Connect with Christopher Nguyen
Lisa Ryan: Hey, it's Lisa Ryan. Welcome to the Manufacturers' Network Podcast. I'm excited to introduce our guest today, Christopher Nguyen. With a decades-long career, Christopher's tech bona fides are second to none. Since fleeing Vietnam in 1978, this multiple-time tech founder has played key roles and everything from building the first flash memory transistors at Intel to spearheading the development of Google Apps as its first engineering director. Today he's become an outspoken proponent of the emerging field of Ai engineering and a thought leader in the space of ethical human-centric Ai. With his latest company Aitomatic, he's hoping to redefine how companies approach Ai in the context of life-critical industrial applications. Christopher, welcome to the show.
Christopher Nguyen: Hi, Lisa thanks for having me.
Lisa Ryan: Share with us a little bit about your background and what led you to do what you're doing now with Ai.
Christopher Nguyen: The most relevant thing about what I'm doing now can be considered a failure, starting after my previous company's acquisition by a company called Panasonic. We all know Panasonic as a global engine. However, many people don't realize that Panasonic is less of a consumer company than an industrial company in manufacturing, avionics, and automotive.
The acquisition of my previous company was the apply Ai machine learning to that global engine. Very quickly, we found that a lot of our, let me call it Silicon Valley techniques of digital-first companies like Google and Facebook, and Twitter run into apparent limitations when it comes to dealing with the physical world. The discussion or debate between atoms versus bits, and we've had to develop a whole bunch of techniques that involve leveraging a lot of human knowledge and expertise. We are automating all of that with machine learning to solve these industrial problems. That's the thesis of Aitomatic, the company.
Lisa Ryan: So how do you do that when you talk about taking that human knowledge? How are you taking what we do almost automatically as human beings and turning that into machine learning?
Christopher Nguyen: Maybe I can share why we do that because too many of us today, that is counterintuitive. We thought the future is only data-driven, and we only collect enough data with sensors on machines, and then we feed them and do these machine learning algorithms, and they'll know and don't predict they'll do everything for us.
It turns out that doesn't apply not today and enough for a very long time to the physical industry. Take the problem of looking at sensors on a machine by refrigeration system and then trying to predict in advance. Is this likely to fail over the next two weeks? Is a compressor going to conk out or something like that? To do that, we still rely on human expertise because it's not in the data we're collecting. It's in their life experience. 30-40 years of seeing various refrigeration systems, models, operating conditions, and so on and building up instead of intuitions in their minds over time. We failed trying to do it the other way. We succeeded in incorporating human knowledge. That's the reason we do that. I can talk about how we do that.
Lisa Ryan: That's interesting because when you have somebody that's been in the job for 20 or 30 years and, as you said, that's that feeling that intuition and being able to take a human feeling and turn it into data, that's just fascinating. If there's an easy way to describe how that happens, that would be great.
Christopher Nguyen: If we learn like humans, we're building learning machines. We can either learn from examples, or we could learn from instructions. Data-driven machine learning is essentially learning. For example, learning by example requires lots and lots of examples before you start to build up some experience around it.
But learning from instruction, you could say if the temperature is too high, but the pressure is too low, then signal something that may be problematic soon. So the basics of clarifying or encoding human knowledge are about capturing some of these rules from the past. Their so-called expert system where people tried to do these things. But advances in technology in terms of machine learning itself are enabling us to do this. We can take the natural language you, and I can speak like this, and then it can be translated into something that a machine can understand. Then we can sit down with a domain expert, a manufacturing assembly person who knows machines well. They can say if the sound that comes off sounds like the knock-knock of an engine, then I know to take that offline. We can take the sentence as is, and now our machine learning algorithm can understand it and translate it into code. That code becomes automated and part of the automated system's knowledge set.
Lisa Ryan: One of the surprising about automation is that it's so widely used, and yet according to your information, only 9% of manufacturers are currently leveraging it. So what are some of the challenges you've seen that have stopped that from expanding wider?
Christopher Nguyen: One of the challenges or surprises that I've learned in the last five years, being part of Panasonic before launching this company, is how the meme or the fear is that robots are coming to replace us. Replace thousands of people, and we just put a bunch of robots doing that; it turns out, the lab we use the word profit with most profitable, the most promising applications are not that right. So it is more about solving the problem of not having enough students than not having enough expertise.
In one example, because it's Panasonic, we also operate in Japan. There are these supermarket chains that run refrigeration systems. There are 10,000 supermarkets in one chain and hundreds of thousands of refrigeration equipment. And three experts in the entire country are qualified to diagnose this. It's very much a human expertise constraint. The solution is to quantify what they know and their lifetime of experience, then try to replicate the scale here in the US, where we're facing the same crisis. We've all come on software in the last 30-40 years. We've outsourced our manufacturing - all the tooling or the physical stuff. Now we're finding that it's not just a like economic risk but geopolitical risk.
Lisa Ryan: Especially with the labor shortage we're facing right now, there are two sides to that equation. Number one, it's great to have automation to do the jobs that nobody else wants. So we can start making people's lives easier while requiring fewer people. But, on the other hand, if you're talking about three people in an entire country that has that knowledge, there's gotta be some fear around, "Well, if I communicate everything I know to the machines, then I'm going to communicate myself, out of a job." So how do we balance that where we can get away from the fear of where we're not that we can work in harmony between man and machine.
Christopher Nguyen: I'll give you an example. Here in the US, you may have heard of a company called Huntsman. They make refrigeration equipment. They are a Panasonic subsidiary, a very large operator. They sell in supermarkets. If you go into the freezer section, you'll see the huntsman logo. To build, run and operate such a network, you need a very large force of service personnel who understand this equipment, have experience and can go out and repair them. Unfortunately, there's a massive shortage of people willing to take these jobs. So what does Huntsman do? Believe it or not, they set up universities, but for schools to try and teach these people, they've been paid very well. So this is a general example where we were short on people willing to take these jobs or right because everybody goes to college and gets a computer science degree. Ai machine learning will try to help those solve those problems first rather than working people out of a job.
Lisa Ryan: Right, exactly, and we certainly need both. I know I was talking to my mother today and her air conditioner went out while she's in Atlanta, Georgia, where it's 100 degrees today, and she's in her late 70s. And nobody can come out. They have nobody to come out until Monday. So the labor shortage is real regarding not only people's health, like my mother's what air conditioning, but also at you just said with 10s of thousands of air conditioning units, how do we get the people to do that.
Christopher Nguyen: That's an important example of when you share it. There's a field called predictive maintenance. You're in manufacturing and instrumentation. Where we try to prevent right failure is better than predictive maintenance is better than even preventive because preventive maintenance, you go out and replace everything every six months. Maybe there's a bit of waste, but predictive means you can try to predict that something is likely to fail.
The value of being able to do so is far more than the cost of that piece of that compressor or the labor to go out; it's a life and death situation. It's not like a Google or where you click on the wrong ad. What he stands for is that you are trying to essentially build intrusion detection system cybersecurity for automotive because soon, cars are computers on wheels and will be hacked. If you get that wrong, someone dies, so I think this is a crucial combination: applying Ai machine learning to the essential processes in our lives. We're still physical people. We still drive cars. We do eat fish, and so on. So the impact of failure can be quite consequential.
Lisa Ryan: Well, the interesting thing you just said about the predictive is that Carrier reached out to my mother yesterday via email to let her know that they sensed something wrong with her system. Still, unfortunately, AT&T was putting in fiber, and somebody cut her line, and she didn't have Internet, so it's a perfect storm. But that predictive maintenance is such an interesting concept because if they can let you know. Hey, there's a good chance we're seeing something that's not working. Then, you can send those people we have so few to fix or do preventative maintenance because they know. That there's a good chance that it's going to fail.
Christopher Nguyen: On a typical factory floor, one failure can easily shut down the line and costs $20,000 an hour. The cost of that screw or compressor you're trying to replace can prevent that from happening. So it has a very, very meaningful economic and human life value.
Lisa Ryan: Well, you also brought up an interesting point that ties in with Ai and automation. That is cyber security. The more that we outsource, automate, and take it out of the human control and put it into the machine control, there are a couple of people out there - one or two - who are the bad guys. So what are some of the things you've learned about the risks? Also, you talk about putting that human intelligence back into the driver's seat when dealing with cyber security.
Christopher Nguyen: You may be familiar with SPYCAR legislation. In the US, SPYCAR stands for safety and privacy in your car. The US Senate likes to have these clever acronyms. But essentially, by a specific year, it was initially envisioned to be 2023 but maybe push that a little more. All cars on the road must have an intrusion detection system. Because vehicles are becoming computers on wheels, they are subject to attack. The way technology has been built is that when people first computerized the car. There's something on the vehicle called the canvas, which you think of as this network. So you have all these sensors and actuators and the processors talking to each other security when that was first built was not top of mind.
Because the cars are moving, they think, who will connect to them and hack them? Now cars are getting connected. There have been demonstrations as far back as 2015. That was when a Jeep Cherokee was controlled and was driven off the road. Because it is life and limb, it is human life at stake. It's not just again clicking on the wrong ad. Congress is trying to get ahead of all this and requiring that manufacturers put these intrusion detection systems into cars.
My company provides some of the intelligence that goes into that. These things have to be a knowledge infection. We can see car communication patterns that have not occurred before. So that may indicate a kind of a cyber-attack going on and then being alerted to alert the driver and perhaps shut down the systems before it goes too fast and cost somebody death.
Lisa Ryan: So, as the hackers get more imaginative and innovative and out with that, is that something that you're just continually monitoring and looking at and patching. How does that work? You're getting better, but the chances are that they're getting better too. So how do you continue to protect these devices, these cars, and trucks?
Christopher Nguyen: It is an escalating ever-escalating battle, and the hope, wish, and the promise is that machine learning or learning machines can be better than machines that are not learning. Computers are pretty dumb. All of the intelligence in a computer comes from humans. We tell them exactly what to do. We tell them, step one, do this step. They're very inflexible, but when you apply Ai and machine learning, the hope is because those machines are now learning. They can adapt. They can adapt sufficiently so that they can see new patterns faster than we expect to have a sort of human response. They essentially can begin to pass themselves, which affords us not perfection, but an additional layer of defenses before things get terrible.
Lisa Ryan: So what is it exactly that you do at Aitomatic? Tell us a little about your services and how you work with your companies?
Christopher Nguyen: Some use cases I mentioned include refrigeration predictive maintenance, identifying and counting the number of fish under the ocean using sonar echograms. This is for keeping fish for something called fixed net fishing off the coast of Japan, as well as cybersecurity, automotive cybersecurity, and avionics. So these are the various use cases. We don't implement directly what our customers do, so these customers have teams called engineers. So you have computer engineers, software engineers, and now you have these emerging people skills called Ai engineers. They use automatic tools, so we have tools, and we have a cloud service that enables them to build these systems. We offer the capabilities that I just mentioned here. They integrate them in specific ways to fit their particular use cases.
Lisa Ryan: We talked a lot about Ai and some of what's happening. We also covered the small percentage of manufacturers that are currently leveraging that. Then there are the automotive cyber security issues. What would be your best tip for somebody looking at taking their manufacturing to the next level of using that human intelligence along with automation? Whether it be the Labor shortage or the technology, what would be some of your best tips for somebody listening today.
Christopher Nguyen: What our expertise is and what we believe in. There are lots of tools and techniques to force someone to get into machine learning to develop Ai solutions. If they think that they have an asset that is human expertise right from years of experience and so on, and they want to apply that to their Ai system, they should seek out tools and techniques companies like ours that specialize in something I call knowledge first Ai.
Lisa Ryan: And so, and this year tool, you mentioned earlier about just being able to talk to it in everyday conversation, it sounds like a knock-knock when it does that or how do you how would you even know what to say to the machine.
Christopher Nguyen: Communicating an international language is the first step on the journey. That starts the knowledge encoding. Once that knowledge is encoded, we then automatically generate what's called the machine learning model. Then the whole thing gets deployed into production, and once it's in production, data is still coming in, and incidents happen. New events occur. All of that is then looped back into rebuilding or refreshing the system, so essentially the whole thing is an operating system learned from human instruction. So as well as more data comes in, it will also learn about what's happening in real-time.
Lisa Ryan: And are some industries better than others when it comes to using Ai, or is it across the board in manufacturing that everybody can benefit.
Christopher Nguyen: There's nothing that is best for everything. The difference between the digital and physical industries is that I used to be part of the digital industry right now at Google and so on. Some tools and techniques work with big data that move quickly in and out, and the process itself is also digital. That works very well for certain classes of companies, I think, for manufacturers for automotive companies, for a bionic company, anything that has a physical dimension, we find that those tools and techniques don't work.
We need to have this human expertise, and so I think, for you know, an Ai company like Automatic that focuses on what's called knowledge first Ai is much more suitable for the physical industry.
Lisa Ryan: On the other thing that just popped into my mind as we look at the graying of America in manufacturing, we have all these people walking out the door and taking that knowledge with them. It also gives them a chance to leave a legacy. So all of those decades of information feel relevant because they're contributing to the company's future and truly leave a legacy with the knowledge they've built up in their career.
Christopher Nguyen: yeah, for us, they're not just relevant. They're essential, literally. Sometimes, this knowledge is lost forever and then has to be somehow built up from raw data. That isn't easy to do essentially like eventually, 50 years from now, machines will learn everything, but what we do the in the intervening 50 years.
Lisa Ryan: Right. We speak a lot on the show about workplace culture and helping employees feel that they're valued, appreciated, and part of a bigger mission. In this case, with that human intelligence, that human factor, you can work it into your conversations. Let the managers say. This is your legacy. We value you, and that's why we're doing it. We're not trying to take people out of the equation. We want to bring your knowledge into the equation.
Christopher Nguyen: Machines, when used correctly, will always augment us rather than replace us.
Lisa Ryan: Exactly. Christopher, it has been a pleasure having you on the show today. If somebody does...