns of traffic and allow those files to settle across the Net wherever they are being used, with some people volunteering some local cache and so forth.

If I listen to a song once a month it won't be cached on my machine. I may just fetch it whenever I play it. But if I play a song every day, I may cache that [in my] machine that may be local to me, and then other people from my neighborhood may pull it from me. And all of a sudden if you allow the traffic to move to natural patterns of use and let natural caching to happen, you don't bring the network down. It doesn't freak everything out because suddenly you have extremely local fractal use of files. And that's great. That's pico cell sharing. I don't know what you want to call it, but it doesn't fuck up the network.

Isenberg: Let's hold the questions and give Larry a few minutes to comment, and *then* we will let all hell break loose.

Larry Lessig: I think Clay is doing two great things. First he is attempting to identify an architecture that second is subject to push-back from the dark forces. I want to reinforce his point, but shift it a little bit. The part I want to bring into focus is where dark forces can do damage. Dark forces cannot have any harmful effect in competitive markets. Put differently, the great thing about competitive markets is that they defeat strategic behavior on the part of competitors. Thus, when we try to think about where the architecture is vulnerable to the dark forces, we need to locate vulnerability at those places where the markets is least competitive.

To do this, I want to suggest a slight modification to Clay's picture.

Think about three layers to the network -- one is content, the middle is code and the bottom is physical. The physical layer: the wires and machines that built the Internet. The code layer: the protocols of the network, the IP, et cetera. The content layer: both mp3 files and web pages, as well as bunch of software that is essential to make the network run including Apache servers and the Microsoft OS.

The essential characteristic of the original Internet -- indeed, the single brilliant characteristic of the original Internet -- is that the code layer has a feature that Saltzer, Clark and Reed call end-to-end. This design pushed intelligence to the edge of the network -- stupid networks in Isenberg's conception of it, with smart applications.

This design meant that network [architects, owners, builders] were not in a position to second-guess how innovation in this network developed. Stupid networks meant the owners of the network didn't have a role in how the network developed. The networks were just shipping packets and didn't have any intelligence to do anything with applications or with the content. This was a competitive environment in the sense that applications and content developed as people at the edges wanted it to develop. That was the genius of the original network.

In a formal sense, the code layer constructed an "innovation commons." It's a commons in that there is no one in control over who gets access to the resource and who does not: everyone has equal access. And it is an innovation commons because the resource that is held in common is the right to innovate --the end-to-end design meant that anyone can innovate for that network.

Now each of the other two layers in the Internet also have commons-like features. But these other layers also had essential non-commons like features -- proprietary characteristics as well as commons characteristics. Start with the physical layer. The original network was run on AT&T's wires. That physical network was owned. It was not a commons.

But a critical feature of this network -- maybe the most important, accidental feature of regulatory history affecting this network -- is that at the time the Internet began to take off, AT&T was a disfavored monopoly. It was not allowed to play games. The network owner was not allowed to behave strategically against new competitors. People were entitled to connect different networks to this telephone network. Thus the original network had a part that was proprietary at the physical layer, but common-carrier like regulation meant that physical layer also had features that were commons-like.

The content layer too had proprietary and commons resources. Much was proprietary --proprietary software, copryrighted material, etc. -- but a critical part of what made this original network function was the range of free and open source software that also lived at the content layer. This content was not exclusively in the public domain -- some was, such as TCP/IP, HTTP and HTML -- but open source and free software is not in the public domain. It is software that is licensed in a way that guarantees that everybody can take the source code and build on it in certain important ways. That makes that software also a kind of innovation commons. No one can close it down. It's constantly there for people to be able to draw upon. Thus here too, at the content layer, there was both controlled and free resources.

At each of these layers then there was an aspect of free resources, an aspect of a commons. At the middle layer, there is essentially a commons -- the structure of end-to-end predates what we have to think of as a commons.

What we have seen in last 5 years is pressure from the physical layer and pressure from the content layer to corrupt the code layer. That's the bottom line. Pressure from the physical layer happens as the network moves from telephones to cable, which is the dominant broadband provider right now. Cable is not regulated in the same way the telephone lines were. It is allowed to discriminate however it wants to. In fact, a Broward county court has ruled that the first amendment means that cable has a constitutional right to discriminate in its deployment of services, which means the obligation that they operate neutrally at the physical layer is constitutionally prohibited. The first amendment does not allow the government to tell the Cable Company that they have to be neutral at the physical layer.

Thus cable companies are developing technologies to enable them to exercise control over the types of applications or content that gets served across the network. Policy based routers make it easier to figure out which content goes quickly and which goes slowly, thus building a potential chokepoint of the sort Clay was talking about.

This is why we need to worry about whether there is competition at the physical layer. If there isn't, dark forces can play strategically. That's why the wireless stuff that Dewayne Hendricks will talk about is extremely important.

The content layer is even more frustrating. Both in the patent world and copyright world, lawyers are securing to existing interests extraordinary power to veto how the network develops.

Peter Cochrane: This is all very nicely reflected in Europe, except for the simple difference that everybody ignores the lawyers. Over here do people actually comply?

Lessig: Yes, there's no such thing as ignoring your lawyers in the U.S. [laughter] Five years ago there was a huge industry called music over the Net. Twelve companies were out there that were keen on finding ways to distribute music across the Net. Now there are two.

What happened is that every single mode of distribution that the RIAA didn't like, the lawyers shut down, so MP3.com, Napster, a whole scad of them, that have been shut down because the content industry has succeeded in convincing the courts that unless they have perfect control over their content...

Cochrane: The reality is that young kids don't give a fig... there is now peer-to-peer networking and for the first time we see CD sales dipping. If you ask yourself the question why breaking up those dozen music distribution companies, which were actually aiding sales, have now led to a new paradigm where...

Lessig: I worked on the Napster case. I think the data is a lot more complicated than that. In fact Napster seemed to have the effect of increasing sales and when it was shut down, that's when sales fell. It's not because everybody went to these other tiered ways of sharing music. Napster really did reduce the number of people who were trading MP3s when it was shut down. Now I agree there are kids out there who are still doing it, but the question is not whether 20% can get around the regulations. You can't think like an engineer when you are talking about law. Engineers think if it's not 99 percent effective it's not effective. Lawyers think, man 60%, 70%...

Tom Freeburg: I submit that you don't know what's happening. I'm very sure of it. What often happens when you make a very public attack on a very public and commonly understood figure, as happened in the Napster case, a whole lot of people say the hell with it and go underground and go right on doing the same thing. The fact of the matter is that now when music sharing goes on you can't find it any more. All you did was shut down the one clearinghouse where the stuff was published.

Lessig: Wait. You can't both say that you can't find it any more and volume is not affected.

Tom Freeburg: I didn't say I couldn't find it, I said you couldn't.

Lessig: My point is that, if the lawyers continue to do what they are doing it's certain to have an effect. Just look at the market cap in these industries. What the lawyers have done is essentially establish a system where unless existing content holders approve, innovation is not allowed. That's Clay's chokepoint. But from my standpoint, this is a chokepoint to worry about because content industry is extraordinarily concentrated. In 2000 five companies controlled 84% of the music distribution in the world. This means that this is one of the most highly concentrated industries there is. And if they get a veto over new innovation that will mean they will make innovation that challenges their business model scarce.

Man: Illegal.

Lessig: Illegal. Here's the bottom line. We should worry about chokepoints where there are opportunities for old industries to kill new innovators. Those are the chokepoints we need to be worried about. And those are exactly the chokepoints that are developing now. So this story is a story about how these old industries at the physical and the content layers have succeeded in convincing the world that unless they are allowed to re-exercise their power through massive expansion of copyright or through the right to have perfect control over what happens at the physical layer, then there won't be anything like the Internet. The reality is that changes at the content and the physical layer will change the character of this end to end innovation commons that we...

Man: It all makes sense, but my question is that why couldn't it be that these measures of control on hardware and software and in networks existed for legally copyrighted material but are not enforced for non copyrighted material? There's nothing illegal about putting up something exactly like Napster for non-copyrighted music. The problem is a complicated one; it's not the technology that's causing the problem. Technology could be used to fix the problem. Bit it seems to me that both the folks at Napster and the users of it and the folks who own the copyrighted material are headed for war because neither of them wants to budge. There's actually a compromise position, which is you could make the technology aware of the copyrights and enforce the law for copyrighted material. In fact, it can be done.

Lessig: I completely agree. I have in principle no objection to the DRM. The reason to oppose the DRM now is that it will be deployed such that the existing interests, the copyright holders, get to determine what the mode of distribution will be for content in the future.

Here's a perfect parallel to the Napster story. The first Napster is the CATV industry. CATV gets born creating an antenna that is stealing other people's content and selling it to their customers. That's what they do. Twice the Supreme Court is asked to shut this "theft" down. Twice the Supreme Court has said no. So when Congress got around to regulating this, it created a very sensible balance. Congress said, copyrighters must be paid but the cable industry has a compulsory right to get access to this content. A compulsory right means that content owners could not leverage their control over content into control over this new emerging industry. If we had that in the music context right now...

All I said is they have the right to rebroadcast, but they must pay for the content they are rebroadcasting. The point is that is a compulsory right, and what that does is you can't leverage your power over content into control of next generation of technology.

The same thing could have happened with music. In fact Napster's chairman, Hank Berry, has always asked for a compulsory right. If they had done that, we could have had DRM, which would protect legitimate copyright from being stolen, but we don't worry about protecting legitimate copyright from being stolen when there is a competitive market here, which is what the compulsory right would in effect replicate. Because for distribution there would be competition and lots of different ways to distribute you wouldn't have to worry about strategic behavior by the five or four...

Man: They are trying that now. There is a proposal in Congress to suggest that the music industry must license its music for distribution by all players, but it cannot favor it's own distribution outlets. The RIAA is lobbying heavily to prevent that, because the last thing you want in a business with high fixed costs and low incremental costs is competition. And the music industry has managed to squeeze that out with things like minimum advertised price with only four players and the battle will probably fought and won or lost in the next six weeks in Congress.

Stephen Kamman: Somebody said that we need to somehow protect the ability for new industries to attack old industries and protect the right to innovate. The thing that frightens me is if you listen to the Microsoft trial, either you understood that there were possibly better ways to build operating systems and therefore Microsoft was a bad thing. Or you thought that this is the operating system, this is the way the world is, in which case you thought, "Why is the Government competing with Microsoft?"

Legally speaking, do we have laws on the books to enforce a case that says the copyright owners are restricting or using their control of distribution to preserve an economic model that technology already is...

Lessig: I've been in that can with all those worms. And I think that the court of appeals decision on the Microsoft case is brilliant. Here's its principle: The court of appeals has embraced the idea that when you have a monopoly and you exercise your power to protect yourself against a competitor who has a different way of doing something that might compete with your monopoly, you have committed something called defensive monopolization. Microsoft has been found to be a monopoly. It has been found to have violated the law. Not just one or two technical violations. The Court of Appeals unanimously said Microsoft systematically violated the law through exercising monopoly power to stifle innovations that might threaten their underlying monopoly.

The browser was one, but it wasn't the only one they were pointing to. Systematic behavior was protecting their monopoly position. That's clearly illegal now.

Man: What about the RBOCS?

Lessig: That's going to be harder now. The problem with this case is, as the Microsoft case reveals, is that okay, in principle you could bring against the RIAA. It would cost 12 million dollars. It's not like the government's ever going to do it. No government would ever go against Hollywood.

So it's not the government's case, it has to be a private case, $12 million to do it. Whose business model includes $12 million to litigate a case that has even a 60% chance of winning? So the law is clear. It's just the gap between the law being clear and the law having an effect is huge.

Man: Something else about the RIAA that Jerry was saying earlier is that there's a big fight right now between the recording industry and the music industry. The recording industry is anybody who makes their music selling CDs and the music industry is anybody who thinks you make your money selling music. And the iron law of bureaucracies is whatever else you were put in place to do, what you are really doing every day when you go to work is trying to keep yourself alive. So the RIAA is in many ways litigating this as hard as they are because they want to stay viable. And the thing they are fighting is not whether or not people get paid for music. People get paid every day from BMI and ASCAP for music played on the radio. It's that they desperately don't want music to go from being a product to being a service.

If music is sold as a service the RIAA is doomed. Their backs are up against the wall in ways that people haven't realized because the minute objecthood leaves data, the minute we get really distributable digital data, the people who make their money selling objects are out of business. Then we get into this three ways split of you got physical infrastructure and you got how the data passes over that infrastructure. That is the thing that the RIAA is trying to stifle. It's much bigger than the music industry, it's objecthood for works of art.

Lessig: Because when it's a service it will be a competitive market and they don't win.

Tom Freeburg: I just wanted to make two observations. History teaches us that when the law comes up against technology, technology always wins in the end. And the time frame by the way is roughly on the same scale as the time frame of technology development. Whatever the Hell the outcome for the RIAA is going to be they are doomed if in fact they can't find a way to adapt to the new technology, or they can't roll it back and it's going to be in the time we care about. Within the next half decade perhaps. However, there are going to be a lot of battles. I'm reminded of something my grandfather taught me when I was a young kid, he said, "Never corner a skunk, they really get ferocious."

Lessig: So last night I resisted that and thought strongly about the resistance. It's not about technology's power to overcome some force. It's about paths. And when an industry succeeds in deflecting the path of development, that has a profound affect. Even if the technology in the end wins. Think about radio. It's a classic example of how bureaucracy was captured by commercial interests in a certain moment that deflected the development of radio for 60 years.

And it takes an extraordinary amount of power. Look at the failure of low power FM. There's no reason it should have failed. It only failed because this powerful industry succeeds in buying off Congress to stop any further...

So I agree with you if we are just talking about competitive markets and technology. You are probably right. But we live in a world where it's competitive markets and protected markets that are protected by a political system that is bought off by campaign contributions in an extraordinarily powerful way. In that world I don't buy the idea that technology in the end wins because bureaucrats always get between the winner and technology.

Man: But the question here is the technology able to escape the effects of whatever supervision or attempts to crush it are put in there, even when it remains public. The old answer is drive it underground, and I really do believe that driving it underground works. It reduces the volume to something. Criminalizing it as much as the DMCA criminalizes all these things also works. But can it be made so that it's just unstoppable?

Lessig: What do you mean by works? I'm confused. The DMCA works in stopping people from... ?

Man: The DMCA works when you have a person you are going to catch for doing something like Dimitri and you put them away. But if a million people are busy using stuff and doing stuff you can't go...

Lessig: So here's the story with content. Let's say the RIAA wins and shuts down all these rogue companies like Napster and MP3.com or buys them up and then channels them to one or two on-line distribution services. Then let that be like this for a year or two years until most people are using these one or two on-line distribution services. Then you've got the buy-in to these services that makes it really hard to five years from now for someone to come out and create some kind of competitor. Not impossible. I'm not talking about perfect control. I'm just saying that you've got inertia on your side at that point. And they have succeeded in using inertia to protect themselves against these other forms of competitors.

Isenberg: Hang on! I'm going to play my three minutes. Mark Petrovic said something that I've been thinking for a long time that goes to Tom's concern and the whole conversation on the floor right now. As the technology roars ahead, there are going to be irresistible forces to adopt that technology. And huge resistance. The ultimate scenario there is all out shooting, dying, bleeding kind of war -- real war, where you get irresistible force meeting an unmovable object and the only solution is conflict. It's happened a lot. We can't ignore that if this stuff is as big as we think it is, and if the resistance is as big as we think it is, war becomes a viable scenario that we have to admit of, and start thinking about, and maybe even start thinking about how to prevent it.

Man: The way that war is fought in this country is that lawyers...

Isenberg: That's not war. That's just legal...

Man: To his point that the first consent decree on Microsoft was a massive patent[?] change that made Microsoft even more cocky. It went from $10 billion market cap to a $500 billion market cap in that time frame. These are not small path changes. And did Microsoft change its behavior? No. Instead of saying you could sell MS-DOS DR-DOS in the same disk drive, but guess what, you still have to pay us even if the customer says they don't want Microsoft DOS.

Lessig: In the sense that Tom says you can't know what's going on underneath, let me make the same argument about the effect of law on lawyers and the industries they regulate. You can't see the effect of these rulings on how it controls how people behave within companies and in universities. When [Princeton professor] Eddie Felten, in the case you were talking about, wanted to publish his paper and was threatened by the RIAA that if he published his paper he would be violating the DMCA, there were people in Stanford and in Xerox Parc who were telling the other collaboratives on that paper that they couldn't publish the paper. Not because of the threats of the RIAA, but because they wanted to be legally conservative in this type of research. And that kind of conservatism permeates the system. So you just need one Eddie Felton and sacrifice one Russian programmer for 25 years of his life and you are going to have 100s of thousands of lawyers around you changing the behavior of ordinary people.

I'm publishing a book called The Future of Ideas. In this book I quoted Courtney Love's amazing riff on the music industry. And I quoted less than 10% of the article in the book. The lawyers from Random House contacted me and said you are going to have to cut the quote, it's too long. And I said in what sense is it too long. They said well we have a policy that is either 10% or 50 words, whichever is less, and this is more than 50 words. I said, this is fair use. They said we don't care about fair use we care about being safe and we say you have to cut the quote.

I said, you know, this is a book about the over-reaching of copyright law! [laughter]

The lawyers said, we don't give a damn about what the book is about. This is Random House and we have our policy and you can take your book someplace else or you can cut the quote. The point is that this is an absurd view of copyright law, but it's Random House's view of copyright law and what is anybody going to do. I mean, that's the underground theft that I think you need to worry about when you start ....

Isenberg: There are a bunch of questions here.

Man: Can you spend more time elaborating on what you mean? I'm confused about the point you're trying to make.

DI: You mean about the real shoot 'em up war?

Man: Yes. What is going to drive people...

Isenberg: Let me defer that. There are a lot of questions on the floor and I'd just like to hear some of them.

Eric Best: My simple question is, do you think there is really something new and fundamentally different going on that is the point of this discussion. Because I want to refer back to what Roxane was saying. As I took it last night she was talking about a circumstance in which capital is withheld because you have these low barriers to entry and you can't basically reclaim a return under these conditions, so why go there.

I'm struggling with, there either is or isn't something about information communications technology and power distributed to individual users that is actually anathema to proprietary capitalism as it's been practiced. Because monopoly positions that are required to get the return on capital investment to make it worth doing are so threatened by the dispersion of power access and ownership despite copyright that this system is broken. It is broken.

Now the question is how to you unbreak it? Because the listening to this, I got to this point: So IT impact challenges capitalism at its core. The proprietary rent-garnering advantage for capital investment return can only be achieved through some mix of monopoly and non-monopoly. We have a problem. My question is do you think there is something new and different going on, or are we just having a conversation we could have had 10, 20, 30 years ago in a different room about some different stuff.

Lessig: Under the original architecture of the network, there was something new and different about the threats to existing power. If that architecture were fixed in the way the idealists talked about it 10 years ago, it would be something new and different.

But the existing monopoly interests realized how this architecture threatened them and they have succeeded in a campaign that the content layer and the physical layer is to react to that to change the architecture to protect themselves against what threat originally produced.

Now, is that something new? Well, this is my favorite quote: as Machiavelli said, "Innovation makes enemies of all those who prospered under the old regime and only lukewarm support is forthcoming for those who would prosper under the new. Their support of the different is partly from fear and partly because they are generally incredulous never really trusting new things until they have tested them by experience."

This is our conflict. People who benefit from the new system don't really get it, they don't go to Congress, they don't buy Congress, Congress doesn't hear them. People who are threatened by the emergence of the new system, they've gone to Congress and gotten the DMCA and the Net Act and all these new regulations. They've also convinced Congress that what they need to do is "deregulate" in the context of cable and everybody to allow them to exercise once again physical-layer kind of control that will re-establish their power.

Best: So that's a 'no'.

Lessig: Yeah.

Man: I'd like to take a slightly different tack on that. The question I've been asked often is, well, if the Internet was always peer-to-peer, what's the big deal now that we just sort of restored it after this sort of five-year imbalance?

And I'd like to suggest that in many ways, computer revolutions are at least as important as social revolutions, if not more than as engineering revolutions. If the engineering revolution we've had is just restoring us to something we had three years ago, the fact that it's being extended to 100s of millions of new users is genuinely new. And that the complexity people always say that more is different. That when you have more people participating, it's a new kind of system, it's not just a difference in degree. It's a difference in kind.

The difference is that a system that has sprung ten leaks is implicitly catchable and a system that has sprung 100 million leaks is implicitly not catchable. And if the edges of the network are really as distributed and if resources can be moved around and aggregated as easily as possible around the edges, I do think that is a different kind of problem.

Lessig: I would agree and disagree. I agree if you have a thousand leaks then that's going to have the effect. But the Napster case is critically important in the following sense: the court has established the principle that Napster can be held responsible for the misbehavior of its users. And in the latest round the court when Napster came back and said we have a system to guarantee 99% compliance, the court said, we aren't interested in 99% compliance. You come back and show me that you have a system that guarantees 100% compliance with copyright laws. Can you imagine if Xerox had to face that kind of test? The point is that's the attitude that goes against Napster.

Now if every peer-to-peer system faces that legal requirement, then venture capital will not go to these systems. VC is an extremely important part of this story.

Raj Sandhu: I agree with you about changing paths because God forbid if AT&T had stifled the innovation of TCP/IP and end-to-end we would still be talking on the phone most of them time. But where is the middle ground here? Clearly both sides have to find a win-win -- this is the art of business negotiation. How do both sides win? How do you do it in some ways without. . . How, if you have your book and you get zero revenues from it, obviously you are not a happy camper, right?

Lessig: I think the critical thing is that people not think that what you are saying is that the world has to be completely free. That's not what I'm saying. When I talked about the three layers, there was one slice that was free -- the innovation commons. But on top of that free layer, there was lots of people making money in the traditional way -- through copyright and control.

Second, when all the pressure is coming from dinosaurs who are trying to protect themselves against extinction, then the rest of the world will say, "look if it's really dinosaurs we are talking about, we don't like them except in movies, so let's let the dinosaurs die and embrace a principle that induces lots of new dominant industries to be born." It's those two things that need to happen in people's minds.

The first one is the hardest. People really do intuitively believe that the idea of a commons or free resources, this is all just academic malarkey. They think that what we really need is perfect property rights and perfect protection of property rights and that's why copyright is great. That's ignorant of the history of copyright and the history of innovation. New innovations always emerge where there is space for people free to be innovative outside the control of existing organizations. Isenberg found himself thrown out, not physically, but spiritually from that organization [AT&T] when he started talking about...

Man: It's interesting to think that Guttenberg could have been shut down by the people that were handwriting all those manuscripts... the same thing with the RIAA. But that succeeded and the Xerox machine was about to go to market and

Isenberg: News didn't travel fast enough to shut [Guttenberg] down.

David Weinberger: Larry, you are the most enthusiastic pessimist I've ever heard... Sean Parker at Napster the other Sean...

Lessig: The one who was silenced...

Weinberger: He says that there are about 57 million registered users at Napster at the time that the court stuff started, and estimated that there were about 10 million registered U.S. voters in that 57 million users. So here is a very substantial voting block that had no effect -- zero. I find that very disturbing.

Lessig: Where the hell is the revolution? There was no revolution because people buy into this mentality that unless there's a system of perfect property, then its all just "theft". They don't have a historical context to say that every use of copyrighted material is not a theft just because the copyright owner doesn't approve of it. But most of the 57 million users thought okay I was getting away with something. And that's the attitude.

Man: Although Orin Hatch said he got more e-mail about Napster then he got about the Gulf War. Hatch was saying Napster is a public issue. I think another one of the reasons that people didn't then exercise is that they didn't really believe it was going to be taken away.

Man: Who were those 10 million voters? College kids. No one cares abut college kids, they don't vote.

Man: Only a third of the Napster user base.

Adina Levin: It would cost 12 million to do [more litigation]? 12 million is cheap. I mean compared to a college student or to me, 12 million is a lot of money, but how many people. Is there someone who rich enough or crazy enough to bankroll it.

Lessig: We've been trying to find an artist who is crazy enough...

David Reed: I'm continually troubled by this question that there is a collection of technologies that enable the perfection of property rights. I don't mean perfect in my ideal. I mean perfect in the sense that the air should be property rights-based because we would allocate it better.

But what I'm particularly struck by, is back in the hoary days of software copy protection, a lot of us sort of fell in love with neat technology for doing copy protection and then the pragmatists among us, Mitch Kapoor being one of them, said this is nuts, we'll get paid for most of the copies, who cares. We'll do some things that make it more likely that we will get paid.

I had conversations with Andy Grove at the time. He said, this is great. You guys are being pragmatic, business is going to be great. Now Craig Barrett in the NY Times said, not too recently, one of the two key barriers that it tells these to the future is one breaking the RBOC monopoly, which I agree with. And the other is perfecting copy protection so everybody gets paid. Philosophically, that's why Intel is into this whole hard disk protection thing because they actually believe they will make more money on processors by enabling content providers to get paid. And what seems to have been the result?

There are two explanations for this. One is Intel has now finally gotten into the content business, albeit mostly as an investment strategy. So that's why they've changed. The other one is that Intel is indirectly influenced by all the RIAAs of the world. And in some sense the RIAAs have been smart enough to realize that in your three layer thing that the strategy at the top won't work unless it's enforced by strategy at the bottom and they've just been really good friends to Intel.

Lessig: I have met with Intel about this question. I don't think Intel is as bad as this makes it sound. Intel is under extraordinary pressure because of another market, the consumer electronics market. And they are playing this game where the RIAA is able to go to the consumer electronics market and get them to agree to anything, because they don't care. So they agree to anything and then Intel is in a position where they are worried, this is a real competitor to the ability to people to use PCs to do stuff. They've got to strike a deal that doesn't absolutely make CE the winner. And also doesn't disable the ability for innovation.

I gave an end-to-end talk and Intel chief lawyer right there said I completely agree, this is exactly right and we are cornered because we want to have boxes that use lots of our chips. We don't want to play this kind of game, but we are cornered into playing it because of the CE market. If they can get a deal that really had a sensible structure I think they would buy it.

Victor Blake: My comment is only minute or so. I think it's much more complicated than that. Keep in mind that technology has reached a point where the description of how an Intel microprocessor works is software. It's PHTL. You can take the software file and build a microprocessor with it. Their intellectual property has essentially been reduced to software. They know that. There is a vested interest there just as there is everywhere else and I thought it would have been more obvious.

Man: You are just saying they want to protect their own idea?

Woman: You don't have a fair use issue on a microprocessor the same as music.

Blake: Absolutely, from a legal perspective I think they pursue the same issues. I did have one other thing to say. I'm not a lawyer, my wife is one so I hear about this all the time. I think also there is a distinction when you talk about 10 million users and why didn't they make an issue of it. There is legislative law and there are interpretations of that law by judges, which builds a history of cases and therefore case law. That's the practice under which these decisions are made. At the same time in parallel it can happen that new legislation is being conducted.

All the decisions that have shut down Napster have nothing to do with what registered voters want. Because before any of them were born, the law the judges are making decisions under and the case history of it, which has nothing to do with computers at all, has already been established. And those are the precedents by which they need to act. And those people are actually, just as in other areas of law, highly constrained in the sense of decisions they can make based on those histories. So it's not entirely un-possible that a decision could be made in a court that seems awful to us, can actually be undone later by legislative change, by those 10 million people when the grow of age and can understand all the...so it may take years for them to grow up.

[Note: Tape ran out before the session break -- David I]

(return to top)
page last modified: July 10, 2002

Copyright 2001 isen.com, inc.