Conversation with IEEE Medal of Honor Recipient Asad Madni
与 IEEE 荣誉勋章获得者 Asad Madni 的对话
Oddly enough, there is no Nobel Prize for engineering. The greatest recognition for engineers, then, is the IEEE Medal of Honor. The list of recipients is a roll call some of the most distinguished innovators in electronics history, including Robert Noyce, Leo Esoki, Lotfi Zadeh, Bob Metcalfe, Morris Chang, and this year, the IEEE Medal of Honor was bestowed upon Asad Madni.
奇怪的是，世界上没有诺贝尔工程奖。那么，对工程师最大的认可就是 IEEE 荣誉勋章。获奖者名单是电子历史上最杰出的创新者，包括 Robert Noyce、Leo Esoki、Lotfi Zadeh、Bob Metcalfe、Morris Chang，今年，IEEE 荣誉勋章被授予 Asad Madni。
We’re honored to have as our guest today the 2022 IEEE Medal of Honor recipient Asad Madni. Madni is lauded for developing the first microelectromechanical-based gyroscope and inertial measurement unit. This MEMS device has been important for keeping steady things that really should not be unstable, from automobiles that are profoundly safer because they are now so much less prone to roll over, to the Hubble telescope, which required absolute stillness to capture so many breathtaking images of the universe for so long.
我们很荣幸邀请到 2022 年 IEEE 荣誉勋章获得者 Asad Madni 作为我们的嘉宾。Madni 因开发了第一个基于微机电的陀螺仪和惯性测量单元而受到称赞。这种 MEMS器件对于保持设备稳定很重要，有些设备真的是不能处于不稳定状态，例如汽车因为更不容易翻车而变得非常安全;哈勃望远镜需要绝对静止，才能捕捉到如此多令人惊叹的宇宙图片。
【NOTE: Madni was responsible for many innovations in stabilizing sensors. The technology used in the Hubble Telescope was Madni’s, but it was different from the MEMS device used in vehicles.】
【注：Madni 负责稳定传感器方面的许多创新。哈勃望远镜使用的技术是Madni贡献的，但它与车辆中使用的 MEMS 设备不同。】
We’re pleased to have as our guest today the 2022 IEEE Medal of Honor recipient Asad Madni. His medal citation reads: “For pioneering contributions to the development and commercialization of innovative sensing and systems technologies, and for distinguished research leadership.”
我们很高兴今天请到 2022 年 IEEE 荣誉勋章获得者 Asad Madni 作为我们的嘉宾。他的奖章上写着：“对创新传感和系统技术的开发和商业化的开创性贡献，以及杰出的研究领导力。”
That’s an incredibly succinct way of putting it, yes. Madni has nearly 70 patents issued or pending, and he’s written over 150 peer-reviewed articles. He’s perhaps best known for his innovative MEMS gyroscope. It’s the heart of stability systems in many critical items, including passenger vehicles and aircraft. That innovation was only one highlight; Madni has been — and still remains — at the forefront of intelligent sensor and system design.
这是一种非常简洁的表达方式，是的。Madni拥有近70项已发布或正在申请的专利，他撰写了150多篇经过同行评审的文章。他最出名的也许是他的创新 MEMS 陀螺仪，这是包括乘用车和飞机在内的许多关键产品中稳定系统的核心。创新只是其中一个亮点，Madni一直并且仍然处于智能传感器和系统设计的最前沿。
He is currently Distinguished Adjunct Professor and Distinguished Scientist with the Electrical and Computer Engineering Department at the University of California, Los Angeles.
他目前是加州大学洛杉矶分校电气与计算机工程系的特聘兼职教授和特聘科学家。
以下是对话全文，您可以点击微信右上方三个点里的翻译功能查看中文
Brian Santo: You are quite famous for contributions to the development of stabilizing technology, particularly what’s called the gyrochip. And rather than have me try to explain it, it would probably be best if I asked you to do so.
ASAD MADNI: I’d be glad to. Let me start by saying that the advent of truly low cost, very highly producible inertial sensors, with no known wear-out, had been a goal of the industry for many years. My company realized division, the system down inertial division, had a Quartz Rate Sensor technology that we refer to as the QRS (later it was known as the microgyro and subsequently as the gyrochip) could actually satisfy this requirement.
Let me explain very simply how it works. It is based on a vibrating quartz tuning fork to sense angular velocity. By using the Coriolis effect, a rotational motion about the sensors’ longitudinal axes, produces a DC voltage proportional to the rate of rotation.
The concept of using a vibrational element to measure rotation velocity by employing the Coriolis principle has been around for decades. In fact, the idea actually developed out of the observation that a certain species of fly uses a pair of vibrating antenna to stimulate its flight. This sensing technique has been the inspiration behind the practical embodiment of the QRS.
The Coriolis force is generated (to explain it for the audience) by an object as it resists being pulled from screen of vibration. Because the quartz’s speeds electric changes in the forces generated show up as changes in electrical charge. And then these changes can be analyzed and converted into angular velocity. So that’s a simple explanation of the fundamental principle.
BRIAN SANTO: Was there a special insight or innovation or a clever manufacturing technique that allowed you to develop the specific product that you ended up with?
ASAD MADNI: That’s a good question. Actually, all the principles theoretically were in place. But the practical embodiment is where the real challenge was.
The tuning fork sensing element was made out of monocrystalline piezoelectric electric quartz, while all the electronics was in an application-specific integrated circuit silicon chip. And the reason for that was quartz simply had better stability and performance in terms of signal-to-noise ratio than silicon. Though it could be manufactured using similar process to those of the manufacturing silicon chips, etched wafer processing photos, photography, etc.
This quartz tuning fork could not be created on the same wafer as the silicon circuitry. We added an extra assembly step. But that was a small price to pay.
What made the gyrochip better than what was available at the time: A) it had no contacting parts, hence no wear-out, dramatically improving reliability; B) much smaller size; and C) much lower cost, since it could be batch processed. So it represented a totally disruptive technology for its time, and something that the industry was actually looking for. And if you’re interested, I can go into the challenges that we faced as we go along.
BRIAN SANTO: Oh, that’s all the fun! What were the challenges?
ASAD MADNI: Well, Brian, let me first say that this technology is referred to as MEMS, Micro Electro-Mechanical Systems, as you well know. And this was much more than a refinement of techniques.
While the semiconductor industry manufacturing processes of deposition, etching and masking, and so on, were used for batch processing in production, a MEMS device offers very significant challenges. For example, let’s just consider something. Even the most complex computer chip contains active and passive devices within them, such as transistors, diodes, rectifiers, capacitors, inductors, resistors, etc. MEMS by contrast contain not only the active and passive devices for the circuitry, but they actually include dynamic devices. The dynamic devices such as tuning fork are physically moving, creating an entirely new set of manufacturing challenges. And it is for this reason that the maturation of processes for MEMS devices took a long time.
Once these challenges were overcome, MEMS became one of the most ubiquitous technologies in use for numerous applications, ranging from aerospace and defense to automotive, medical, industrial, communications, and consumer electronics. Today, any piece of electronics that you have or carry in your pocket does indeed have a MEMS device.
BRIAN SANTO: So that’s actually quite a fascinating subject. Integrated circuitry was a plein air then, and the industry is going through great effort to make sure that it’s mostly plein air still. That seems like an unusual approach if you’re going to stick something that’s moving, or build something that could move, into a plein air construction. Was that part of the difficulty?
ASAD MADNI: Well, yes, that is. And so therefore, there are different processes. Today, of course, you know, technology has matured over the last 30 years. And now you have the same type of sensors in silicon, because they have reached that level of maturity. So their processes are different. They have like, you know, bulk processing and surface micromachining and deep…
BRIAN SANTO: That’s now. How about at the time?
ASAD MADNI: At the time, we were the pioneers. I mean, we were using the available application-specific integrated circuit technology that was state-of-the-art. And that was, you know, standard processors. And building separate quartz forks and implementing them on that.
One of the things that I think I should mention is, remember, this is a math-based device. So to get a signal-to-noise ratio, because you’re detecting this Coriolis force, the level of the signal that you get, the signal-to-noise is dependent on the size of the fork. The sensing element. And this is another very big challenge, because as we went through reducing the size of this tuning fork, from aerospace and defense (which is really pretty large size, inches or more), down to something that was just a few grams and a very, very small thing (smaller than your small fingernail), we had to go through some tremendous innovations in terms of physically reducing the fork size, and actually increasing its performance, which is quite contrary to normal theoretical thought process.
So we had some very outstanding innovations. And that kept us in the forefront of the technology with reference to aerospace and defense, which was the first part. And then of course we go to the automotive side.
BRIAN SANTO: Did you have patrons or potential customers lining up to back this technology? Were your aerospace and military customers already eager and prepared to use it? Basically, the question is, Who were your first customers? And how did you expand from there?
ASAD MADNI: Okay. So initially, we wanted to introduce this as a disruptive technology for our customer base, which was the aerospace and defense sector. [UNINTELLIGIBLE] International Division had its primary customers. And they loved it. When we started building this… and remember, you have to understand this was in small quantities because there’s aerospace and defense market, it’s not a human market. And it found excellent, acceptance, initially in highly classified programs and subsequently on major A&D programs. I mean, this was from UAVs, helicopters, missiles, aircrafts, you know, all the aerospace and defense sectors.
And while we were sitting fat, dumb and happy and enjoying our customers’ graciousness, the Berlin Wall came down. And just like all other aerospace and defense companies, we all had to pay the price of peace. And the A&D sector business started declining very, very rapidly. And we had to do some real soul searching.
Because here we were sitting on this tremendous technology, which nobody else had. And the market sector for which it was designed was collapsing right in front of our eyes. So we had a choice: either stick our head in the sands and let nature take its course, or do something about it. And after a lot of soul searching and intense marketing studies, we decided that there was a tremendous opportunity in the automotive market for electronic stability control and rollover prevention, where they were looking for angular rate sensors to prevent the accidents that were taking place due to electronic skid as well as rollover.
And remember, braking systems, anti-lock braking systems that they came out, everybody knew that they were doing some good for us, but they were never any theoretical data showing how many lives had been saved. Whereas as far as electronic spin out and rollovers, there was theoretical data and practical data that had been taken showing how many deaths had been caused.
So here was a market. The challenge was, we had no experience in this market. But we had a technology that could address this market. And that was a very bold and major move that I had to take, together with our chairman at BEI, to pursue this market.
BRIAN SANTO: My recollection, if I recall the timeframe correctly, was that the market for sport utility vehicles was new. And there were a couple of manufacturers with models that became notorious for rolling over. Is that the same timeframe to that? Did that secular trend help in any way?
ASAD MADNI: Yes, it did. You’re absolutely correct, Brian. Remember, these sensors that we transformed to go for electronic stability and rollover prevention. Just first, let me give you, by way of background, a single-axis angular sensor for aerospace and defense is anywhere from $1,200 to $1,800. A full inertial measurement unit that had full six degrees of freedom in three of these sensors, three accelerometers and full circuitry could go in several 1000s. And with GPS, augmented to go up to 100,000.
But our challenge here was… and the highest, by the way, the highest number of sensors we have sold for aerospace and defense was about 10,000 a year for a classified program.
Now to transfer that to the automotive market, where you’d have to start shipping very quickly ramping up from 3,000 a month, to millions a month, was a tremendous task. And to bring the cost down $100, then to $50, and then to $25 in a very short period of time. But remember, it’s part of the braking system.
And this yaw sensor (this angular sensor is called a yaw sensor), it sits towards the middle front of the car. And what it’s detecting is the lateral movement of the car. And so that becomes the behavior of the car. And with how the driver applies the brake or how the driver turns the steering wheel, you get the information like the direction in which the driver is trying to turn, how fast, how much, how much pressure is being applied due to torque, which becomes the intent of the driver. And the computer compares the intent of the driver versus the behavior of the car. And in real time takes control of the brakes and applies the brakes in such a manner that it prevents this fishtailing which then eventually leads to rollover.
So our biggest customer (we had many customers, but the biggest customer) when we went into the automotive market was Continental Teves. Continental Teves and Bosch were the two largest brake manufacturers. Bosch is developing their own silicon-based approach to the their sensor. Continental Teves teamed up with us, and they were using our sensors.
The very first cars to introduce it was actually the General Motors stability track system. It was launched on the Cadillacs and the Corvettes. And then what happened, Brian, coming to your point, which was a very, very important event.
They have these newspaper editors, such as yourself, that do research on cars and do test driving on cars. Well, one of them took (I think it was a Mercedes, if I’m not mistaken) an SUV on the Autobahn in Germany, and did what is called an elk maneuver test. An elk maneuver test is where you make very acute turns simulating the condition of an elk moving up on the Autobahn. Well, when you made this a few turn, the car rolled over. And that created a very sensational effect. I think Mercedes recalled their cars for retrofitting.
Deiss, remember, was running Volkswagen, and he decided that he wanted every single car, every single VW, to have electronic stability control and rollover prevention. And while this was going on, we had barely learned how to produce the sensors in smaller sizes. We were trying to put our automation in place, innovations in design, breakthroughs in manufacturing.
Our customer came and said, Well, from the few thousands a month, we want hundreds of thousands a month. And don’t tell us you can’t do it. Tell us what does it take to get there. So essentially, we hadn’t even learned how to walk. But we have started to run now. And so this is where the real break came and the market just exploded worldwide.
And our Chairman, Charles Croker, and I both are very active in lobbying on Capitol Hill to make sure that ESC and rollover, the ESC system especially, got to mandated for all cars, and eventually we were successful.
And I have to tell you that we had a lot of scars on our backs. We had to do several things. First of all, having come from the aerospace and defense sector, we had become victims of our own successes. And when we decided we were going to go into the aerospace and defense market, I had a long talk with our engineers. All of them had grown up with me pretty much. And they said, Well, we’ve done more complex systems than automotive can ever demand. I said, Yes, you have. But right now what we have to do is, you’ve got to develop these things with not one additional cent of cost. You have to design to cost. Because the automotive industry is totally unforgiving. There is no loyalty here. Somebody makes the thing for two cents less, you’re done for. So you have to design to cost.
And some people accepted that challenge, some did not. There was a hubris that had set up, and eventually I had to tell these people, I’ll provide you the proper training. I will send you to Switzerland to learn how to handle quartz from watch manufacturers; we will get automotive experts to talk about the quality assurance systems, which by the way, were far, far more demanding than aerospace and defense, contrary to what we had thought.
Well, some of the engineers agreed, and others (out of the hubris) were reluctant. And at that point, I had to take a very major decision and pretty much put my job on the line, which was to tell them that just the way an individual can outgrow a company, a company can outgrow an individual as well and have to let them go. I was questioned by the Board of Directors for that meeting, even though I was on the Board of Directors, being the President and Chief Operating Officer.
But after we were successful, and with a lot of scars on our backs, we became the largest supplier of yaw sensors for the automotive industry. And at that point, those that had left wanted to come back, and my question to them was, Why?
So that gives you an idea of what we went through. It was a remarkable defense conversion story. As I said, in spite of the scars on our backs, we paid the price, but it was a truly successful defense conversion story. CNN featured me on their shows from source to power shares. Everybody got really excited about this technology.
So we were the pioneers. We were the trailblazers. And now, of course, there are a lot of people in the market. But I’ll tell you this much: While achieving this process, when going through this defense conversion, I would have to tell you that this was one of the most unexpected and gratifying applications. Because to invent and commercialize technology that is pivotal in saving human lives all around the world on an ongoing basis, was truly a great personal reward for those of us who gave it all we had. And it was a great experience.
BRIAN SANTO: I’m fascinated about how that eventually got you to participation in and contributing technology to the Hubble Space Telescope. And I’m particularly interested in a line that I found intriguing and that, as a non-engineer, I can’t help but wonder about. The Hubble Space Telescope had the quartz MEMS gyrochip.
ASAD MADNI: I will give you a full explanation on that. So first of all, to answer your first question, How did we get into it? The BEI Technologies, which is listed on NASDAQ, was the company. Charles Crocker was the chairman. I was the President, Chief Operating Officer and the Chief Technical Officer. Now, under us, we had about 13 companies. The MEMS part was the Systron Donner Inertial Division. I came from the Systron Donner side. I was the Chairman, President and CEO of Systron Donner.
The major assets of Systron Donner, which is owned by the British giant EMI, was disposed of when the defense budget started to decline. And Charles Crocker of BEI Electronics and I worked together, and BEI Electronics acquired the major assets of Systron Donner, including the Systron Donner Inertial Division and three other divisions. And the joint company became BEI Technologies, Inc.
So under us, the Systron Donner Inertial Division is the one that was in the MEMS area, and the inertial navigation.
We had another company, which was strictly in position sensors. We had another company that was an industrial encoders. We had another company that was in the motors and actuator business, and so on and so forth. We had a total of 13 companies here; and one in France, two in the UK. But we had one division that was called the Precision Systems and Space Division that was based in Little Rock, Arkansas. And this division specializes in absolutely the high end, the highest resolution optical encoders.
As you know, a simple encoder is the device that… let’s say you’ve got an LED, and you’ve got these slots. And as it goes through, you count the number of pulses. And so you know how much it has been rotated and what speed it’s rotating. Absolute encoders are ones where you have multiple tracks, and so that when you shut the power off and turn it on, it comes back again. They’re usually 8, 10, 12 bits resolution, sometimes maybe even 16.
But this company specializes in extremely higher. When I tell you high end, what I’m talking about is 21, 22, 23, 24 bits of resolution. You think about that. Two to the power of 24 minus one. Okay? So they were the prime suppliers of encoders, for pointing systems for the aerospace and defense market.
Now, the Hubble Space Telescope, which is a low Earth orbit telescope, meaning it’s 150 miles away from Earth, orbiting at about 17 to 18,000 miles per hour. The way it took the pictures is, it first had to lock on to a guide star. Guide stars are bright stars called Cepheid variables. And so that becomes like a reference point. And then while it’s going at 70, 80,000 miles an hour orbiting the Earth, it scans a portion of the sky, collects those images and keeps doing that and then these images are fused. So you can understand the accuracy that is needed to be able to do that. Accuracy meaning in terms of pointing and in terms of stability.
So we developed this encoder- and motor-based system that worked in conjunction with the fine log guidance systems that the gyros are sitting, the mechanical gyros, but ours was the optical encoders and the motor-based system that moved in azimuth and elevation to keep the Hubble pointed at the guide star, so this was a star select.
Now if you’d ask me, for your audience, well what kind of accuracy does that mean? I can tell you in our gradients and so on, but most people won’t understand. So let me just say: If you take a US quarter dollar and hold it up in Los Angeles. And the Hubble Space Telescope is in, say, San Francisco. It would have to point at the center of that quarter dollar from San Francisco to Los Angeles.
BRIAN SANTO: Boggling! Boggling!
ASAD MADNI: And here comes even more difficult specification. Because of the Van Allen radiation belts, the stability of that pointing had to be such that, over a 24-hour period, the beam could not move by more than the thickness of the quarter dollar.
BRIAN SANTO: Absolutely extraordinary.
ASAD MADNI: It is. So this is the extremely slow-motion, dual-axis servo control system that we developed for the Hubble Space Telescope. In 1990, it was launched. And ’til today, it is there performing exceptionally well.
BRIAN SANTO: All the other problems with the Hubble telescope, that was one of them. If you’ll excuse the expression, “stable” elements.
ASAD MADNI: It certainly was. The problems with the Hubble, which I think was one of mankind’s greatest achievements, was the misalignment of the mirror. And when the whole world was pointing fingers at the United States, we sent our crew that actually went in there and in front of the world fixed everything. And we started getting the images back. And our pointing system played a major role in that, in providing us with these remarkable images that have enhanced our understanding of the universe.
BRIAN SANTO: Do you have a favorite image from Hubble?
ASAD MADNI: Indeed I do. It is those two tall pillars. I forget what they’re called now. You remember the two really tall pillars? They look something like bananas.
BRIAN SANTO: Yes!
ASAD MADNI: And what people don’t realize is that the height of each of those is one million light years. From the bottom of that to go to the top of that, it would take light one million years. And every little finger in that picture, I forget what that Nebula was called now. But twin towers, if you remember what I’m talking about.
BRIAN SANTO: Yes, I do.
ASAD MADNI: And if you look at each of the little fingers, each finger would give birth to a Star Solar System orders of magnitude bigger than our solar system. So yes, that was my favorite.
BRIAN SANTO: You look at that stuff, and it’s just so vast. You can say how big it is, and it’s still almost meaningless, because you just can’t wrap your head around it.
ASAD MADNI: In spite of that, humanity just can’t get along with each other, The egos get in the way. We don’t realize that we’ve been blessed with tremendous intelligence. But yet at the same time, when you’re looking at the magnitude of the universe, we are basically irrelevant.
BRIAN SANTO: I would not argue with that at all. So we get to the point where people have recognized that your contributions have been absolutely amazing over the years, you get an IEEE Medal of Honor for it (well-deserved in my opinion, obviously). But I can’t help but wonder. You already have an IEEE medal named after you. How much bigger of an honor is the Medal of Honor? At this point, is this like the second thing in your trophy case?
ASAD MADNI: No, I would say that the medal that was named after me was very gratifying and humbling. Because it was family and friends that IEEE approached. And they established the honor, added that medal in my honor for my 50 years of contributions to the profession, for philanthropy, and for mentoring the next generation of students.
IEEE Medal of Honor is actually a different thing. As you well know, there is no Nobel Prize in engineering.
BRIAN SANTO: Right.
ASAD MADNI: But in the fields of electrical and electronic sciences and engineering, this is the world’s highest honor. So if you were to say, What would be the Nobel Prize for electrical and electronic sciences, it’s the IEEE Medal of Honor. When you consider the kind of people that have received it — Claude Shannon, William Shockley, [UNINTELLIGIBLE], John Bardeen, Nyquist, Pierce, Darlington, Kalman — I mean, these are the giants that have transformed everything. So to receive this I would say is a very, very humbling experience for me, the most gratifying experience. And I would have to say that this in the trophy case would probably be number one.
BRIAN SANTO: Congratulations.
ASAD MADNI: Thank you. And I say this in a very humble manner.
BRIAN SANTO: I imagine you’ve thought a bit about the future. What do you expect? What do you see in the future of engineering and the future of society?
ASAD MADNI: Great question. So let me start off with the following. And I’ll give you some caveats as we go along.
Today, if you just take a look 30, 40 years ago, we have unbelievable technology at our disposal. The three most important examples of where our advancements have taken place are as follows, in my opinion: low-cost miniaturized sensors, utilizing MEMS and nanotechnology, which makes them ubiquitous in everyday applications in our world. Second, miniaturization and the increased density of memory chips, together with cloud computing for data storage, computation, data manipulation and signal processing. And three is artificial intelligence and machine learning which, as we go forward, I will just address as artificial intelligence. To provide intelligence that can handle large amounts of data and perform previously unimagined tasks.
I’ll first give you a simple example where it’s affecting us tremendously. An area of great importance where advances are being made as engineering strives to better human lives is human-centered technologies. Enabled by converging engineering advances in sensing, computation, machine learning, data communications, all of which draw on the machine intelligence to help understand, support and enhance our human experience.
The challenge for us is to create technologies that work for everyone, while enabling the tools that can eliminate the source of variability or difference of interest. And this is from medical applications to cultural applications to several things. We’ll see great advances here, but if I were to be a person that’s predicting the future, this is what I would say. Artificial Intelligence will play a role in our lives that we cannot even imagine at this point. But I’d like to give you a little bit of a background if I may, Brian.
BRIAN: Sure, please.
ASAD MADNI: By way of background, artificial intelligence is more than 2,000 years in the making, dating back to the ancient Greeks. To protect his island from pirates, it is said that the first king of Crete received an unusual gift from the Greek god of invention and blacksmithing. A bronze robot known as Talos. Like clockwork, Talos was conceptually programmed to circle Crete thrice daily, throwing stones at nearby ships. I’ll come to it and I’ll give you further details of an essay that myself and Professor Achuta Kadambi of UCLA wrote on this topic for that 50th anniversary issue that I talked to you about and we shall be sending you.
Artificial Intelligence relates to form of execution that is demonstrated by machines that traditionally has been associated with humans or animals. The ancient robot Talos, for example, is defending an island, right? An action that is traditionally performed by humans. Likewise, the self-driving cars of today seek to replace a human driver. These examples, both ancient and modern, fall under the realm of what Professor Kadambi and I call weak AI or weak Artificial Intelligence that is preprogrammed to address tasks that would have been given to a human.
The question that arises is this: If AI has been here all along, if Artificial Intelligence has been here along from Talos to self-driving cars, where will the field go next? And this is what Professor Achuta Kadambi and I wrote an essay on for that 50th anniversary issue. The untapped future of Artificial Intelligence, the revolutionary progress awaits, we believe, lies in Strong Artificial Intelligence, Strong AI. Machines act as a teacher to humans. When humans learn from such machines, it is possible to receive unexpected insights that yield a change in practice.
And what do I mean by that? One future of Strong AI lies in scientific discovery. And I think you’ll enjoy this. A disruptive tool to unblock stagnated fields of science. In fact, this is a field where AI must be used, we believe. Where a human can only apply the same known techniques in their arsenal, the unexpected insights from an AI might be the proverbial wiggle that is needed to get the wagon wheel out of the rut.
To see the impact of AI on scientific discovery, let us consider, for example, the field of physics. While certainly a meaningful field that merits continued study, the last 30 years have seen little progress on fundamental questions like explaining the wave collapse. And the wave collapse in quantum mechanics, is a wave function collapse that occurs when a wave function, initially in a superposition of several eigenstates, reduces to a single eigenstate due to interaction with the external world. This interaction we call an observation.
Part of the challenge is that physical observations have become much more expensive to collect the so-called Big Science. And also difficult to interpret by humans. From Newton to Einstein, we have seen a remarkable jump in the complexity of observations required to validate the theory.
The modern physicist, however, has something that neither Einstein or Newton had: ever-increasing computational power. This motivates a new paradigm for physics, which we refer to as artificial physicist. The artificial physicist could operate in a way that is almost contradictory to human. Where a human can test a small set of curated theories on a sparse set of data, a machine can test a huge number of compututorial possibilities on massive datasets. That is certainly a radical change in approach, but hopefully one that can be a radical change in results.
For example, consider a computer program that can rediscover Einstein’s famous equations. We have not yet observed a technology that can automatically intuit these equations. One of the challenges is that Einstein’s equations are human interpretable construct, but a solution with AI might build upon in symbolic equation generation.
However, we believe that the road ahead to scientific discovery is not easy. For the moment, human engineers and computer sciences will have to create the artificial physicists. We will struggle with questions of interpretability. If the artificial physicist were to be based on a deep neural network, how does one enforce human interpretability? In other words, what I’m saying is, How does the output of the artificial physicist guarantee an output equation that meaningfully maps to what humans can interpret?
The future of AI, we believe, lies in grappling with these nuanced challenges. There are multiple frontiers that could be explored. The first frontier lies in the interpretability. If a machine has to teach humans new insights, both partners will speak the same language. Imagine if a hybrid team could be formed with the two physicists working together. One is an artificial intelligence, an AI; the other is real.
A second frontier relates to novel algorithms and architectures to implement AI. Today, neural networks (what we call deep learning( is the dominant approach to implementing Weak AI. However, such methods are preprogrammed rather than self-thinking.
Yet a third frontier of AI lies in unblocking traditional fields; not just physics, but chemistry, medicine and engineering. And I might point out that the word choice of “unblocking” is deliberate. It is one thing to use AI as a tool to augment human performance in a field, much as computers augment the author searching for a word definition. It is however entirely different to have the AI drive the research team unexpected and meaningful directions.
An example of unblocking an action can be found in the optical sciences. Progress in optical design long held that Fourier coated apertures are optimal. With the advent of AI, optical scientists have been successfully using AI algorithms to create unexpected aperture masks that depart from and also outperform Fourier masks.
In a nutshell, what I’m trying to say, I’ll try to summarize with this particular statement: For thousands of years, humans have been teaching AI to do our chores. It might be time that we let AI teach us how to innovate in new and unexpected ways. I hope that it’s not too long.
BRIAN SANTO: It was not too long. What you just described, I’ve used this analogy before on this podcast a few times, and it never stops being appropriate in different places, I think. Years and years ago, when Deep Blue beat the chess master Garry Kasparov. After that, Kasparov actually became excited by the possibilities of a human chess master working with a computer. I’ve heard attributed to him that he coined the term, “a centaur”: something that’s half this, half that. Half human, half computer.
ASAD MADNI: Yes.
BRIAN SANTO: And it’s become, I understand, a common holding in the chess area that a chess Centaur will be any human Grandmaster alone or any chess computer alone. The combination of sheer computational power comes up with combinations a human might not arrive at on their own. And it takes the human to understand what the ramifications might be. It sounds as if that’s somewhat analogous to what you envision.
ASAD MADNI: You’re right, Brian. You’re very right. And you described it very eloquently by giving that example of Kasparov and Deep Blue. Very, very true. You’re actually right on the money.
BRIAN SANTO: Fascinating. So you’re optimistic, yes?
ASAD MADNI: I am very optimistic. I also believe that there are a lot of skeptics in this world, and they should be. Anything, any progress that we make in this world, depends on whose hands it is used. A knife can be used for chopping onions and cutting food or for killing somebody. Vice versa, we’ve seen abusive technology. So these are things that we have to be cognizant of. And as AI starts to develop and starts becoming more human in its thinking capabilities, I think our young students, the next generation is going to have to be very proud properly taught about ethics, not just from religious and moral standpoints, but also ethics from the standpoint of dealing with this form of intelligence, and these capabilities that you have. So they’re just for the benefit of society, and that it is not used for inappropriate tasks.
BRIAN SANTO: A tool is neither good nor bad. It’s who’s using it, right?
ASAD MADNI: So well-said. So well-said.
BRIAN SANTO: Professor, thank you so much for your time today. It was a truly enjoyable and illuminating conversation.
Our guest today was Asad Madni, the 2022 IEEE Medal of Honor recipient.
If I understood Madni correctly, he was referring to a photo from the Hubble Telescope that depicts what astronomers call a star-forming nursery. This one has been named Free-floating Evaporating Gaseous Globules, or frEGGs. The image is the one we used on this podcast episode’s web page. The source is NASA.
And that concludes another episode of the Weekly Briefing. Thank you for listening.
↓↓ 更多权威专访，请点击阅读原文