Transcript: Sean Doyle, Amicas Inc., 3/30/99
Video Tape 1/2
C: I bought that open sources book, and have been reading it and thinking about, and there has been a tremendous amount of press lately with Apple and all that. But one of the things I wanted to ask you about was a more historical question, was how close you feel to this particular history, being an MIT graduate, how close you feel to MIT hacker culture. I mean it's a story the that the open source people like to tell repeatedly, especially through Eric Raymond and this history of hacker culture.
S: well I'm afraid I'm really going to disappoint you because you see, when I was MIT I really didn't do any hacking at all. I was in the economics department and I worked a little bit on Troll. Which was sort of this thing at Sloan school.
C: but you weren't part of that while you were there. That part I think I knew.
S: well I worked with troll the senior year and the year after I graduated and then I left to go down to Washington and that stuff was just running on IBM mainframes. And I sort of knew a lot about the hacker culture, but was a little bit distant from, because it, it seemed a little bit too self referential. There just wasn't a lot of interest in the arts or politics or stuff like that. It was very insular. Maybe it was just the people that I met there but it seemed a little bit exclusionary for someone like me who wasn't in computer science, who was really more interested in social things you know social justice that sort of thing that I seem to of long... that I'm not working on anymore. It was really, when I was at the Fed in Washington and they switched over the research division to a Unix network based on Sun workstations. Then all of a sudden they needed a lot of software written and I was writing a lot of statistical software things and found software more interesting, but the way I learned was through differentusenet groups and different mailing groups or different Usenet groups for Mathematica.
C: and you mentioned before the free software foundation had some free...
S: that's right that's right they had some tools, so that we build all of these GNU tools, or things layered inside of Emacs that worked really well, but you know it probably helped that I had some connection to MIT and that I sort of knew of the people there and stuff, but I felt like I never joined that community until I left.
C: do you feel like it's become more political sinse then? Becasue now, Especially Richard Stallman obviously but other parts of the story are told as if it's a very political culture one that's very opposed to especially, almost entirely with respect to problem ownership of information...
S: see when I worked at the Fed it wasn't an issue at all, 'cause anything we'd write , it was just available, it was government stuff, I mean I think the only things that were hidden were some of the money models or things like that there were used for predicting exchange rates and interest rates. But the fact it was free and can be shared with academics only added to its value and definitely made it worth while. You know it really wasn't I guess I'd didn't really think about the politics I like the politics, I like the sort of thing about sharing information I find most patent things offensive even though right now were doing one or a few of them for Amicas. I think it's less to like actually shove anyone out of the market, more to become notorious,to signal that we care about a stuff but I don't see how any of the stuff is enforceable or that we have any intention of enforcing it. You know it would be a much more interesting thing if we could just go completely open source on our stuff and just share it. So when you say "the politics of it" it's like it's there is a sense in which the politics a few years ago seemed to be very restrictive that we could look at, you know, I could know that there was a program out there that would do something very useful to us, and I couldn't really look at a because, you know, if you sort of like, it's one of these hard things, if I just copy the program it's illegal if it was I just didn't understand something happened or what the function calls were or how you call the system or stuff then I think it's a learning exercise but then you have to go back and write the program from scratch you know it has been restrictive and the politics of it had been I found a little bit mixed.
C: and there's clearly a large split in the community, the community such as it is, right now especially over the licensing issues.
S: well there's the GPL and then there's the LGPL and the LGPL seems to be fine for what we do. The GPL, actually if we were going to open source some of our stuff, since it sort of forces everyone else, since no one else can build commercial product on it, you know that might be the better way to go. You know it's not like a poison pill defense but it is this thing where once you put out there you can't stuff it back in the bottle, and if different people started contributing to it there wouldn't be a way where Agfa or one of these companies could take the fruits of it and just repackage it, they would have to submit their stuff back out to the community. Since most of these systems are pretty badly written and bloated stuff like that it would probably be a good thing to shine some sunlight on them. There was this other rumor, it was in a document I saw from Microsoft and I haven't seen anything on the IBM Website to verify this, that said that DB2 for Linux was going to be free.
C: That's different from the Jikes thing?
S: yes DB2 is just their industrial strength database which right now is one of the big component costs something like Amicas, so if the database could be free, you know, it's, you really could make a system that people could download and install pretty much for nothing. And there are some other libraries that we have to purchase from people so to do something like that you know it would be a lot of work but, but it would be an interesting business model because it would say that the value really is in the service that we provide, the branding of it, the testing of it, the integration with different components it's not in the artifact itself and I find that appealing as a vision, whether it's appealing as a business model I don't know.
C: Can I ask you specifically then why Amicas or you and Adrian chose to go with Windows and Java as opposed to as opposed to trying to run it all on Linux or Unix or whatever?
S: Oh, It's called marketing, partially. Also Linux, when we started three years ago, wasn't as big as it is now. So and we didn't go with Unix because as far as we could tell from talking with people at hospitals, everyone hates Unix. They feel differently about Linux but there seems to be this distinction that if you get like Solaris I mean you have to keep sending people to schools, to Solaris schools or to the other brands of Unix schools. Those people are expensive. The utilities that run on them are inaccessible to normal human beings even though there is lots of problems with the NT kernel, the stability of the NT. A lot of IS departments are standardizing on it because you know they can run Word on it, there are tools that they these guys already know how to run and that gives the perception of a lot more control. The price is also pretty good because you can run on top of this dirt cheap hardware and you know it's eight hundred dollars for the license. I think with Linux what's happening is that people are starting to realize now that stability does matter a lot and tools are getting better and the growth curve seems to be there so that's changed. If we were making the decision now, we would probably make it the same way we did it then, which is, right now it's NT and were looking at Linux for the future, if you had asked me a year ago I would have said that I was very interested in Java OS which is the operating system written in Java, which is now sort of released and it's a product but I don't hear about it outside of the nerd community, so, and if your trying to sell to the IS community than you want them to buy something they think they know how maintain. Right now the reputation of Linux, which is mostly how I know it, is that it is rock solid and scales pretty well. Microsoft claims to the contrary that their thing scales but we know that that's false. Though for something, a system like Amicas the actual load on the system isn't as much as you get with one of these big ecommerce sites, so I think the scaling issue in that dimension doesn't really matter, both systems are fine. We have had some discussions with Sun where they've want us to port the system to Solaris, and we told them "oh we'd love to do that but we want customer who says the that's what they want," and so far we haven't found one. You know we found people who said "I would feel more comfortable if this were on Unix," and we say "this this make a difference between you buying the product and not?" And everyone so far has said no.
C: and you found that most hospitals don't use Unix they use...
S: the use Unix for their legacy systems and they're migrating everything to NT, because they're probably using NT for a lot of their other like departmental servers, and if they can use the same people and not send them off to Unix school then they have a big cost savings there. I think the entry costs for training someone thats like useful in Unix vs. useful in NT it's much lower entry cost for NT right now. You could argue about the cost of the whole organization, but right now I don't think they look at that. They look at it and they say, gee I can take Joe average programmer or Joe average someone and I can make them into an NT domain person and they can manage passwords, they can create new accounts, stuff like that. And in Unix, it just seems more mysterious, I don't how true that is, but when I used to use Unix a lot of years ago it certainly was the case that there was a lot more security. So I don't know what's going to be different there with Linux, except I think that if Linux is going after that market and Red Hat wants to own it, they know what they've got to do. And it probably wouldn't be that hard to do it better than NT, it's an awful lot of work, but there is such a rigidity in NT and a disregard for third party vendors of software, you know when Windows 2000 comes out, which is basically NT on the inside, they're already sort of warning people: oh you have to build to the latest APIs, these things are gonna change, your software is gonna be incompatible. I'm hoping what happens there, is that people look the cost of you know getting a new version of word, getting a new version of excel, new version of all the database stuff, and compare that with the cost of having to upgrade all their stuff for, to change over to Linux.
C: and that there might actually be breaking point there, that seems quite possible.
S: It's also probably the case, from everything I've heard, that Windows 2000 is going over time-budget, it's probably buggier than they would've wanted. And so if you really want to compare stability and stuff, there'll be people who want all these new features clustering, stuff like that, but they're going to come at a price of instability. Whereas the Linux stuff, it seems like they care a lot about clustering, you know NASA is doing a lot of work with Beowolf, and it looks compelling.
C: now with the Linux tools that are part of the open source discussions, they are mostly infrastructure, things that there seems to be a large enough community of developers that want to work on it, that wanna improve it, that wanna participate in this, would you find that in healthcare? Would you find community of developers in healthcare?
S: yes. I think that you would. You know particularly at university hospitals, look at the UC Davis DICOM library that we used to based our stuff on, MSU took up to work on. There was an ad on the comp.protocols.DICOM, UCLA is looking for DICOM programmers. Malinkrodt, you know they built a toolkit. In some ways, it doesn't have to be that many people to sort of like set the standard. Duke has done a lottastuff with HL7. In some ways the closer that you move it also to standard protocols and standard tools the more crossover there's going to be from the other parts of the computer science department, and from universities that are affiliated those medical schools. So then you could have like interesting term projects, how you enhance this or how you build that, and I think that's the engine by which a lot of these things happen. Also there's a lot of small companies that would probably benefit by sharing stuff back and forth because the market right now is sort of big enough that by sharing stufyou don't necessarily have to compete so directly. Particularly if it becomes a more service oriented, if the industry over the next few years becomes more, if the services are shared back and forth rather than the actual version of the software that you're running or the actual server that you're running on. So, you know in some sense as the components start going open source or becoming free, it seems like all the application software is going to go largely, I mean this is a guess, because I haven't really seen this happen but, it might go the same way. If you're charging for value rather than for how much it costs you to put it together, then people are going to charge either for the expertise of building the network, or they are going to set up some sort of micro-payment system where you get charged per transaction.
C: yes that seems like the way you guys want go with at the level of the radiology study. And you think that's gonna be more likely?
S: well, you have to keep an open mind, but its hard to tell what can happen. I think it would make us more attractive if, not only the IS department but sort of the central administration of the hospital knew how they could use this to monitor costs and, for some things, like if it's a cost per study going through being compressed, that takes CPU power, but serving a study over the web is really like pretty much zero cost on the server so I don't see any particular reason why you would want to meter it, except for security or or charge for it, but there may be models that the hospital wants to impose on it, for that each time referring physician looks at that information they get charged for it. But I tend to think that wouldn't happen, cause I think they perceive that there's this value in the synergy there and so some mistakes might be lower, if people can look at the images. But you never know what people are gonna want to do.
C: well I have a really difficult time parsing this because I'm trying to think about, economics has traditionally divided the world into products and services or raw materials, products and services, and healthcare to the extent that it's been included in economic models has has been a service, then know you've got this new layer of services on top of, services to services that makes thinking about the problem in political-economic terms very recursive.
S: well, maybe it's recursive or maybe it's sort of like what happened to the phone system you know, Scott McNeeley has this jargon phrase he calls 'web-tone' that's post we like dial tone, where you have your Jini devices, you got all these other ways of these different entities on the network finding each other and authenticating each other and stuff like that. And in the same way that hospitals outsource their electrical power generation, for the most part, they may have emergency backup, to their phone system, maybe their heating, maybe not. Maybe they'll put outside a lot of their information service stuff as well, because there's nothing particular about medical data that makes it store any differently, well there are different security requirements, you've got different security requirements, you've got all sorts of authentication things, and that you want to get stuff back the way when it went in, but there's nothing really magic here.
C: And the thing that gets me sometimes is that people often say that healthcare lags behind other industries in this respect, in terms of computerization. If you think about this way I don't believe it because this is happening to other industries at the same time, because the internet makes it a specific change that is different than the computerization of the financial services industry in the '70s or Sabre for airlines.
S: I actually have no idea what that means, that its lagging behind. Because certainly in many ways in terms of image processing or some very high-tech things is probably way ahead. And if that means what's on people's desktops, ever since 19- maybe 84 or so when I arrived at MGH there's been a computer on pretty much everyone's desktop. Computers have gotten better, the computers there, were maybe they were inferior to the ones of the Fed maybe not, Fed had some Unix workstations but then they had a whole bunch of people with dumb terminals, and it seems like most hospitals that I have been to, seem to have a serious commitment to you know networks and automation etc.
C: yeah, I think that particular refrain, the medical computing lag, is a cipher for the organizational structure of healthcare that, the changes that come with turning healthcare into a market-driven system that requires a split between management and the professional services that constitute healthcare, so you can then make financial decisions about whether two hospitals can come together or buy other hospitals, or whatever, first appears, before that you don't have that you have doctors making managerial decisions and you got hospitals, in boxes, and they very rarely merge, so when people say that medical computing lags behind because they don't understand that, travel agents were always hooked together, the stock market was always something that was linked up together.
S: yeah but I don't know about a lot of manufacturing industries, people who work on manufacturing, they might have computer consoles there but it's a different sort of animal, you know farms probably lag really for behind, I won't talk about amusement parks or you know. In some sense, hospitals are very complicated little communities and so to say that this thing as a whole is lagging behind the make much sense. There's one other thing doesn't sort of the the utopian hope behind all of this, as the cost of all the computer stuff goes down, and the bandwidth stuff costs go down, then been maybe less of the per transaction charges or costs, the friction, of the system that maybe other values could dominate other than just monetary ones, you know, that you might be able to more flexibly make decisions based on people's medical needs, because the cost for that might actually be a lot lower. If you've already decided that you need a second opinion on something, do you go to the guy down the street or you go to someone on the other side the planet.
C: there's a cynical version of me that says once you separate those two decision-making capacities, people who make management decisions are going to make financial decisions never medical decisions, and then doctors are still going to the...
S: right, but if it turns not to be a financial decision much anymore, then maybe autonomy, I mean this is clearly idealistic and, but to the degree that you can make things non-monetary transactions or that they are equal in some sense, maybe the to central administration who won't care what network it's in. You know I'm sure there's always they wanna sell the information to drug companies but maybe they could have, if they build in the anonymity stuff, maybe they could have their data and eat it too.
C: [much laughter] that's funny, on the open source web site says also "yes, you can eat open source." There's a question that relates both to open source and healthcare in different ways, which is about this problem of sharing the risk of making decisions. In the healthcare environment, it's usually focused around order entry, and a relationship between the financial decisions that are made with respect to cost-cutting or cause control there, and the medical decisions that need to be and, when I was at HIMMS that was one of the questions that I asked people, like Healtheon or IBM if I could in get to that point, was, "is there any notion in the business community of the importance of sharing that risk?", understanding that, if the doctor can't make decision or is constrained in some way, then the organization that controls those constraints should share some of the risk. And I think uniformly, this question was something people didn't understand, and people didn't seem to think there was a problem there. For one thing they can clearly separate between clinical and business decisions even if they don't understand at the ground level what the differences is or if there is the difference there. But another things I don't think anybody really wants to take the first step there.
S: you know, I think you really need to ask, like the doctors' union, like the AMA or something, what they think, because I think they get stuck with the responsibility, when you get your M.D. it means you are liable. So, if they can't do their job because of over-constrained stuff, which is the story the gets fed to the media and TV shows all the time, you know "human tragedy!", you know "I had to treat these bullet wounds, clean them out with only a Q-tip, because of a computer error, or they only ordered children's bandages with Flintstones on them, rather than these big things for AK-47 holes." You know all those things, it's just very hard to figure out what's apocryphal what's scare, even if things like that happen are they representative? Or did equally stupid things happen in other regimes?
C: I wouldn't even know how do research on that. It's a question that's come up repeatedly to me, I wouldn't know how do that without, the closest I've come is thinking about inventing a series of people, a series of imaginary people, to send through the healthcare system, to find a what decisions are being made about them, when they go from admissions, to the nurse, to the doctor, and if the doctor can't do this, but that's a big test. There is part of the problem of getting access to the site, and then if you actually tell enough people what your doing that you have to stage a situation, you never gonna have something...
S: you know there are these different accreditation agencies, and they probably have to care about this to some degree and you might wanna look at how they measure success and failure for an organization, cause I'm not sure what exactly happens if you lose your accreditation but I'm sure your insurance rates go up or there's procedures you might not be able to do or not get reimbursed for, which is the same is not being able to do them.
C: yes I'm sure there must be several...
S: it was a very big deal in the MGH in-house newsletter, like two years ago, something like JC...
C: JCAHO?
S: Exactly. That they were coming then that, and it was important not just to be polite to them, but to be polite to everyone, because you know, the standards are there for a good reason, and you adhere to them, not because your being watched, but because they're good practice. That's the spin gave on it, and I think that's the right spirit to take it too, though I'm sure they behave that way because they're being watched.
C: the other side that that I was imagining a question about with respect open source is the, with people running something entirely on Linux or using the tools that are open source, it seems to me that there might be question of liability there, in the sense that nobody in particular has to... but its not the case that Microsoft is probably gonna take responsibility either.
S: yes. But I think actually that's really the only thing you can charge for at a certain point, someone integrated these components tested them, got it through FDA...
C: so you think that's, that's sort of a question I'm getting at, then the service therefore becomes a sort of assumption of liability,
S: right, it's the branding, and componentization, then maybe making all the tests that were done on it open so that people can review them, and you might say if you use this in a way that hasn't been tested then you're at your own risk, but you could elaborate all the scenarios there. There's a part of me that thinks that's actually one of the most valuable things that could happen, because suppose you have different groups that were working in this open source way, or even one that was just working in this open source way, you know they can't handle the scenario, then you could document that it couldn't be handled, so was one of the things that wasn't prescribed use and you could fix it, that would all be in the open. Whereas right now, in a conversation with a customer today for instance, "when this fails because this information, you know, you've got an emergency patient, and they come in and we give them the name John Doe, then how do we fix the information later?" And with a lot of it, it' s just like "beats us," you've got to go manually to the places that sent the DICOM images and then you've got to delete it manually and then we resend, or we can send it with new identification information, but then there's a possibility that someone will try to read the study twice. Because one time it' s called John Doe, and then it has the correct name, because there's nothing in DICOM that says you can manage something remotely or delete an image that you just sent, and their notion of identity is so messed up that you can't say "oh I'm sending this, but this one paves over the earlier one, it should throw an error if it has a different patient name, but you might wanna have it be the earlier one be the error or the new one's an error.
C: or "there is another file with this number, do you want to replace it?"
S: Right, but these are important decisions, because you don't want to throw away original data because then you have to expose the patient to radiation if you wanna get that image again. I think a lot of hospitals have rules that say that you can't just have the report, you've got to have the underlying data around in order to do billing.
C: So you got to that from liability and Linux...
S: how tdid I get there? Oh, oh... he if you can make not only the code open, but a lot of it is sort of the use scenarios, because that's sort of the next layer above this, a lot of the Linux stuff is really just plumbing, how you put together networks, web based stuff, but there's been very few applications, this is showing my ignorance, in Linux which are like end-user applications, there's Gimp which is the Photoshop equivalent. But as you get these more socially complex cases people collaborating in medical environments were different people have to sign the name and they are legally liable, the fact that there are scenarios and known failure modes might actually be a good thing, where we had to tell this customer today, "can't you just resend the data and have it pave over," and thing is well, I don't know if I sent to a Philips or an EMED workstation. It's not defined in the DICOM standard what it's suppose to do. So how would I know, and even if we could experimentally find out, we can't rely on it, because that might change in the next rev. So to the degree that you have these systems that have their own little internal models of the universe, which can be quite ugly and wrong, and a glue system like this thing that Amicas is, that takes different components and and puts them together, the degree to which that gets, what things you decide to open up for a site customization, it's hard to figure that out ahead of time, but it might be easier if there were a lot of people working on the problem at the same time. So then the liability, I'm not sure how exactly this would be figured out, I suppose you'd have to throw some very good lawyers at it, but that we could certify worked in the way expected through a series of test cases that we had.
C: and that would be, the product that you would sell, or the service, or the solution [laughter, okay, maybe just a smile]...
S: the solution right.
C: I think that's a very interesting change, that you can basically pay for a reduction of risk, there's obviously expertise going on and all kinds of productive work, but what you're really paying for, what the competition on price is about, is the reduction of risk.
S: Right. Right. It's reduction of risk and then it's also, there's a lot of unnecessary rigidity built into almost all of these large scale workflow network systems. I don't know if I ever told to about, in the old R*Star system we had, the pushbutton interface we had, it was like an automatic teller machine, where you could push on these, you had these 2K monitors side-by-side and you had these pressure buttons and we went into this one hospital, I can remember where it was, St. Paul or something like that, there was this one set of customers that wanted this change, and 3M actually had had this, I don't think it was an anthropologist, but someone there watching how they actually used it, and the thing they were very surprised by, was it turned out that the people that were using it weren't the physicians by and large, it was the nursing stuff. Because it turns out that the nursing stuff often reads in ICU's, looks at the x-rays to see if tubes have been removed or stuff like that, and you don't really need to be radiologists to know that. The doctors tend to be more disease based, and they would go between the different ICUs, where the nurses tended to be more geographically based, so they have a certain floor. So so the navigation for like having each of them go through their work list was different, so it was just like the top-level menus you have that you'd wanna change. And they just wanted there to be two modes where you can say, hit one button if you're a doctor or a nurse, and 3M didn't want to it because you know because nurses don't make purchasing decisions it's the doctors that do. But I looked at that and I thought, if there were some hospital administrators that were in the loop and they knew that there was a different possibility here, then there's at least a possibility of feedback, and with a lot of complex social locking arrangements that you have for something like healthcare, you wanna get the most collaboration you can with the people involved. And there's also, i think, a possibility for hospitals sharing information, maybe the way, because if a were all open, they could be just like that the Fed where we made these little templates so that people could push the keys down on Emacs to fit a certain terminal and we built in all these functions that were useful to us, and which is very easy to get, it got put on some FTP site and people could download it and we sent copies to Richard Stallman, I don't know if it ever got any distribution. But the idea that you could have other groups of economists that might wanna share those same functions, you really just need one other collaborator for it sometimes to be useful, for the initial part, even if you just have one group that makes it and another group that sends in bug reports that may be useful too.
C: but in your case, and it sounds like in the case of 3M, I mean you achieve that Amicas only because, both you in Adrian having been in the healthcare environment for...
S: that's right, by I worry a lot about that right now we are too distant from that link, I mean I am spending all my time nerding away and I'm not spending my time talking with doctors about design. Now we think we have a pretty good idea of what it is that they want, but the next level that the system's going to in terms of how it works with the whole organization, I don't think anyone knows what that means yet and I think that means a whole lot of stuff on site working with people. To the degree that you could have not only multiple hospitals collaborating with us but if they could collaborate with each other, they could take the form of some sort of user group, that would be the most passive thing for them, where they would just sort of try to make their requests much more coherent, and decide among themselves what were the priorities of the things, but if there were intelligent ways to open up customization of the system, then we could work with people in a different way. It is gonna be hard to figure out how you do the gatekeeper function right because if there was another vendor in there the wanted to include something that basically destroyed the system as we now know it, if they wanted to contribute the code for free, at what point do we say no? so we have to figure out ahead of time what were the important attributes that had value for our customers that we were going to make a rationale about and then stick by. You know if someone wanted to from my point of view, technically destroyed system by integrating Cobol or something with it, but it turned out to be in some customer's interest, there's no reason for me to stand in the way just because I have some sort of aesthetic problem, but if there's a safety problem or a consistency problem, then maybe things need to be redesigned and you sort of, get together with that other group to draw the right interfaces.
C: there's also a problem there that, a) that things change very quickly and healthcare, but its hard to figure out what everyone...
S: the change the quickly that the lagging behind everyone else, [laughter]
C: the change quickly, but they also change because...
S: what do you mean by change quickly? Most of the changes that I look at seem to be glacial...
C: yes, well so maybe it doesn't actually change quickly on the ground, the lets say, decisions about what the structure of healthcare should look like have been changing very quickly, but none of them have been implemented at the level of, at the broad level on the ground in healthcare, but every six months there's a new commonsense consensus on the way market healthcare should be run, the organizational structure, the degree of vertical and horizontal integration of firms, all of the stuff is producing this vast amount of theorizing about...
S: yeah, but how much of that stuff matters? I mean, there was this old story and I'm sure it's told in a lot of different forms but I heard it first from Robert Solow, there was this man and his wife, and the man made all the important decisions in the family and the woman made all the trivial decisions, like the wife made the decisions about where he worked, what he ate and who his friends were, etc. whereas he decided on like the family's policy towards Israel [laughter]. You know, you're right there does seem to be this sort of fad and things turn over and there are these different models, but maybe it's just a period of flux and really nothing's changing, because there's no model that really fits very well, so it's more like the fashion industry, I mean I just don't know.
C: well I felt that so strongly at the Partners Telemedicine Center, because that was all they had was this top-level of buzzwords and models and ideas and it did have effects on the ground, but only in isolated and scattershot kinds of ways that it affected people who were supposed to be doing clinical trials in telemedicine. It does have an effect but it seems to me that its random and probably not very lasting in most cases. I doubt that Tele-dermatology has really changed the practice of dermatology at all. But there is the sense that if you put these, if Amicas becomes the way that radiology is done at MGH then you've got the problem that you changed the workflow and the effects of that they wont blow up in your face, but they will come back to you as the new way that things are done in healthcare, that's what I mean by things change quickly, it's not that technology changes everything its not organization changes everything but they go back and forth, which is for lack of a better word, dialectic.
S: and also that if you have standards organizations, not like DICOM, but like different medical associations, nurses association's, disease associations, that have annual meetings and stuff like that, to the degree that they, this isn't exactly what Amicas is now, but if you have some XML based rules for defining the space of certain decisions sorts of things or the taxonomies that different things fit into, what the names of the exceptions were called and stuff like that, you can start building up this common language, which really needs to be built up from the ground up, I don't think any attempt a top-down rationalization is gonna make it because they're always are these sort of funky side cases, and...
C: But wait, if we are talking about medicine, you don't think that language already exists?
S:It's not formalized enough to be useful for...
C: It's formalized in the sense that everybody learns the same things in medical school, but its not necessarily all written down, it's not in machine-readable code.
S: Well, its not in machine-readable code, but it may also be, even though it's a very rich language, it's one that also has its own obscurities, and the only way to make it more rational is to make that itself a goal in public process. I can't tell a doctor; "Oh don't call this part of the body after the person who found it in the 1800s," you know, I don't think that would be productive at all. But the names for certain disease processes, how things are linked together, so there's like all these different medical communities between geneticists vs. molecular geneticists there are large gaps between how those guys talk, but even a larger gap between those who are doing like are geneticists and radiologists for these other groups, it might be information to the radiologists that this person has this sort of gene, if they understood that this had something to do with bone proteins or something. Right now I don't think there's any way that any of the systems talk back and forth, you could just have this esperanto or a tower of babel thing, where everyone has to make a common language, and I think that would faile miserably...
C: well there are attempt to do the like with the national every of medicines little thesaurus, well big thesaurus...
S: right, but getting that out in a way that people can use it and modify it, I don't know to what degree that process is open, that a relatively small hospital or research place that was inventing new categories for things how big a difference they could make their, how fast or slow that taxonomy should change. I think that in some sense something like the internet is lowering the transaction costs of updating those libraries, changing them..
C: determining how much inertia is needed to get something changed...
S: that's right, but even with something like XML one of the nice things, is there is this, I don't think it's a of valid part of the standard but certainly talk about enough that its a good idea that eventually it'll happen, called XMi, XML data interchange. So I have one XML scheme over here and one over here and their both about patient data, but they're "incommensurate"...
C: Because they've been developed by two different communities,
S: that's right that's right... they are incommensurate in the Feyerabendian sense. Then you can define the third set of XML that defines the mapping between them, just like Feyerabend's stuff it doesn't say you can't talk about it, it just as they are not completely identical.
C: but wouldn't that in some sense is make it identical, how would you preserve the difference between two terms that our...
S: well, I think that something for human judgment to work out, because when your translating languages or when your translating between different theories you have to have things were some judgment is applied, but you might be able to codify some of that stuff, in terms of how the mappings are done, or you might have to have it just go in as indeterminate in one system from another. But that you can have these syntaxes where you can write down the original information and and then you can define translations between them or at least the parts that can be translated, means that the cost of interaction between these communities might be lower. And it might be in the interests of both to maintain the conceptual systems that they have because of training issues or one particular view is very good about thinking about a certain sort disease that's chronic, and one is very good thinking about when the a patient is already better something [laughter]. You know, or epidemiologically or something like that, that they don't have to have an exact correspondence...
C: but they would still be able to share data that was generated in totally different domains?
S: right. So, how those things get talked about, what sort of, cause I would think, there these different professional societies and that's one of the things, since everyone is an information worker now, that, that's one of the things they're going to have to worry about, how they define the boundaries, of what these things mean.
C: there's certainly no more common refrain in the healthcare world, than and I think both on the ground and and at the upper tier level of psycho babble, than standards.
S: right, but the standards are very tricky things, I could be wrong here, but I think the ACR has all of these diagnostic codes, but they are somehow copyrighted, there are restrictions on just using them or you're supposed to pay some sort of fee...
C: yeah well with the ICD codes, too they require, it's not an exorbitant fee, but nonetheless, but I don't think they're copyrighted...
S: I'm trying to remember, there a number of complications that I don't quite understand, in a lot of these things you might want to build something much more like copyleft, that says, this group owns it you can go around selling it as your own, and if you make any substantial changes you go to put back in the thing, which might mean that if you're selling product that is substantially different that you have this original thing and then you have some differences to it and those things are clearly separated. But it seems like there's a value in these different societies, like branding things, and saying this is the right way to think about it, or saying that when we published papers, we all go against this database, you know, and then and presumably they would have something for new experimental data, where you would just make up your own categories. And that would be fine because people would know what was inside the domain and what was different or new.
C: yeah that seems like a much more difficult problem than building the tools to deal with, even if everyone move towards XML...
S: yeah, the tools themselves, the implementation side is pretty trivial, it's like what do people really mean?
C: what do they really mean, and how you deal with the problem of when the American Dermatalogical Association insisting that this is the right way to dermatology, when European groups or South Asian or whatever others might not even be involved in that process? Or it becomes a de facto international standard.
S: in the worst-case, you have your XML doing voltage converters when you go between countries and sometimes some things work and sometimes they burnout. Presumably if it's software and everyone's done their job right, then at least you know when you've goneout of range rather than burning down the whole system. There is no way to force everything into compliance, or make everything all the same except either by having a very rich interaction where everything is growing and feeling like it's going to be out of control or that you just make it incredibly rigid where everyone pays lip service to it like they do DICOM and they do their own thing on the side. We did find some very new wrinkles in the DICOM standard this week, because Terry's new DICOM library started enforcing all of these restrictions that are stated in the DICOM protocol, and so far out of the six vendors or so that we tested, two of them failed, luckily they were small companies like Picker and Kodak, that no one has ever heard of. But you can just imagine, they are not going to change their stuff. And you know on one level they were like relatively minor so we have to relax the stringency.
C: what were they?
S: Oh, if in DICOM you have a study date and time, you got a series date and time, you've got an image date and time, then you've got an acquisition time, then you've got a transfer time, all these different times. But for all the images that share a common DICOM study instance UID you are supposed to have the same study date and time— that's a property of the study— and they don't, every one is different. Within a study, if I get 20 images or hundred images for this CT exam, they all have different times.
C: presumably that's because they've open taken series right?
S: right but that's why you have the image time parameter to capture the difference, or a series time for the time that series was started at. And whether this is Terry's reading of the protocol or not, it's definitely the way GE does it, it's the way Philips does it, you know a few others. But you know, no one ever enforces the stuff because things would just start to fail. And part of that's probably because the standard isn't that clearly written or just so humongous that it is impossible to digest. You know in some sense it might be, it could be evil in the sense that I could say oh we passed the DICOM conformance test because Agfa has this DICOM conformance tool that you can test against, but it only tested an image at a time so the thing could pass that but not work with another system. The philips MR studies also can be routed to some EMED systems, we found this out of the Boston medical center little while go because the philips added some garbage to the DICOM, which is valid in DICOM, you just put it in a private group. So these people pay like lip service to the standard and if done all the work, it's all complete, but then its like they just threw in this little curve ball that probably means if you want it to work well then you got to use it with their system. To the degree that this stuff becomes open source, or that there are better test for setting conformance standards, it's probably, in the long run it will end up lowering the transaction costs for everyone involved. And then hopefully other things can dominate the decision processes, if it really was equal that everyone in the hospital could have a Mac on their desk or Linux box or Windows box, and it was completely indifferent to the organization which one they had, because so much software inter-operated and could send the stuff back and forth, I don't think they even bother to have corporate policies on the stuff. But there is a big cost of ownership issue and... that's why they've got to standardize so,
C: I want to get you on tape talking about evil since you just mentioned it, it's one of the things that you explained to me very early on in the advanced imaging lab at MGH, because there was a server that was not sending something between two terminals in the room, and he said that that was local evil not global evil, and it occurred to me that this was an important concept.
S: oh I don't know, it's just a term I used to say that something's wrong, but usually it means something is wrong on purpose, there was agency behind it. I can remember that particular one but I think it may have been with some GE equipment where it has this default where it likes to send things in its own private format rather than in DICOM, if you give it a choice. And even if you've configured the GE device to only send things only in the DICOM protocol, and I don't know why they would done something like that it doesn't serve any backward compatibility problem, it's really just an exclusionary sort of thing. And at the time I don't think I understood, now there's like with a lot of work there's ways of only proposing certain transfer syntaxes rather than saying I'll take whatever you give me, because we were just trying to be open. And now we're significantly more restrictive in the way. So I guess there's evil like that...
C: one of the other examples that you had was when you are doing CAS, and there was something with IE3...
S: yes oh yes, there are so many things with IE3 that are completely evil, like here's one of them: in the HTTP protocol there's this thing called the user agent field where a browser announces to the server who it is, if you look at IE, it announces that it is Mozilla, which is the Netscape, not trademarked, just the code name, then it's like Mozilla, something semi-colon compatible, or something like that, there are different variations of it for different versions, why did they do this? Well because a lot of the web servers were like sending out certain code that said if it were Mozilla they would serve the stuff down, or they would send out something very simple or stupid that would look very ugly. And so what Microsoft did, and you could say this was the good side of the argument, that they wanted to get the same data feed that Netscape would get, so that they could render it and make their best shot at looking good that would cost the minimum of changes on the web. But you see I don't buy it, because it turned out that they didn't support things when they first came out, like I don't think they supported tables when they started doing it, later on their versions of Java script were so different, there was no way they were compatible, it just added a tremendous complexity. So what I thought when they first released it, I think it was with IE2, I'm having trouble remembering, it was just a way of pissing out on the internet and saying there's no law that says we have to follow these internet standards were to do we damn please, and were so big that you can't stop us. Microsoft just did something similar with XML where they are violating certain RFC's in the latest IE 5, and it's like apparently the number of these bugs are in some of them have been known to them for over year in different microsoft XML implementations, or different IE-like things, and Microsoft just doesn't care. Maybe their bug list is so huge that they can't get to the bottom of it, but I think it's much more likely that even though they are on the committees at the web consortium, they're not going to follow it because they perceive it gives them some advantage or the you're compatible with them or you're compatible with some other person or you've gotta do the work to be compatible with both that they really haven't subscribed to the standards process, or buying into it as a whole organization. So I view it as evil in that way I mean they obviously have the talent do it, they obviously have the resources to do it, they've obviously done the work, it's just that they'll have this little twitch were they won't support a certain MIME type or they'll support some things differently than.
C: which forces everybody else to either work with that were not be compatible...
S: yes apparently in the new versions of Excel and Word you can save things out in XML, but they're something about other going to encode it in some weird way or make it like templates within HTML or something very nonstandard, so they're saying they're saving in open format when, in fact they're not.
C: yes I know even with the most recent version of Microsoft Word if I generate something in HTML code and then I check the syntax it generates billions of errors, because it doesn't meet the 4.0 standard, it doesn't meet of 3.2 standard...
S: and why? A mean the standards are relatively well written, it's not like DICOM, I mean the people who did a lot of the web RFC's they spent some time on them they went through a review process they got some of the ambiguities removed, it's a good set of processes. You know, it's like what they did with the Java virtual machine, they could have implemented the JNI the Java Native interface, to be just slow and made their stuff much faster and in Sun couldn't have complained about it. But they didn't want to do it, there was just all the stuff they wanted to see what they could get away with. It's an organization that I don't think has a good public sense of itself for the size that it has, I think it's much too small-minded and greedy. And I think they would actually do much better if they were better corporate and internet standards neighbors. But they're not going to, I don't think there's a learning process in place there.
C: another question on evil, one of the things that I was thinking it might encompass, was the sense that there was not necessisarily agency behind it, but that these kinds of in compatibility issues can happen as a result of a lack of communication, which might involve agency at some level...
S: well, I think of that more as stupidity then evil [laughter]. No evil is when there is an opportunity to do something, and understanding that there is an opportunity to resources and all that, and then you do something just to spite the other person. You know I'm sure it's like in messy divorces, where you would rather sell the property at half it's value rather than have it go to the other person.
Tape 2 of 2
{4:15}
C: In a Large hospital organization, when you throw a bunch of money at a word [telemedicine] without any solid technical ideas, or solid technical ideas, just some buzzwords.
S: I think that's what a lot of telemedicine has turned out to be...
C: Yeah, but not because telemedicine per se doesn't have anything around it, but because there was just a lot of energy generated around it, a lot of heat and light...
S: Right, a lot of good things got done, but it means all things to all people and its.
C: That and it was broadsided by the internet, kind of exactly what happened with your company. You could have been a telemedicine company, but it turned out that doing what telemedicine wanted to do was the standard, the stantus quo on the internet, there's no big shakes about it, no reason to call it telemedicine, just call it business as usual, and I don't think [Partners] telemedcine ever realized that. Instead they just entered into these pathological modes of trying to figure out what telemedicine meant and they reduced it to clinical business modelling, that they would in some sense, go into constituencies in the hospital and tell them how to... create business models for how to generate revenue in Dermatology for example, I guess it sounds like a good idea...
S: This is what I was saying before, that I hope there's a space opening, that there is room for a lot of good ideas that don't necessarily make anyone any money, just like sharing of good ideas, like what you think, ideally what the scientific community is about , except that in medicine there is less of a perfect, less that things are repeatable in a certain way and maybe theres a bigger moral component in that you're suppose to do right by the patient, whereas in science you're not supposed to say anything about things you don't know, whereas in medicine that may be all you have to say is how much you don't know. Um, I'm just hoping that there is gonna be a different notion of what a community that happens over time. I was talking to someone the other day about one of the things that they were saying was very difficult about starting a company of 150 or so, was that as soon as you started talking to someone they would say, "oh excuse me, who do I bill this to, whhich project do I bill this conversation to?" Because they were so, the microeconomic theory stuff was done so well that they that's how they worked. And I don't remember my economics very well, but I think it was Oliver Williamson who said, well, why do you have firms? so that you can minimize these kinds of transaction costs, and do a lot of things informally. And it may be the case that what we think of right now as a hospital or as a firm, may change. But who knows.
C: Hopefully, if I continue doing this project then that's something I can watch. I mean it hasn't happened yet, I think hospitals still work the way they always did. But it could very well change, at least some parts of it will change. Adrian clearly has a lot of excitement about the internet as something entierly new with respect to the problem of the firm, because you really do now have all of these possibility for all these new organizational forms, that are not set in stone.
S: Right, as soon as RCN installs cable modems in Somerville, I'll probably spend the evenings at home rather than here, or maybe some days when I just want to work on certain tasks, and if we hire people in the short term, where they don't need direct supervision, then maybe some days it makes sense that I'm not here, so that I can focus and get things done. You know right now I have this split role, that I find wearing some times, where sometimes I have to talk to customers, talk to people about specifications, and other times I just have this list of problems and I need to go through them by five o clock or four oclock or before Jonathon leaves so he can make an install that Kevin or Adrian can test. And when I'm in that mode I would just, I would far prefer being alone. I mean there are times where its great to be ina room with Jack to be able to think through a problem, but its also good to be in a room where there's not a lot of noise, and no interruptions. So yeah I think the freedom to have these different forms for the different roles or the multiple personailties that I have to display. But unfortunately, cyberspace hasn't caught up in Somerville yet. [interruption]
{14:10}
C: Is it openosity or openicity?
S: Openicity, oh, I think...
C: openosity sort of has this...
S: Like pomposity? "We're more open than you are..." [laughter]
C: what about Apple's foray? You know they named it Darwin...
S: Yeah, I thought that was sort of... they 're very clever with their names. I suspect that what that have, the base source code is pretty good, but I also suspect that it's pretty lame and late.
C: I think its probably pretty good for the Macintosh developers community and
S: Yes and the people that are already there and that have power PCs and that are using the Apple servers, its probably a great thing and there's probably enough there to sustain it.
C: but you know, the thing that was surprising about to me was not there was really no discussion of the quality of the source code at all, it was all about the license, and what got reproduced was this infighting...
S: I suspect actually that the code is very very good, maybe not as good as Linux because not enough people have looked at it in the same way, but I think its based on Carnegie Mellon's Mach core, and that was a very serious effort by very smart people that worked on it for a very long time. So I suspect that the core stuff in there is probably pretty good. But in some ways the real value of Apple would be if they shared the operating system for their desktop stuff.
C: Absolutely, well they tried that for a very short period of time with respect to Hardware, by licensing the OS out...
{16:00}
S: Right, well maybe if they decide they want to be a hardware company, because the thing is, Apple can't decide if they want to be a hardware company or a software company. They tried to decide they were gonna be a software company and licence the hardware and then realized they were losing their margins. If they want to be a hardware company, why not give away the software and let other vendors build to spec or let it run other things.
C: If they did that, then there might be more interest in the Linux community for Mac as well, because they're certainly a possibility for those machines,
S: Right, but I think Linux already runs on powermacs,
C: It does, well, it doesn't run on mine, unfortunately,
S: Oh, why not?
C: Oh, because its one of the new G3s and everything is USB and firewire so there's basically no drivers written for any of them
S: yeah, right, but if you had more nerd backbone, you would just go ..
C: Yeah, yeah, just get right in there and do it..
S: Right, so it will probably happen eventually, that stuff will be out. One scary thing i've seen recently, is that some people are talking about at the end of the Microsoft trial what they want is to have Windows made open source. And what I'm worried about there is that if that happened is that you'd have this huge polluted thing just sitting out there and people would just adhere to it because it was the standard thing to do. Probably a lot of good would happen to it in that it would get fixed, but I think it would fragment the open source community, I can't imagine that micorsoft would write any kind of license that would be, in anyone else's long run interest. Whereas the spirit behind a lot of the open source stuff is that you really are trying to do something in a community. Even though you're trying to make money and you don't necessarily, the way that you poison the well, right is you put it under GPL and no one else can commercialize it, but that's not really that poisonous, it just says something about that, that stuff's gonna be out there.
C: And I think that means, although I don't think its the right word, but it is the sense of turning it into a commodity, in the sense that it becomes something where the value has to be added to it in some other way.
S: Right, or, you know I'm reminded of Ivan Illich was at some sort of conference, I can't remember if I've told you this before, and it was, someone was giving an overview of what the conference was about, they talked about how they wanted to "produce some good ideas," and he just lost it at that point and said, "Oh and you want me to consume them!? Does everything have to fit into the apartheid of production and consumtion?" Maybe you know, these people keep stressing community, and like Sun has their community license and that's the buzzword du jour, but I think there is something important there, where maybe it's like, some things are becoming more like a commodity, but because they can be shared, there are shared tools and that maybe it doesn't mean community in the traditional sense but it may move some transactions out of what we normally think of as the economic realm. You know, that it might be cheaper to like buy this thing from microsoft, but if I invest in spending the time to learn this then I can share it with other organizations there's this higher chance that we'll all have something more usable.
C: Yeah, I'd like very much to be able tothink through that, I was just listening to one of Eric Raymond's speeches on the Cathedral and the Bazaar, and near the end of it he explicitly invokes Marcel Mauss and the Gift economy, as an admittedly incorrect way of characteristizing what's going on, but there is some truth to it in that, because there is the sense that's what's going on is an obligation that is not strictly economic, it is not strictly a credit/debt relationship, it's one where the transaction that we enact produces this sort of promise, or in Mauss' terms it's this biting relationship, it holds on to something through time, it has to be repaid, it has to be reciprocated, but it doesn't have to be done immediately, and it isn't calculated in the economic sense. So there is something to that with respect to Open Source as a commodity in that you've got competition at one level, and collaboration underneath it or on top of it that allows, that kind of makes that economic realm less clear. And I think that's probably what frightens people, that might be what the businessman's reaction might be to free software is, on the surface is that it muddies these categories.
{20:30}
S: Right but I think that's going to in turn creates this opportunity, where say, the people who did the muddying are gonna get paved over a little bit, because maybe they are trying to charge for things that in some sense have no value anymore, and uh, as some always say that new has to replace old, but there is the sense in which, i think information is finally getting a value, not just in the sense that I can like buy a brute fact from you, but [laughter], right "Brute facts! Get your brute facts here! Hot dogs, Brute Facts!" But that there's like this relationship, or this learning that can happen if they are willing to share what they have, then the value of the whole thing is bigger, there's actually probably a parallel, and I don't know very much about this, but apparently in the airline industry there is a fair amount of sharing between different airlines and manufactures about failure analysis, that there was something that was, I don't know whether it was mandated or just good PR, but there is a lot of effort that at least used to go into the different failure modes— how things fractured, how things weathered— stuff like that was shared, and that might be an interesting thing to look at, because here you have these organizations where there is so much economic competition, but probably the bigger pressure was that if you started having these things crashing right and left, the value of the whole enterprise would go down. And so there are probably some interesting analyses there.
{eof}
Interview clip with Sean Doyle: Open Source
Interview clip with Sean Doyle: Open Source
Last Modified 11-Sep-99 9:32 PM ckelty@mit.edu
Go Back to the Start
