WHEN THINGS START TO THINK | Chapter 3: Digital Expression

May 15, 2003
Author:
Neil Gershenfeld
Publisher:
Henry Holt & Company (1999)

A few years ago I found myself on the stage of one of Tokyo’s grandest concert halls. I wasn’t going to perform; Yo-Yo Ma was, if I could fix his cello bow in time. The position sensor that had worked so well at MIT was no longer functioning after a trip around the world.

To a casual observer, Yo-Yo’s instrument looked like a normal cello that had lost a few pounds. The one obvious clue that something was out of the ordinary was the racks of computers and electronics behind it. These made the sound that the audience heard; the cello was really just an elaborate input device.

Sensors measured everything Yo-Yo did as he played. A thin sandwich of flexible foam near the end of the bow detected the force being applied to it, and a rotary sensor measured the angle of his wrist holding the bow. Strips under the strings recorded where they contacted the fingerboard. Polymer films measured how the bridge and top plate of the instrument vibrated. In a conventional cello these vibrations get acoustically amplified by the resonant cello body that acts like a loudspeaker; here they were electronically amplified by circuits in the solid body. The cello itself made no audible sound. A small antenna on the bridge sent a radio signal that was detected by a conducting plastic strip on the bow, determining the position of the bow along the stroke and its distance from the bridge.

Earlier that day we had spent a few hours unpacking all of the shipping crates, attaching the many cables, and booting up the computers. Everything was going fine until we tried the bow. The position sensor that I had developed to measure its position was erroneously indicating that the bow was fluctuating wildly, even though it wasn’t actually moving. Fortunately, I had packed a small laboratory of test and repair equipment. Given the amount of new technology being used in a hostile environment (a concert hall stage), something unexpected was certain to happen. Probing the signals let me narrow the problem down to the fine cable coming from the bow antenna. Opening up its protective insulation and shielding exposed an erratic connection, which I could patch around with a bit more surgery on the inner conductor. After I packed everything back up, the signals checked out okay, and I could resume breathing.

There was an expectant buzz in the full hall as we connected the computers, the cello, and Yo-Yo. The second unexpected event of the evening came as I began to make my way back to the control console where we were going to do some final system checks. I was only a few feet into the audience when I heard the unmistakable sound of Yo-Yo Ma controlling racks of electronics. The concert had begun. As some newfound friends made room to squeeze me in, I felt the way I imagined I will feel when my children leave home. The instrument had become his more than ours; he no longer needed us. As he played, the hall filled with familiar cello lines and unfamiliar sonic textures, with recognizable notes and with thick layers of interwoven phrases, all controlled by him. I sat back to enjoy the performance as a listener, no longer an inventor.

I was equally surprised and delighted to be there. This project was both an artistic and a technical experiment, asking how the old technology of a cello could be enhanced by the new technology of a computer. Yo-Yo’s real allegiance is to communicating musical expression to a listener; the cello is just the best way he’s found to do that. By introducing intelligence into the instrument we could let him control more sounds in more ways without sacrificing the centuries of experience reflected in its interface and his technique. This was an ideal way to explore the role of technology in music: there was an enormous incentive to succeed, and it would be easy to tell if we didn’t.

My contribution had begun years earlier with my love/hate relationship with bassoon reeds. As an enthusiastic amateur musician I was just good enough to be able to get hopelessly out of my depth. One of my perennial stumbling blocks was reed making. To create a reed, cane from a particular region in France is seasoned, shaped, cut, trimmed, wrapped, sealed, shaved, tested, aged, played, shaved some more, played again, trimmed a bit, on and on, until, if the reed gods smile on you, the reed finally speaks over the whole range of pitch and volume of the bassoon. Almost imperceptible changes in the thickness of the reed in particular places can have a salutary or devastating effect on the color of the sound, or the dynamics of the attacks, or the stability of the pitch, or the effort required to make a sound. The reward for all of this trouble is playing one of the most expressive of all musical instruments. The same sensitivity that makes it so difficult to put together a good reed also lets the reed respond to subtle changes in the distribution of pressure and airflow as it is played. When it works well I have trouble telling where I stop and the instrument begins.

For a few weeks, that is, until the reed becomes soggy and the whole process starts all over again. The cane softens, cracks, and wears away, reducing the reed to buzzing noodles. This is why I was ready to embrace the convenience of electronic music synthesizers when I first encountered them. I was a graduate student at Cornell working on experiments in the basement of the Physics building, crossing the street to go to the Music building to play my bassoon for therapy. I didn’t realize that I was literally following in the footsteps of Bob Moog.

A few decades earlier he had walked the same route, leaving his experiments in the Physics building to invent the modern synthesizer. He was struck by a parallel between the human ear and the transistor, which was just becoming commercially available. Both devices respond exponentially rather than linearly. This means that each time the amplitude or frequency of a sound is doubled, the perceived intensity or pitch increases by the same fixed amount. A low octave on a piano might span 100 to 200 Hz; another set of twelve keys farther up the keyboard might go from 1000 to 2000 Hz. The same thing is true of the current flowing through a transistor in response to an applied control voltage. This is what let Bob make circuits that could be played like any other musical instrument.

Moog began building modules with transistor circuits to control the frequency of oscillators, the volume of amplifiers, or the cutoff of filters. Patching together these modules resulted in great sounds that could be used in musically satisfying ways. As Bob’s musical circuits grew more and more capable, he spent less and less time in the physics lab, until his advisor got rid of him by giving him a degree. This was fortunate for the rest of the world, which was transformed by the company Bob then founded. The progeny of the Moog synthesizer can be heard in the electronic sounds in almost any popular recording.

At Cornell this legacy came to life for me when I encountered one of Bob’s early synths still in use in the basement of the Music building. It belonged to David Borden, the composer who had used Bob’s original synthesizers to start the first electronic music ensemble. David also had a useful talent for breaking anything electronic by looking at it; when one of Bob’s modules could survive the Borden test, then it was ready to be sold.

David was in the process of updating Cornell’s studio to catch up to the digital world. The studio had languished since Bob’s time as the department specialized in the performance practice of early acoustic musical instruments, leaving it with the earliest electronic ones also. As David filled the studio with the latest computerized synths, I found it liberating to come in and push a button and have the most amazing sounds emerge. Even better, the sound was there the next day, and the next month; I didn’t have to worry about it becoming soggy.

I quickly realized that was also the problem. The nuance that made my bassoon so beloved was lost in the clinical perfection of the electronic sounds. Unlike the bassoon, there was no ambiguity as to what part I did and what part it did: I pressed the button, it beeped. I began to give up on the technology and go back to my real instrument.

As I walked back and forth between the Music building and the Physics building I realized that I was doing something very silly. A bassoon is governed by the same laws of physics as the rest of the universe, including my physics experiments. It may be a mystery why we like to listen to a bassoon, but the equations that govern its behavior are not. Perhaps I could come up with a set of measurements and models that could reproduce its response, retaining what I like about the bassoon but freeing me from the Sisyphian task of reed making.

Directly comparing a bassoon and an electronic synthesizer makes clear just how impressive the bassoon really is. Even though its design is centuries old, it responds more quickly and in more complex ways to finer changes in more aspects of the musician’s actions. It’s not surprising that when playing a synthesizer one feels that something’s missing.

From across the street in the Physics department, it occurred to me that, instead of leaving the lab to make music, I could stay there and use its resources to update the synthesizer with everything that’s been learned about sensing and computing since Bob Moog’s day. As I began the preliminary analysis that precedes any physics experiment, I was startled to realize that for the purpose of making music, computers are beginning to exceed the performance of nature.

There are relevant limits on both the instrument and the player. Displacements of a cello bow that travel less than about a millimeter or happen faster than about a millisecond are not audible. This means that the data rate needed to determine the bow’s position matches that generated by a computer mouse. Even including the extra information needed to describe the bow pressure and angle, and the positions of the fingers on the strings, the amount of gestural input data generated by a player is easily handled by a PC. If we use the specifications of a CD player to estimate the data rate for the resulting sound, this is also easily handled by a PC. Finally, the effective computational speed of the cello can be found by recognizing that the stiffness and damping in the materials that it is made of restrict its response to vibrations that occur over distances on the order of millimeters, at a frequency of tens of thousands of cycles per second. A mathematical model need not keep track of anything happening smaller or faster than that. Dividing the size of the cello by these numbers results in an upper limit of billions of mathematical operations per second for a computer to model a cello in real time. This is the speed of today’s supercomputers, and will soon be that of a fast workstation.

The conclusion is that with appropriate sensors, a computer should be able to compete with a Stradivarius. By the time I appreciated this I was a Junior Fellow of the Harvard Society of Fellows. The regular formal dinners of the Society provided an ideal setting to develop this argument, but not a Stradivarius to try it out on. That came when I met a former Junior Fellow, Marvin Minsky, at one of the dinners. He called my bluff and invited me down to the other end of Cambridge to visit the Media Lab and take the experiment seriously.

Marvin introduced me to Tod Machover, a composer at the Media Lab. Starting with his days directing research at IRCAM, the pioneering music laboratory in Paris, Tod has spent many years designing and writing for smart instruments. Classically, music has had a clear division of labor. The composer puts notes on a page, the musician interprets the shorthand representation of the composer’s intent by suitable gestures, and the instrument turns those gestures into sounds. There’s nothing particularly fundamental about that arrangement. It reflects what has been the prevailing technology for distributing music—a piece of paper. Notes are a very low level for describing music. Just as computer programmers have moved from specifying the primitive instructions understood by a processor to writing in higher-level languages matched to application domains such as mathematics or bookkeeping, a more intelligent musical instrument could let a composer specify how sounds result from the player’s actions in more abstract ways than merely listing notes. The point is not to eliminate the player, it is to free the player to concentrate on the music.

Tod had been discussing this idea with Yo-Yo, who was all too aware of the limits of his cello. While he’s never found anything better for musical expression, his cello can play just one or two notes at a time, it’s hard to move quickly between notes played at opposite ends of a string, and the timbral range is limited to the sounds that a bowed string can make. Within these limits a cello is wonderfully lyrical; Yo-Yo was wondering what might lie beyond them.

As Tod, Yo-Yo, and I compared notes, we realized that we were all asking the same question from different directions: how can new digital technology build on the old mechanical technology of musical instruments without sacrificing what we all loved about traditional instruments? We decided to create a new kind of cello, to play a new kind of music.

My job was to find ways to measure everything that Yo-Yo did when he played, without interfering with his performance. Sensors determined where the bow was, how he held it, and where his fingers were on the strings. These were connected to a computer programmed by a team of Tod’s students, led by Joe Chung, to perform low-level calibration (where is the bow?), mid-level analysis (what kind of bow technique is being used?), and high-level mapping (what kind of sound should be associated with that action?). The sounds were then produced by a collection of synthesizers and signal processors under control of the computer.

Tod, who is also a cellist, wrote a piece called Begin Again Again . . . that looked like conventionally notated cello music (since that is what Yo-Yo plays), but that also specified the rules for how the computer generated sounds in response to what Yo-Yo did. Yo-Yo’s essential role in the development was to bridge between the technology and the art. He helped me understand what parts of his technique were relevant to measure, and what parts were irrelevant or would be intrusive to include, and he helped Tod create musical mappings that could use these data in ways that were artistically meaningful and that built on his technique.

While the physical interface of the cello never changed, in each section of the resulting piece the instructions the computer followed to make sounds did. In effect, Yo-Yo played a new instrument in each section, always starting from traditional practice but adding new capabilities. For example, an important part of a cellist’s technique is associated with the bow placement. Bowing near the bridge (called ponticello) makes a bright, harsh sound; bowing near the fingerboard makes a softer, sweeter sound. In one section of the piece these mappings were extended so that playing ponticello made still brighter sounds than a cello could ever reach. Another essential part of playing a cello is the trajectory of the bow before a note starts and after it ends. Not unlike a golf swing or baseball throw, the preparation and follow-through are necessary parts of the bow stroke. In addition, these gestures serve as cues that help players communicate with each other visually as well as aurally. These influences came together in a section of the piece that used the trajectory of his bow to launch short musical phrases. The location and velocity of the bow controlled the volume and tempo of a sequence of notes rather than an individual note. Yo-Yo described this as feeling exactly like ensemble playing, except that he was the ensemble. The computer would pick up on his bowing cues and respond appropriately, in turn influencing his playing.

Developing the instrument was a humbling experience. I expected to be able to use a few off-the-shelf sensors to make the measurements; what I found was that the cello is such a mature, tightly integrated system that most anything I tried was sure to either fail or make something else worse. Take a task as apparently simple as detecting the motion of the strings. The first thing that I tried was inspired by the pickup of an electric guitar, which uses a permanent magnet to induce a current in a moving string, in turn creating a magnetic field that is detected by a coil. Since a cello string is much farther from the fingerboard than a guitar string, I had to design a pickup with a much stronger magnet and more sensitive amplifier. This let me make the measurement out to the required distance, obtaining a nice electrical signal when I plucked the string. When this worked, I proudly called the others in to try out my new device.

To my horror there was no signal when the string was bowed, even though I had just seen it working. Some more testing revealed the problem: bowing made the string move from side to side, but my sensor responded only to the up-and-down motion of the string that resulted when I plucked it.

Plan B was to place between the strings and the bridge a thin polymer sheet that creates a voltage when compressed. By carefully lining the bridge with the polymer I was able to get a nice measurement of the string vibration. Pleased with this result, and ignoring recent experience, I called everyone back in to see it work. And then I watched them leave again when the signal disappeared as the cello was bowed. Here the problem was that the bowing excited a rocking mode of the bridge that eliminated most of the constraint force that the polymer strips were measuring.

A similar set of difficulties came up in following the movement of the bow. Sonar in air, sonar in the strings, optical tracking, radars, each ran afoul of some kind of subtle interaction in the cello. It was only after failing with these increasingly complicated solutions that I found the simple trick we used. I was inspired by a baton for conducting a computer that was developed by Max Mathews, a scientist from Bell Labs who was the first person to use a computer to make music. His system measured the position of the baton by the variation in a radio signal picked up in a large flat antenna shaped like a backgammon board. This was much too big to fit on a cello, but I realized that I could obtain the same kind of response from a thin strip of a material placed on the bow that was made out of poor conductor so that the signal strength it received from a small antenna on the bridge would vary as a function of the bow position.

The most interesting part of the preparations were the rehearsals at MIT where the emerging hardware and software came together with the composition and the musical mappings. Each element evolved around the constraints of the others to grow into an integrated instrument. When everything was working reasonably reliably, we packed up and headed off for the premiere performance at Tanglewood in western Massachusetts. It was hard to miss our arrival when, for the first time, Yo-Yo went on stage to try out the whole system.

A cello can only sound so loud. Beyond that, players spend a lifetime learning tricks in how they articulate notes to seem to sound louder and to better cut through an accompanying orchestra. There’s nothing profound about this aspect of technique; it’s simply what must be done to be heard with a conventional cello. But we had given Yo-Yo an unconventional cello that could play as loud as he wanted, and play he did. He unleashed such a torrent of sound that it threatened to drown out the concert going on in the main shed. This was great fun for everyone involved (except perhaps the people attending the concert next door).

A similar thing happened the first time he tried the wireless bow sensor. Instead of playing what he was supposed to be rehearsing, he went off on a tangent, making sounds by conducting with his bow instead of touching the strings at all. After a lifetime of thinking about the implications of how he moved his bow through the air, he could finally hear it explicitly.

As the rehearsals progressed we found that eliminating many other former constraints on cello practice became easy matters of system design. It’s taken for granted that the cellist, conductor, and audience should all hear the same thing, even though they’re listening for different things, and are in different acoustic environments. We found that Yo-Yo was best able to play with a sound mix in his monitor speakers that highlighted his performance cues, different from the sound mix that filled the hall. Such a split is impossible to do with an acoustic cello but was trivial with ours.

The biggest surprise for me was the strength of the critical reaction, both positive and negative, to connecting a computer and a cello. Great musicians loved what we were doing because they care about the music, not the technology. They have an unsentimental understanding of the strengths and weaknesses of their instruments, and are unwilling to use anything inferior but are equally eager to transcend all-too-familiar limitations. Beginners loved what we were doing because they don’t care about the technology either, they just want to make music. They’re open to anything that helps them do that. And in between were the people who hated the project.

They complained that technology was already intruding on too many parts of life, and here we were ruining one of the few remaining unspoiled creative domains. This reaction is based on the curious belief that a computer is technology, and a cello isn’t. In fact, musical instruments have always been improved by drawing on the newest available technologies. The volume of early pianos was limited by the energy that could be stored in the strings, which in turn was limited by the strength of the wood and metal that held them. Improvements in metallurgy made possible the casting of iron frames that could withstand tons of force, creating the modern piano, which is what enabled Beethoven to write his thunderous concertos. Popular musical instruments have continued to grow and change, but classical instruments have become frozen along with much of their repertoire. The only role for new technology is in helping people passively listen to a small group of active performers.

What our critics were really complaining about were the excesses of blindly introducing inferior new technology into successful mature designs. Given that a Stradivarius effectively computes as fast as the largest supercomputer, it’s no wonder that most computer music to date has lacked the nuance of a Strad. Just as with electronic books, rather than reject new technology out of hand it is much more challenging and interesting to ask that it work better than what it intends to replace.

Here our report card is mixed. We made a cello that let Yo-Yo do new things, but our instrument couldn’t match his Strad at what it does best. I view that as a temporary lapse; I’ll be disappointed if we can’t make a digital Stradivarius in the next few years. That’s admittedly a presumptuous expectation, needing an explanation of how we’re going to do it, and why.

Luthiers have spent centuries failing to make instruments that can match a Strad. Their frustrating inability to reproduce the lost magic of the Cremona school has led them to chase down many blind alleys trying to copy past practice. The nadir might have been an attempt to use a rumored magic ingredient in the original varnish, pig’s urine. Better results have come from trying to copy the mechanical function of a Strad instead of its exact design.

By bouncing a laser beam off of a violin as it is played, it is possible to measure tiny motions in the body. These studies have led to the unexpected discovery that modes in which the left and right sides of the instrument vibrate together sound bad, and modes in which they move asymmetrically sound good. This helps explain the presence of the tuning peg inside the instrument, as well as why the two sides of a violin are not shaved identically. Instruments built with frequent testing by such modern analytical tools have been steadily improving, not yet surpassing the great old instruments but getting closer.

Progress in duplicating a Stradivarius is now coming from an unexpected quarter. Physicists have spent millennia studying musical instruments, solely for the pleasure of understanding how they work. >From Pythagoras’s analysis of a vibrating string, to Helmholtz’s discovery in the last century of the characteristic motion of a string driven by the sticking and slipping of a bow (found with a vibrating microscope he invented by mounting an eyepiece on a tuning fork), to Nobel laureate C. V. Raman’s more recent study of the role of stiffness in the strings of Indian musical instruments, these efforts have helped apply and develop new physical theories without any expectation of practical applications. The work generally has been done on the side by people who kept their day jobs; a recent definitive text on the physics of musical instruments starts with a somewhat defiant justification of such an idle fancy. Computation is now turning this study on its head.

The world isn’t getting any faster (even though it may feel that way), but computers are. Once a computer can solve in real time the equations describing the motion of a violin, the model can replace the violin. Given a fast computer and good sensors, all of the accumulated descriptions of the physics of musical instruments become playable instruments themselves. This is why I expect to be able to create a digital Stradivarius. I’m not smarter than the people who tried before and failed; I’m just asking the question at an opportune time.

I also think that I know the secret of a Strad. When we sent the data from Yo-Yo’s sensors to almost any sound source, the result still sounded very much like Yo-Yo playing. The essence of his artistry lies in how he shapes the ten or so attributes of a note that are available to him, not in the details of the waveform of the note. Much of the value of a Strad lies not in a mysterious attribute of the sound it makes, but rather in its performance as a controller. Where a lesser instrument might drop out as a note is released or waver when a note is sustained, a Strad lets a skilled player make sharp or soft attacks, smooth or abrupt releases. This translates into effective specifications for the resolution and response rate of the interface, something we’ve already found that we can match. This is the second reason that I’m optimistic about making a digital Strad: much of the problem lies in tractable engineering questions about sensor performance.

For a computer to emulate a Stradivarius it must also store a description of it. Right now this sort of problem is solved most commonly in electronic musical instruments by recording samples of sounds and playing them back. This is how most digital pianos work. The advantage is that a short segment of sound is guaranteed to sound good, since it’s a direct replica of the original. The disadvantage is readily apparent in listening to more than a short segment. The samples can’t change, and so they can’t properly respond to the player. The result lacks the fluid expression of a traditional instrument.

An alternative to storing samples is to use a computer to directly solve a mathematical model of the instrument. This is now possible on the fastest supercomputers and will become feasible on more widely accessible machines. Aside from the demand for very powerful computers, the difficulty with this approach is that it’s not much easier than building an instrument the old-fashioned way. Even if the wood and strings are specified as software models they must still match the properties of the real wood and strings. That’s exactly the problem that’s remained unsolved over the last few centuries.

A better alternative lies between the extremes of playing back samples and solving physical models. Just as luthiers began to progress when they moved their focus from how a Stradivarius is built to how it performs, the same lesson applies to mathematical modeling. Instead of trying to copy the construction of an instrument, it’s possible to copy its function. We can put our sensors on a great instrument and record the player’s actions along with the resulting sound. After accumulating such data covering the range of what the instrument is capable of, we can apply modern data analysis techniques to come up with an efficient model that can reproduce how the instrument turns gestures into sounds. At that point we can throw away the instrument (or lovingly display it to admire the craftsmanship and history) and keep the functionally equivalent model. You can view this as sampling the physics of the instrument instead of sampling the sound. Once the model has been found, it can be played with the original sensors, or used in entirely new ways. Soon after we were first able to model violin bowing with this kind of analysis I was surprised to come into my lab and see my student Bernd Schoner playing his arm. He had put the bow position sensor into the sleeve of his shirt so that he could play a violin without needing to hold a violin.

Granting then that a digital Stradivarius may be possible in the not-too-distant future, it’s still fair to ask what the point is. The obvious reason for the effort is to make the joy of playing a Strad accessible to many more people. There are very few Strads around, and they’re regularly played by even fewer people. Even worse, the instruments that aren’t in routine use because they’re so valuable suffer from problems similar to those of an automobile that isn’t driven regularly. If we can match the performance of a Strad with easily duplicated sensors and software, then playing a great instrument doesn’t need to be restricted to a select few.

Beyond helping more people do what a small group can do now, making a digital Stradivarius helps great players do things that no one can do now. As we found in the project with Yo-Yo, traditional instruments have many limitations that just reflect constraints of their old technology. By making a virtual Stradivarius that is as good as the real thing, it’s then possible to free players from those constraints without asking them to compromise their existing technique.

Making a Strad cheaper, or better, is a worthy goal, but one that directly touches only the small fraction of the population that can play the cello. There’s a much more significant implication for everyone else. Right now you can listen to a recording of Bach’s cello suites, or you can play them yourself. It takes a lifetime to learn to do the latter well, and along the way the suites don’t sound particularly good. But when a Stradivarius merges with a PC then this divide becomes a continuum. The computer could emulate your CD player and play the suites for you, or your bowing could control just the tempo, or you could take over control of the phrasing, on up to playing them the old-fashioned way.

A wealthy executive was once given the chance to conduct the New York Philharmonic; afterward when asked how he did, a player commented, “It was fine; he pretended to conduct, and we pretended to follow him.” That’s exactly what a smart musical instrument should do: give you as much control as you want and can use, and intelligently fill in the rest.

Now this is an exciting consequence of bringing computing and music together. It used to be that many people played music, because that was the only way to hear it. When mass media came along, society split into a small number of people paid to be artistically creative and a much larger number that passively consumes their output. Reducing the effort to learn to play an instrument, and opening up the space between a PC, a CD player, and a Stradivarius, points to the possibility that far more people will be able to creatively express themselves. Improving the technology for making music can help engage instead of insulate people.

There’s a final reason why it’s worth trying to make a digital Stradivarius: it’s hard. The most serious criticism of so many demos of new interfaces between people and computers—say, the latest and greatest virtual reality environment—is that it’s hard even to say if they’re particularly good or bad. So what if they let you stack some virtual blocks? The situation is very different with a cello, where the centuries of wisdom embodied in the instrument and the player make it very easy to tell when you fail. The discipline of making an instrument good enough for Yo-Yo Ma ended up teaching us many lessons about the design of sensors and software that are much more broadly applicable to any interaction with a computer.

Constraints really are key. If you venture off without them it’s easy to get lost. One of the most depressing days I’ve ever spent was at a computer music conference full of bad engineering and bad music. Too many people there told the musicians that they were engineers, and the engineers that they were musicians, while doing a poor job at being either. Increasingly, we can make sensors for, and models of, almost anything. If this kind of interactivity is done blindly, everything will turn out sounding brown. Technologists should be the last people guiding and shaping the applications; accepting the challenge presented by a Stradivarius is one demanding way to ground the effort.

Toward the end of the project with Yo-Yo, I asked him when he would be ready to leave his cello behind and use our instrument instead. His answer was instant. Rather than speculate about the irreproducible joys of his Strad, he jumped to the practical reality of making music. He tours constantly and needs to be able to get out of a plane, open his case, and start playing. Our system took the form of several cases of hardware tended to by a small army of students, and it took a few hours to boot up and get working. It’s not until our technology can be as unobtrusive and invisible as that of the Strad that he could prefer it instead. It must always be available for use, never fail, and need no special configuration or maintenance. That imperative to make technology so good that it disappears lies behind much of the work in this book.

WHEN THINGS START TO THINK by Neil Gershenfeld. ©1998 by Neil A. Gershenfeld. Reprinted by arrangement with Henry Holt and Company, LLC.