One Laptop Per Child Talks

Jim Gettys delivered the opening keynote about the One Laptop Per Child project at Free and Open source Software Developers' European Meeting (FOSDEM 2007), a 2 day event, recognized as "The best Free and Open Source events in Europe."

Below is a transcription of Jim Gettys, "One Laptop per Child" (OLPC) keynote introducing the children's laptop from One Laptop per Child - a potent learning tool created expressly for the world's poorest children living in its most remote environments.

The full video program is on olpc.tv and a related interview is here.

---

Announcer: member of the board and also Vice President of software of the One
Laptop Per Child Project. He will of course speak about One Laptop Per Child. Jim.

[Applause].

Jim Gettys: Thank you. This is a picture of a school in rural middle of nowhere Cambodia. It's about six hours to drive there. Nicholas Negroponte built a couple of schools, a few years ago and discovered that the cost of the laptops was really expensive. The cost of the fuel for the generators for the laptops was really expensive. It was quite eye opening.

One thing I will start with is I want to try and reset peoples of view of the world here. The first problem when the kids were given laptops was in persuading the parents to let them use the laptops when they were taken home because the laptops were worth more than the houses that were going home to. Once they got past that point, the next realization of a day or two later was for most of these kids this was the first time there was any artificial light in their house.

That's the bulk of the world. Most kids go home at night to houses with no electricity and no light.

OK. This is sort of what I'm going to cover in this talk. In some sense, this being a geek audience, I'm going to try to explain to you why we designed the machine the way we did. We've got the luxury that Linux people typically, and free software people have not had of actually being able to design our own hardware from close to scratch. This has allowed a set of freedoms in design that you typically don't have has allowed us to go in directions that you can't normally do in the Microsoft ecology at least not unless you are Microsoft.

There are realities about hardware design that people don't fully appreciate. It's very different from software. There are realities of what ingredients you can actually get to make your sausage. You can only make as much sausage as you can actually get the ingredients for. It isn't like software where you can always run off a few million more copies. It just doesn't work that way.

Some parts of the recipe you can substitute and those are called second sources. But other parts, you can't. They are single source suppliers and if they stop being able to make your component you're in real trouble. If you make a lot of sausage, you can actually get some custom ingredients made. Depending on how long in advance you start depends upon how unique and new the ingredient can be.

So full custom BLSI is a three-year lead time. You can build an ASIC in as little as three or four months, if you know what you're doing. We’ve actually built a couple of application-specific ICs in the machines that we have here. This is the so-called BTest2 machine, the second Beta test build we have done.

So, in any case, what we are out to do is to educate kids or help them educate themselves. That's what we're about. We are not about free software per se. However you will see how the two go hand-in-hand later in this talk.

There is something on order of one to 1 1/2 billion the children in the developing world. This is a lot of kids. So if you do the math, how many machines per year should we be doing if we can build a machine that lasts five years? How many laptops were built last year? Does anybody know in this audience? I think it's like 65 million or 70 million, something like that. You can see that if we do this for real we should end up building a lot more laptops than everybody else put together.

This is the reality of what the school in the capital of Nigeria looks like. OK? This school is fortunate; it has electricity in one room. There are no lights in the regular classroom. And that's the capital city. If you go outside the capital of Nigeria it doesn't get better it gets worse.

Here is another picture of kids from that school, I believe.

Another thing I want you to think about is again I had breakfast with someone this morning, who had never focused on the fact that most kids only get five or six years of elementary education. So the people we are trying to serve, the kids we are trying to serve are the young kids. Not the middle schoolers. Not the high schoolers to which we have been primarily catering. We are trying to fundamentally help kids who are learning to read and getting their very basic education.

In some parts of the world however what you consider a basic education is quite interesting. I was talking to a guy by the name of Greg Mortensen, who has been building schools in very northern Pakistan. By the time those kids leave fifth-grade they have learned five languages. Quite impressive.

In any case, if we just continue the way we are going that's a risk by itself. Doing nothing is a risk. Using computers in education, even in the developing world is quite old. That's an Apple II. That's Seymour Faber there who unfortunately was badly injured a few months ago. This is I think from whenever Apple IIs were current so that would be the 1960s or something like that. So this is not a flash in the pan. The difference is we are trying to do this at scale, because we believe that there are network effects that take place when every child has a machine.

We are called One Laptop Per Child for a very good reason. We don't believe that that sharing machines or computer labs are an effective way for kids to use them. Do we share are our pencils? It's very hard to use it as your book if you are trying to share. It's hard for two people to read the same book at the same time. So we mean it when we say One Laptop Per Child. It's the basic premise of the project.

OK. So there are lots of efforts all over the world. We are likely to launch in the countries that are green here. Thailand is iffy, courtesy of a change of government. That always makes things much more entertaining. Governments are not necessarily stable.

Our premise is that kids all over the world and this is true so long as they get decent nutrition, are bright all over the world as they are here. So we think that if we can get machines into the hands of every kid that this is a really fundamental tool. This presents a huge set of challenges though as you might imagine.

Another thing is we don't believe that kids are just passive consumers of content from the content providers of the world. We believe that kids are created. For those of you that have children, almost certainly your children keep journals, at least in their early years. As you may know reading and writing are intimately related. So writing a journal is something that is fundamental. I have watched my own daughter who is now in fifth or sixth grade able to actually write English reasonably well, which was quite remarkable.

This is a program called [PAM PAM] that is under development. This particular version isn't out yet. But the intent is for kids to be able to compose their own music.

The other thing is we want kids to be able to experiment. This funny gizmo on the left, we did this for a demo with a so-called [ATEST] boards at a conference last May or thereabouts. That's a photocell on the end of a wooden spoon. The photocell costs less than a dollar. This allows you to conduct music. You can detect what the light is and figure out the rhythm of that you want to get just by detecting what the light is.

So we wanted to be very easy to plug things in. One of the unique things about this machine is the audio port can also be used just for DCM. You can use it as an A to B converter rather than just for audio in.

The network is really fundamental. We want kids to be able to share. We want kids to be able to work together. If you watch your own children with their friends you'll often see two kids in front of the machine busily teaching each other about to the particular thing they're playing with. Similarly, we want the teachers to be able to work with the children. So the network is really fundamental.

However this again, that brings a whole set of new challenges. There is no infrastructure in many parts or most parts of the world. How do we get the network out to the middle of nowhere?

So, we believe that it is just as important that the kids communicate with each other as with the Internet. Now we want them to get at the Internet in a serious way. So the intent in fact is to get Internet connectivity to all the schools.

Depending on where you are, there is either good or terrible connectivity. Thailand, for example, has 30, 000 schools. They claim to have connected of 30,000 of them. On the other hand, if you go to Nigeria, you get a very different answer.

Our point of view is that you can actually build networks on your own very cheaply if you know how. And the other observation is for a lot of things like cell phones bandwidth that isn't being used goes onto the floor. So there are interesting games you can play by governments twisting the arms of cell phone carriers to allow the bandwidth that's not being used for revenue paying phone calls to go toward Internet connectivity.

This is something in the middle of nowhere, I think it was in Pakistan, done some years ago setting up a network.

This is going across the disputed border from Pakistan into Kashmir, if I remember correctly. These kinds of radios are now very cheap. You can build them out of any old tool. This is a picture of somebody hacking away in downtown Athens it turns out - making their own network out of cheap components.

So what we're intending to do out all of this is actually ship what's called mesh networking. The individual computers are able to forward packets on behalf of nearby computers. You also notice we also have antennas pick up. That's makes a significant difference in the sensitivity of the radios.

James Cameron has been doing experiments in the outback of Australia, which is about as good as it ever gets. So it's a least one extreme of what you can do with the 211 even without directional antennas. When he's had a line of sight he's had it up to nearly 2 km, which I'm really amazed by. The more realistic thing of course is 300 or 400 m.

And then of course, you start thinking about meshes of meshes that get interconnected by various point-to-point sorts of links. And then you have to back haul however you can back to the Internet. So that's the kind of architecture that we're working on.

I'll talk a little bit more about how that's driven the hardware design here. I'm going to go into what our design trade-offs have been. This is not the typical laptop.

Obviously, we care about it being safe and you really don't want to have a 3 kg a laptop on a six year old. It's needs to be pretty light. You also have a feature that it can be smaller than usual because guess what - kids are smaller than we are. Kids like things that look like toys. It has to be low power.

Some people think, “Oh we should ship all the laptops and/or desktops that we don't use anymore here to the developing world.” Well, as I demonstrated before, the numbers don't work just in terms of quantity. And then you think about the power plants required to power these things. I mentioned the laptops are often consuming 25 W. If you ship a million laptops that's a 25 MW. Isn't this wonderful? You really need to worry about things like power and scale.

So obviously we have to make these things cheap or inexpensive, if we're going to be able to build them in any quantity.

So our big challenges are infrastructure, how do we get what kind of power, and how do we get connectivity? Obviously as Thailand has unfortunately been demonstrating instability is also a major challenge.

You have to worry about the physical environment. Kids don't get on school buses in the developing world. They walk home. We would like a machine that can survive getting rained on, at least when closed. We worry about that sort of thing.

You will notice that it has a rubber keyboard. It isn't that the keyboard is less expensive that way although that maybe true. It's that we have to worry a lot about water and dust. A conventional keyboard will not survive.

So in any case, there are other indirect costs we worry about, which is if our software is big and bloated we have a problem. I think Vista has shown how big and bloated things can get. We need to fight this tendency hard in Linux. I believe people are now beginning to take this more seriously.

But now it's down to where of the costs go in most laptops. If you try to make a really inexpensive laptop it turns out that sales and marketing would be your biggest single expense. So basically we have Nicolas Negroponte traveling the world as our President of Marketing. He's quite an amazing force of nature, shall we say.

Then there is a lot of expense that goes into supporting big bloated software. Another major cost is the display. The flat panel displays on your machines are expensive, and as you will see they also won't even work where we need to be able to have a display work.

A typical laptop today -

[Laughter].

- is often consuming 20 or more watts. Think one or two W. OK? Something a little over a year ago we got together to really go over our first major design review. We understood that building a machine that looked like it was going to take a steady state power of order of something like two to three watts was not good enough. The reason why is the reality of the first point.

A small kid cannot generate much power. We want to have a situation where the child can rely on power being available by being able to generate it themselves with some sort of a hand generator. It turns out that the crank on the machine is a bad idea. But there are several generators we made as external devices.

We really want the kids to spend most of their time actually able to read and write. So our goal has really been more like 10 to one. If the child has to generate power, we want to not to be spending most of their time in generating power for the machine and that sort of thing.

There is another observation we figured out. The mesh network demands of the wireless needs to be on preferably all the time. That's an interesting problem. If the CPU is on all the time and the CPU is taking a couple of watts, the numbers don't add up very well. And in particular, if the child is not confident there will be power when they need it then the first thing that they'll do, inventive little creatures that they are, will be to try and disable the wireless so they will have power when they need it.

We finally realized that most of the time when you look at your laptop not much is going on. The screen is being refreshed, and we are forwarding packets. And that's it. Most of the time your CPU is doing nothing. Why is it on? Tell me again why your processor is on. Why are the processors in the audience on now? How many of you are actually using your machines this instant?

I don't see anybody's hands go up. Well, yes, the wireless isn't working. Which means I actually have an audience.

[Laughter and applause].

So necessity was the mother of invention. From our point of view we want to keep the processor off as much as possible when you are in a situation where there isn't much power available and leave the wireless able to work and possibly the display on so the kids can be reading something.

So how do we solve this problem? Well, there are two solutions to this one of which, one piece of it is called the Decon chip, which we designed which allows us to leave the LCD on and turn off the processor board. So we can have the screen on and turn off the CPU, which even with a low-power CPU would still be consuming a few watts.

That's processor is something like five or seven dollars something like that by the time all is said and done. It's a small RAM chip that does the display.

We also made a deliberate decision to use a wireless chip that nobody had heard of made by Marvell. It actually has an R9 processor in it and about a quarter megabyte of RAM so it's able to forward packets even if the CPU is off. That's why we selected the Marvell chip. There is no other chip like it on the market right now.

So at the end of the day, depending on what you're doing, we think we'll be down one half to 1 1/2 watts area depending on whether you're using the screen or the wireless, that sort of thing. So we think with 22-watt hour batteries we should have a pretty long life. This is a typical number obviously.

Obviously a piece of this is how quickly can we wake up the machine? If we want the machine to appear to be alive, and yet we are suspending it right and left we are going to have to wake up the processor quickly. It turns out the Geode is not ideal. It takes about 25 ms to do the wake-up and it takes about a frame time to get to the display back from the DCON chip. So I think it will be about 50 or 70 ms. So, this is just at the edge of human perceptibility.

I did do some measurements on an iPAQ handheld and showed that from the time it came out of reset to the time that Linux was able to schedule processes took about 10 ms. on a 2 MHz strongarm. When I asked the kernel community, how long does it take to come out of suspend, it was a very fun trick question for a good part of last year. The typical answer was in the one to two second range. The reality is, it should be measurable in milliseconds.

OK. So there are other things that we will do like drag the panel slowly whenever possible. This is not particularly novel to us. Other people are doing this sort of thing. Obviously it changes the rate at which you update your screen depending on what you are doing. The screen itself is very novel, as you will see.

As I was talking about before, we are using this Marvell chip that has more detail about it. Marvell is implementing a standard in development called the A0211S. It's able to do forward packets without the processor on at all. We are just bringing this up right now in a more serious way. That's what was going on last week. We showed that it was technically possible to do, about eight or 10 months ago. And now we are making it happen for real.

We obviously care about power management in a really serious way. Most notebooks are often taking 10 seconds to wake up. I want them to wake up in less than 100 ms. I think that's really doable.

So the other thing we obviously had to worry about was cost. This is really made much more like consumer electronics. Its injection molded plastic. You need volume to afford the plastic injection molds. They are hideously expensive. They are by far the cheapest way to make a thing if you are building it in high volume. But you can't afford to do it unless you are making things in high volume.

So you have a chicken and egg problem. You must have large volume in order to get prices down. We are also getting rid of most of the distribution costs basically, what's happening is the countries will be ordering directly from Quanta, which is the ODM involved.

For those of you who don't know what in the ODM is, most of the laptops in the world are actually built by a set of companies whom you've never heard of. Quanta computer is the largest of them. One out of every three laptops in his room of almost any manufacturer you care to name it was made by Quanta. So we have the thousand pound gorilla working with us. It's really quite fun to see how that operation works.

Oops! I hit the wrong button by accident. OK. So we think the power will be way down. The 12-watt worst case is if you have all three USB ports sucking as much power as possible. The more typical case is going to be down to the low ranges for most things you're doing.

It's not a very fast processor. It's sort of what we all had five or seven years ago. It's a G0GX2. We have 128 Meg of RAM. Thankfully, open office is not something we want to inflict on an eight year old anyway.

[Laughter].

So that's not a problem. So in fact, most software just works. There is a megabyte of serial flash for the firmer, which we will get to a moment. Some USB ports and there is an SD card slot. There is wireless audio and also a camera in the machine. There is half a gigabyte of RAM and flash using David Woodhouse's JFFS file system. That does data compression. For typical programs and data it has a field about twice that size.

We also have a new touchpad. What are we are actually going to ship this or not, I don't know yet, because we are still slightly over budget and we may pull it out. I don't know. But in fact, the one I have here can either be used as a conventional touchpad or the larger area can be used with a stylus for learning how to write, strangely.

There is a VGA camera, so kids will be able to take pictures. So the idea is to be able to actually write.

OK. So, I want to speak a little bit about power systems design. This machine will take more or less anything nominally between 12 V and 24 V. You can plug it into a car battery. You don't need one of these bricks that are expensive, and that produces exactly 18.5 V because guess what - lithium ion really likes exactly the right voltage charge. We know what happens with lithium-ion if something goes wrong these days as Alan Cox demonstrated.

[Laughter].

So we are not using lithium ion. Our original intent was to use nickel metal hydride. We are also going to be using another chemistry, which has just become available. It's not lithium-ion but another lithium battery chemistry, which is similarly safe and ecologically friendly. Lithium-ion also ought to be recirculated.

We worry a lot about what the lifetime is. A notebook manufacturer sees batteries as a profit center. And so they are perfectly happy to have you buy a new battery every year because you have recharged 300 or 500 times, aren't they?

That's not adequate for us. So we worry a lot about whether it is possible to charge the battery many many times. There is a gang charger that we've just got samples of back. It has a car battery in it and you can charge 10 batteries at once. There are also solar cell panels that we will feed in it.

But the ultimate fallback for the kids is a generator. There is a crank being built. There is also a pull cord device that looks very promising if we can make it electrically robust that some former media lab folks have started a company called Potento to build. That's much more efficient than a crank. The crank only uses the little muscles in your arm. With the pull cord you get to use your shoulders or use your leg, that sort of thing. It's a much better idea.

The display itself is new and novel. It costs maybe 40% of what a panel of this size would normally cost. Most of the cost of an LCD is not the LCD. Most of the cost of what you have in front of you is in fact in the three color filters and a florescent light bulb, which is high-voltage and therefore requires an inverter that costs money and in how to drive the electronics.

What's more, a conventional display does not work in bright sunlight. Many or most kids in the developing world get taught some or all of the time out of doors. So we have to have a display that can actually be used in bright sunlight. So even if we could afford conventional flat panels, which we can't, don't work.

Mary Lou Jepsen, who is our CPO, has invented a new form of LCD. In color mode it is backlit, and we have this funny pattern you see - red green blue, red green blue and so on. In the front slit grayscale mode it is a 200 DPI display. This thing is 1200 x 900 resolution. So we have a lot more pixels on the screen than most of the laptops in this room. We want people to be able to read books nicely, that sort of thing.

In color mode, depending upon the ambient lighting, the resolution is anywhere from about 800 x 600, if there's no ambient light. Remember in this situation, you actually still have intensity of information occurring at every pixel. So the effective spatial resolution is not a simple function where you say, “Oh, I have one per pixel therefore, it works out to the following resolution.” So as the ambient light goes up the color saturation goes down, but the effective resolution of the display is going up. It is a complex situation.

However, if you do test patterns, in a fully dark room with backlight the worst-case is an 800 x 600 resolution. In a lot of other environments it's effectively somewhat higher.

Power consumption is very low. We do not use the fluorescent light bulbs. We use LEDs. They are color matched LEDs, but we don't care that the color fidelity is exact. And so we can take basically all of what the LED manufacturer does. So it costs a dollar for 12 LEDs, which are also replaceable, because part of the problem is, in places like Libya, the lifetime of the LED with the elevated temperature is expected to be something like three to five years. It may need to be replaced once during the life of the machine and that is possible to do. We can actually replace that pesky backlight should it fail.

So in the worst-case this display consumes about one W of power. Most of your displays in the audience here are taking between six and eight W of power. In grayscale mode at 25 Hz refresh, this display takes about 100 mW. So this is a fundamentally different flavor display than you are used to.

As I said, part of the trick of making our power budget is being able to turn off the CPU. Obviously we have to leave the dRAM being refreshed, but not active. So they are usually a chip between the CPU and the TFT called TCON. We added a megabyte of memory and ended up with what we call DCON. It drives up the funny pixel thing for us. It also can do some color anti-aliasing. This is part of how we get to the point where we can leave the display on and the CPU off. So we suspend the RAM.

This is the difference between what it would normally be with the DCON and the meg of RAM.

The other ASIC we built is a thing called Café. This is called sliding down the slippery slope. The geode turns out to have a flash controller that sucks. So we had to do something. Once we had to build a chip to interface to NAND ourselves the countries wanted expandable flash so we added SD and also a camera interface out of the whole thing. So that's the other chip we did.

The dual-mode display is actually usable in the sunlight. That for example is a real picture of a classroom out in sunlight. This shows conventional IBM ThinkPad in bright sunlight, and how much you get to see. And there is our display next to it.

OK. So I'll quickly run through what we did a lot of - oops. I guess we've lost some of our pictures. My regular laptop is in the shop. I've been having real trouble. This is what a production line looks like. This is the first machine built in November. I had fun watching them build the second build three or four weeks ago.

We went through a whole pile of pictures we aren't going to be able to see of different mechanical designs. A lot of the mechanical improvements will be in the next so-called B Test-3.

We worry a lot about the safety. There are no hazardous substances in this. We are trying to make something that has no sharp edges or corners in case somebody falls onto it. We are trying to make it as moisture and dust and dirt proof and smooth as we can. You will note to things like - guess what - one rope, and you have a shoulder harness for this guy. That's why there are two holes.

What the ultimate environmental difference between using conventional textbooks with all the paper, which actually is not too very ecologically friendly, versus building electronics is a very interesting question. We are having some people look into this - what the environmental differences are. Making silicon is also interesting. So I don't know what the wash is. But environmentally it's not obvious that it is any worse.

OK. I want to move onto software. We care about open source from the point of view that we want the kids to be able to learn about computing and the teachers to be able to do so. We want those who are really interested to be able to do things. We are obviously worrying about software bloat in a serious way. We are worrying a lot about systems security, because we are building a very large ecosystem. We hope to build between five and 10 million machines next year. That would make us a very big target.

The other problem is theft. We would like to be able to detect theft in a way. Sun opened up what's called Open Firmware without even a press release. So we are using that rather than a conventional BIOS. We had to use Linux’s boot loader for a while, but Open FirmWare ended up being a better bet. Linux is big enough that getting everything shoehorned into 1 MB flash was an interesting challenge.

We care a lot about having complete control. We are using Linux, BIOS and Open Firmware for this. This allows us complete control over the resume path. So we aren't running through ACPI stuff which, when we measured it, was tens of milliseconds. Even if the BIOS doesn't do other stupid things. I've noticed BIOS’s that seemed to want to sit there for 10 seconds for arbitrary reasons probably having to do with the fact that some device out on some USB might ever do.

We also want to be able to have kids see how machines really work. There is a basic set of applications being worked on. Remember that our user interface environments have been designed for adult developed country office workers, not six or seven or eight year olds who are just learning how to read. Do you think that this is a good match folks?

Do you think our conventional desktops whether they are Gnome or Vista or Macintosh are good match for young kids? I don't think so. Would we care about is enabling collaboration between kids and their teachers in a really serious way and a UI that is approachable and discoverable by the kids.

There is the concept of a zoom interface, which is the local machine, and you can see your buddies and every body who is on the local mesh. So we are bringing the facilities for collaboration and making collaborative applications really very far forward in the UI.

So again, these are the different sorts of modes. This is a mockup of journals being built right now. So here in the upper right one you are seeing sets of kids working around common activities, whether they are browsing the web or painting or stuff like that.

If your application is big slow and power-hungry, I'm afraid that you're out of luck. We often have a lot of choices for any given tool. We obviously care about small and simple. So Abiword is very nice, but it's too much for a seven-year-old. It's great for my 12-year-old by the way. She uses Abiword in preference to either Microsoft office or Open Office. For a 12-year-old Abiword is just about the right thing. But for a seven year old it's too much. So the Abiword folks are building a nice simple version for the younger kids, which is great. And again that's also a collaborative aspect for Abiword these days.

There are lots of ways in which we can use help. I have machines to give to people who are helping us. I even have a couple I will probably give away while I'm here. Some people question why we use GPK and PANGO. Well it turns out it has the best localization ability. We care about language coverage in a way that is really fundamental.

Oh, that's really too bad. The two things that I really wanted to show most I can't show. I have this wonderful logo for people who work on bloat. It has an overweight Tux, very sad with the international symbol of - you know. So this is Bloat Busters. Who are you going to call? We call on ourselves. Let's fight bloat. We want to keep Tux his svelte slightly pudgy self. Well, it was in the files that didn't - it was in the presentation so - it got lost – Macintosh’s have their problems. What can I say? Macintosh’s have their problems.

OK. So, we care a lot about end-to-end services. We think of servers as performance optimizations rather than being necessary. So for example our chat system that we are worrying about you say uses a protocol but will work even when you don't have a server accessible. Because we want kids in the village someplace away from the school to still be able to chat.

So think about that sort of thing when you were doing things. You really can't have systems management in any formal way courtesy of things like Avahi and Anycast and so on. A lot of the manual configuration can go away now.

So we have just built our second build of machines. There will be a third build later in the spring at which point we will do much larger education trials. It looks like end of July is when we should end up in a high volume production.

I want to end by saying thank you. This project without your efforts over many years would not be possible. So thank you all.

[Applause].

We might have time for maybe one question, I don't know.

Yes sir.

Man: I have a question.

Jim: We are all nerds here.

Man: You said that the screen was 1200 x 900?

Jim: Yeah.

Man: You said it had one MEG of RAM? That doesn't quite fit.

Jim: Six bits per pixel. The pixels are packed very carefully. If you do the math you can just to fit that into a megabyte.

[Laughter].

Jim: Good question. Very nerdy question. LCD’s give you in reality about six bits of grayscale. So that's the answer to how that works, barely.

Yes sir.

The question is - how do you do localization if you are using mostly images? It is indeed the case that icons are not completely transferable culturally. They are more so than text is, but yeah, I can imagine that we are going to have to build different icons for it.

Understand that we want kids to learn to read and write. We really do. It's just that the machine has to be usable by a child, who is just in the process of learning.

One last question. The gentleman there with the red shirt.

Well actually right now, they break off easily.

[Laughter].

In the next build that we are making them rubber so they will bend. [Laughs], they will bend enough to push them around on the axis. Again, that's a very good question. These were not strong enough. We have people who have been taking these machines and dropping them on marble floors –

Man: On purpose?

Jim: Yes, on purpose. Several hundred machines after they went through thermal testing, and so on and so forth ended up being tested until destroyed. So we could figure out what needed to be strengthened.

Now a few of these strengthenings are in this version. The next version, the so-called B-Test3 machines will have most of the results of that work making a more robust machine. It takes a long time to modify injection-molding molds. The lead-time on those is measured in months. You can turn new PC boards faster. And you can do a new sheet metal much faster. But new plastic? Oh, it's expensive and it takes a long time.

Thank you all.

[Applause]