On today’s show, we interview Geoff Seyon. Geoff is the Founder & Managing Partner of STUDIO IXORA, a venture studio focused on applying Explainable AI to industries such as Finance & Healthcare. Geoff is an Entrepreneurial Technology Leader with 27 years of experience, inclusive of 15 years in Globally Distributed Software Delivery, Digital Health and 10 years of hands-on experience with various Artificial Intelligence approaches & technologies. Prior to assembling STUDIO IXORA, Geoff spent 15 years assembling, advising and leading multiple teams that have created 7 Digital Health and 10 AI New Ventures. Before that he also had spent 5 years at Sapient where he grew an office in India to 700 people and received the 2002 Sapient Founders’ Award.
Geoff holds Master and bachelor’s degrees in electrical engineering & Computer Science with a Concentration in Artificial Intelligence from MIT, as well as an MBA from Harvard Business School.
In this episode, you’ll learn:
- How do overseas engineer programs match up with the top ones in the United States?
- Are there any skills that our engineers currently lack?
- How hard is it to learn how the hardware and software interact in a computer?
- What is the difference between Artificial Intelligence and Machine Learning?
- What part of our mind is being replaced in the 4th industrial revolution?
Help us out!
Help us reach new listeners by leaving us a rating and review! It takes less than 30 seconds and really helps our show grow, which allows us to bring on even better guests for you all! Thank you – we really appreciate it!
Connect with Guest:
- Connect with Geoff on LinkedIn
CONNECT WITH SHAWN:
https://linktr.ee/ShawnflynnSV
- Shawn Flynn’s Twitter Account
- Shawn Flynn’s LinkedIn Account
- Silicon Valley LinkedIn Group Account
- Shawn Flynn’s Facebook Account
- Change to Shawn@thesiliconvalleypodcast.com
Disclaimer to the Transcripts:
Intro 00:00
You’re listening to The Silicon Valley Podcast
Shawn Flynn 00:03
On today’s show, we have Geoff Seyon, who is the Founder & Managing Partner of STUDIO IXORA, a venture studio focused on applying Explainable AI to industries such as Finance & Healthcare. Geoff is an Entrepreneurial Technology Leader with 27 years of experience, inclusive of 15 years in Globally Distributed Software Delivery, Digital Health and 10 years of hands-on experience with various Artificial Intelligence approaches & technologies. On today’s show, we talk about the talent of engineers that’s overseas, the advantages and disadvantages of outsourcing for a Silicon Valley company, and what is the difference between artificial intelligence, machine learning, and much, much more. Stay tuned for an amazing episode. Enjoy.
Intro 00:48
Welcome to the Silicon Valley Podcast with your host Shawn Flynn who interviews famous Entrepreneurs, Venture Capitalists and Leaders in Tech. Learn their secrets and see tomorrow’s world today.
Shawn Flynn 01:11
Geoff, thank you for taking the time today to be on Silicon Valley.
Geoff Seyon 01:15
Shawn, it’s a privilege to be here, man.
Shawn Flynn 01:16
Geoff, you’ve been in the valley now for three decades. Can you talk about kind of what you’ve done in this time? But before that, tell listeners, how you even arrived here, a little bit about yourself, your past and, your history.
Geoff Seyon 01:31
Absolutely. So I got to MIT, really on the back of a book that my dad gave me on the coming war between Japan and the US and the 1980s. And of course, we all know, at this point in time that nothing much happened in AI in the 1980s and 90s because that was the most recent winter of AI. But since then, a lot has happened. So I studied computer science at MIT. I graduated and I ended up in the consulting space because again, even though I wanted to do AI, I figured just let me just go learn other business problems and try to solve those with the technology mind that I had been given. And I ended up working at a company called Sapient Corporation for five years, right out of MIT. And for two of those five years, I ended up helping grow and scale a company in India. And that’s now quite an amazing company, which we call Sapient Publicis now. And it’s about 100 companies, which is part of a conglomerate in France. But the foundation for the technology delivery were the 700 people that I had the privilege of managing, and those 700 went on to lead an office which became over 16,000. And those 16,000 are now the bedrock of that conglomerate called Publicist Sapient.
Shawn Flynn 02:46
Can you tell me more about your experience there in India with that outsource team and kind of finding the engineers? And were they what you expected when you got there?
Geoff Seyon 02:55
So I came from a small country, small island, if I pointed to it on a map, you probably couldn’t even find it. Even people from Trinidad and Tobago don’t know what town I came from. So, for me going to India, where every city block felt like the population of my whole country, was eye opening. And on top of that, we had some amazing engineers at Sapient. And we interviewed them with a pretty high standard because we were in front of some of the world’s biggest and baddest Fortune 500 Enterprise Class clients, delivering fixed time and fixed price software solutions.
Geoff Seyon 03:33
When we got to India, the level of technical immersion blew me away. We were recruiting mostly out of a system of schools called the Indian Institute of Technology, and some other affiliated schools at the same tier. And these kids were just incredible, such a pipeline. And what I was blown away by is that this was not a different type of talent than what we had hired at Sapient. Certainly, they had not been given the same opportunities that we had had growing up and studying in the US. They had just the same potential, they just didn’t have the same exposure that we did. And so part of what we had to do as a mission, to survive the 2001 recession, and subsequently then, 9/11 was the *flattened company that existed largely at the time in North America and in Europe, in the developed world, and integrated with a team of counterparts based in Gurgaon, which is near New Delhi in India. And that back end was not the low end of the work. It was a complimentary part of how we delivered software for the client. And learning that was probably one of the most important things that taught me about innovation and human capital development and global software delivery.
Shawn Flynn 04:55
Can you talk more about building a global company and the human resource behind it and the complements and building those skill sets?
Geoff Seyon 05:04
So you are probably not going to have a hard time finding good technical talent when you go abroad. There are certainly amazing engineers in this country as well. There is a general notion that you will find stronger technical skills abroad and stronger communication skills in your client geography. That’s not always the case. What’s more important to focus on, I think, is the culture. Are you aligned with the counterparts that you’re recruiting from anywhere else? They don’t have to be in India. They could be from Eastern Europe. They could be from Africa. They could be from Latin America. But at the end of the day, all of these different cultures bring something different to the table. The question that I think we, as managers in Silicon Valley, have to be able to make better and better now that we’re running out of places we can hire people from is what is that our company is supposed to be great at? And then instead of trying to say, what does that geography have to offer me or what software developers exist in that particular city or from that particular school, ask instead, “How do I find the people who will help me go from good to great as a company?”
Geoff Seyon 06:20
And to do that, I think there are three phases that you have to go through today, which is very different from what I had to do in in the early 2000s. We had to physically go there and set up an operation. You may still have to do some aspects of that today. With Upwork, Fiverr, and general freelancer type of outsourcing application, you can find great people quite easily. That’s the first step: finding them, giving them an initial project, understanding how they work, whether they fit in, whether they align, that’s the first step. The second step is then growing them, investing in them, understanding what it is you have to do to teach them your protocols, how you work, your methodologies, your technologies. And the third step is keeping them because once they become good, they are no longer going to cost 20 bucks an hour. They’re going to cost 50 bucks an hour and you’ve just trained your competitors. Human labor, right? So there’s quite a simple formula in my mind, but it doesn’t take away the elbow grease that is necessary. I feel at the end of the day, if you’re an innovation, you got to be in human capital.
Shawn Flynn 07:30
Do you think right now, with Silicon Valley, the traffic gets worse and worse every year, housing prices going through the roof, it’s harder for people to try to even live here. Do you see that outsourcing will be growing and growing in the coming years and that more of these huge corporations will hire even their initial teams overseas to work on projects or startups here in general? They might just have one person here but then have all the engineers in India and China?
Geoff Seyon 08:02
So it’s a crucial question. So the first part is, it’s inevitable that we have to figure out how to scale outside of Silicon Valley. So some of the stats I pulled up recently from… There was a Barron’s article in February about Silicon Valley has nowhere to grow. It says that we have about 436,000 or more tech employees. That’s just tech in the lower part of the valley, down here on the peninsula. And it grows by about 15,000 per year. Then up in San Francisco, they presently have about 100,000 or more employees employed at companies like Salesforce, and they grow to about to a thousand per year. And a lot of that is Salesforce. So of course, we can’t put these people in Silicon Valley, the demand isn’t going to go away. You’ve looked at the stock market over the last several years. You’ve looked at the penetration of some of the major platform companies. You’ve looked at the amount of venture capital which has gone into everything, from SAS, to big data to what have you. Digital health is big. I’ve spent a lot of time in that. Now, people are raising several billion dollar funds to invest in AI. It’s not going to stop. The demand isn’t going to stop.
Geoff Seyon 09:12
So the question is, where do we get people from? There’s only one solution, we have to go globally. Now, globally, these services like Upwork and others, they have tens of millions of people. So how do you find your complementary 10,000 that fits into your team? And I think it’s actually quite easy, given that it’s not Google trying to do this hiring, right? We’re not trying to take Google from 50,000 in the US to 100,000, globally. That’s a harder problem. We’re looking at a startup, which typically has a few hundred people. So they will find a few hundred amazing people around the world that somehow are configured and aligned with what your corporation is supposed to be the best at. It should not be too hard. And so I think, given the immigration issues we have here, given the housing issues, given the fact that once again, we have a whole new generation of software managers, I think we should invest in learning some of these skills of global delivery and innovation management.
Shawn Flynn 10:11
Can you talk about some of the struggles or surprises with your global team that you managed, getting that information back and forth, or just managing overseas and talking to another department that’s in another country?
Geoff Seyon 10:25
Yeah, so I will start with the surprises, whether I was doing this in 2001, or even more recently this year. Last year, I found that for the portfolio companies that I built, you will always find talent that you didn’t expect to be there. The question is, how well are you paying attention? And I think this is where startups have an advantage. But startups are much more precise about what they need. They’re much more assertive and alert of talent which comes their way. They’re also much more willing to make an investment because, you know, for startups, we will take whatever we get and make it work. But that also means that we’re willing to work with people and grow them professionally to match the needs of our corporation. So I think it’s quite easy if you put the time and effort into finding people, to shape them towards your standards and your ideals, in terms of culture, in terms of your communication modes. And the technology is also important, but it’s also the least important of the three. And so I think, if you spend the time listening and looking, you’ll find amazing people.
Geoff Seyon 11:38
One example for me is we found… We were on a trip to Bangladesh; it was a personal trip. Typically, you would expect things like artificial intelligence, deep learning, machine learning would first go to places like India before they go to places like Bangladesh. Mostly because of different statuses of the universities, different capabilities, different standards of living and what not. We ended up finding a guy who was selling his household possessions, his bike, his motorbike, and things like that, just to buy a graphics processing unit board so that he could complete his bachelor’s thesis in deep learning. And this kid’s dad had just died, and he managed to finish his degree. But when he showed us what he had done, and some of the public contributions to the field of artificial intelligence that he had made, we were blown away. This kid barely had any mentors. He was mentored by a couple of professors from one of the notable international universities in Bangladesh. But other than that, it was tough going for him. But yet, he was able to do things which were, I wouldn’t say the same as our counterpart teams who were trained at places like Stanford and UCSF, but he was able to catch up pretty fast. He just had to be given the opportunity. And fast forward a year and a half after working with us, we actually helped him by writing recommendations, encouraging him, and giving him guidance and feedback. We got him into a US PhD program so that he could come here and continue to further his education, but also continue to develop his progress in the field of deep learning and AI. So that blew me away. That’s one of the positive surprises that you will find if you spend the time looking. And I have hundreds of them from over the years. But that’s a more recent one.
Geoff Seyon 13:26
The challenges are that sometimes, things which are obvious to us as innovators, entrepreneurs, and business managers here are not so obvious. We don’t realize how much time it takes to teach someone the conventions of doing something. So I’ll give you an example. If you’re trying to work with offshore workers who are helping you with mining data or automating things like a chat bot or other forms of data gathering, like that for textual type of training of AI systems, you might find people who are better than you expect, in terms of the written or spoken English skill. But teaching them how to combine that skill, which most American graduates from most American colleges would be able to combine with the language skills, the humanities skills with the technical skills, and then be able to say, “Let me also figure out how I can take the dialogue that I wrote or I conducted, this interview that I conducted here, and translate now into the types of technical structures and divisions that can be used to train the machine.” It’s hard. And in vice versa, if you ask a person who understands it from a technical point of view, they may not quite understand the business context. And they’re not willing to spend the time as a call center worker in order to do it because they see, societally, as this is a task that’s below me. I’m a software engineer. I’m not going to pick up the phone or text back at customers. That’s not my paygrade.
Geoff Seyon 14:59
So, the reality is that, unfortunately, unless we provide guidance to the global teams, and make the investments in them, they want to continue on the pathways that their respective countries and the local industries have left them on, which is about to be replaced, just like all of the other industries here by machine automation. And they have a golden opportunity to rise exponentially towards the tangent of adding value at the Silicon Valley level, but it’s going to get away from them unless we start reaching out and investing in communities that we care about around the world.
Shawn Flynn 15:38
For them to rise up to that level, would it be more influenced in the university schools or more additional coaching at the professional level? What needs to be done to have them compete at a level that’s here in Silicon Valley?
Geoff Seyon 15:55
So that’s an excellent, excellent question. I won’t quite say it takes a village because the concept of a village is very different globally. A village is probably 100,000 people in a place like India. But I would say it takes the commitment of a corporation. The corporations are the villages, because they reach into their respective communities where their people come from. They provide transportation, they provide housing, they are almost like a small country. That’s what it was for us, right? We have to do everything, from providing transportation, security, and housing for us as expats. Some of the local hires as well, all the way down to decide which vendors we buy meat from.
That’s a difficult decision when you have 100 expats on the ground in New Delhi. So these companies have the infrastructure. The question is, are they making the investment in the education system? So for us, when we were in New Delhi, we partnered with IIT to influence the curriculum. And then, even when we got graduates that joined our company from whichever school, whether it be tier one, tier two, whatever… We made the first three weeks of the training program where we invested in them, but there was always feedback. And so you have to keep investing. And then you send them back to do more recruiting at their own colleges. And that influences the curriculum. Then maybe some of them will go back to the high schools, and so on and so forth. And then some of them leave your company, inevitably, they all will leave. If you have 20,000 people, you know, the 700 people I work with back in India in 2003, probably are no longer there. I’m guessing maybe 70 of them at most are still at the company. So over a longer period of time, a larger number will go elsewhere. The question is, then what happens? Well, they’re probably going to go to other companies and influence them with the high standards that you set. And so this is how communities change over time. And these general ingredients seem to be the things that allow places like Silicon Valley to evolve in the first place or places like Boston, Austin, Texas, North Carolina. Places like Berlin, great arts programs, great universities, a cluster of companies, lots of startups, and then a ripe and a socio-economic environment. It’s the soup that matters. So I’m not saying that tiny companies have to boil the ocean. But tiny companies get big pretty fast. When I went over with Sapient, it was 50 people. By the time I left, two years later, it was 700. And at one point, we were hiring 200 people per month. So you can make a big impact pretty quickly.
Shawn Flynn 18:32
For companies like that, getting the talent for 200 people a month, you’d mentioned there’s talent everywhere. And you also mentioned the university ITT. Can you talk about the number of graduates that come out of these universities every year, and kind of the competition level that they’re at?
Geoff Seyon 18:51
Yeah, so I think in one particular month, if I remember the statistics correctly, not 200 but I remember one month, there was a statistic where we had hired, in Gurgaon, 77 people. To get to those 77, we went through 13,000 resumes. So that’s about a half a percent pipeline rate from applicant to actual hire. It is difficult. And in our case, we were building a consulting firm, and a lot of the filter, a lot of the challenge was in English communication and client facing skills. Technical was usually not the problem. It was probably about a 50% pass rate in India, as opposed to a 20% pass rate at the top places like Boston and San Francisco. So the technical challenge was not the big hurdle. It was making sure we found people who were aligned with the culture and the communication values. But let’s talk about the stream. And this is why I think it takes some visceral learning on the ground. There’s a big difference when you hire grads from tier one schools in whatever country you operate in, okay?
Geoff Seyon 20:03
So there’s a system of schools called the Indian Institute of Technology, IIT. And I think right now IIT, across, there are about 16 of them now. When I was there, they were only about six or seven. IIT has the space for about just shy of 1000 computer science graduates a year. The Indian Institute of Technology system takes applicants in from some standardized local tests. They get 1.5 million applicants into that pool who are competing for those less than 1000 computer science seats. So if you did the math on that, you have a pipeline that’s coming in, and you’re getting 0.1% through the other side of the funnel. So let’s try to compare that to MIT. So at MIT, I looked at the most recent class here. MIT will typically graduate in Computer Science, approximately, just shy of half of the class. The class size, the undergraduate class size is about 1100 people. So just shy of half of that, let’s just call it, you know, 550 will graduate with a computer science degree or an affiliated degree that’s tied to computer science. But MIT is only taking in right now 21,000 applicants or so. Only 21,000 people are applying to schools like MIT. So what that means is that that pipeline there is 2.5% as opposed to 0.1%. Or if you want to put it in financial terms, the schools in India are 10 basis points, whereas MIT is at 250 basis points, in terms of throughput. It’s a big difference. It’s a 25x difference. So obviously, the quality of grads selectiveness is going to be very different.
Geoff Seyon 21:57
Now, where MIT has the advantage is that you’re exposed holistically to a lot more types of things. And Stanford and Harvard and a lot of the top tier schools, especially engineering schools, that are more holistic in terms of what we’re trying to do to build successful Silicon Valley class businesses. But this is emerging in India and in China, and a lot of growing economies where the school systems about large numbers of computer science graduates. And it’s an opportunity for us to partner with these geographies, because certainly they don’t have right now as much venture capital as we have access to here. But is that the only differentiating factor? Right now, it is one of the larger ones. There are certain other aspects of Silicon Valley that differentiate us. But who knows? In 20 years’ time, is it going to be the same? And if you’re still building companies, if you’re on the third fund 30 years from now, where your portfolio company is going to be focused. So it’s an investment that’s worth making today.
Shawn Flynn 23:01
But these engineers that are graduating, I mean, maybe they don’t have the same soft skills. But are there any technical skills maybe they’re lacking? For example, maybe they’re studying a type of computer language that’s a little outdated. Or maybe they’re lacking in engineer skills when it comes to the hardware. Is there anything that maybe the general population might be lacking or steering away from that it would be needed in the future with the new developments in AI?
Geoff Seyon 23:30
So it’s a good question, Shawn. I would… I’m not sure it’s the right question to necessarily ask because, again, if you think of the world of skills as a portfolio, whatever skill you’re looking for can probably be found somewhere, right? And I will make a bet that the geographies that focus on a particular technology or particular type of way of thinking or particular skill set, because of larger populations and becoming better and better institutions… Increasingly better institutions in those geographies will probably outcompete our specialists here in the United States. They’re doing it already. They have been for the last 20 or 30 years. A lot of them migrate here and join companies like Google, Facebook, Microsoft, and Apple. But that’s not the only factor, the technical skills and the alignment, and whether or not the language is current with what we need is not typically the factor. It’s the other intangible factors. So it’s things like the culture. It’s things like the language. It’s things like the timezone to be quite frank. Is there an opportunity to have a physical interaction with them? Either we are traveling there to spend time with the teams or they’re traveling here. Are they permitted to travel here because of visa restrictions? And the last factor I would throw out is probably the educational one, which we talked about just a while ago.
Geoff Seyon 24:59
If they are coming from educational systems which are very different from the ones that we are building these companies in, it’s going to be more difficult. Right in the woods, it will be harder to hire a computer science grad out of a liberal arts school in a country in Eastern Europe and have them write software for a bunch of MIT or Stanford engineers here. But that’s not necessarily saying it will be incompatible, right? You have to figure it out for your company.
Geoff Seyon 25:31
So there’s this function, which is a function of culture, language, timezone, physical, and educational proximity. That’s the question to ask, because the way I’ve experienced it, is you’re growing your company, and it has a certain focus on those elements. The people that you bring in to complement your team should also have, as a group, the same focus, and that weird makes it easier for you to grow together. If it’s not, you’re going to find that a lot of energy is being put into convincing someone why the culture is one way or the other, or convincing someone why you chose a certain way of making a technical or business decision, or convincing someone that this is the right language to speak, or the right language to use in this scenario, instead of actually just getting the job done, because it’s all about execution at the end of the day.
Shawn Flynn 26:23
What about learning how the hardware and the software interact?
Geoff Seyon 26:27
That’s a great question. So I think that’s one area where right now, we would have an advantage in the US because, in general, the education system is a little bit more holistic. So at MIT, for example, the curriculum is almost inseparable. When you do computer science, you do electrical engineering. When you do electrical engineering, you do computer science. I think it’s a good measure of a great engineer as to whether they can understand what will likely happen at the hardware level when they write a line of coding. And vice versa, it’s a measure of a great hardware engineer, if they have a community of software developers that are writing software in their platform, I think there’s a lot of things that are missing to teach that in the current curriculum. And I don’t understand why I don’t see many schools do it that way. I think it’s the right way to do because what happens is when you don’t understand the underlying hardware. And this is where it becomes a challenge for global populations, if you don’t have a rack of servers, like an engineer at Google would have to just go roll up and run some massive parallel algorithm on, you’re not going to be able to write the same type of code that someone here is writing, or you’re going to have to write millions of lines of code as opposed to a few hundred lines of code. And you can still get the job done.
You’re just not getting it done in an alignment with what the companies here would need you to know. So I think there needs to be some work we all do, in terms of introducing the difference between, for example, just as one example: the world is changing very rapidly from serial architectures of the CPU to the parallel architectures of the GPU. And the GPU is just a hack, we use it to draw the graphics on our screens. But eventually, some very smart data scientists realize, “Well, this is structured as small, tiny processes, which can be used in the way that a bunch of parallel processes could also be used.” And in my opinion, I can even send you an audio clip of a nine-year-old, such as my daughter, and explain the difference, right? If you ask them, what is faster for kids crossing a street? Should they all cross the street in a straight line? Or should they all cross the street, holding hands in parallel and just cross? Hopefully, the crosswalk is wide enough to cross all at the same time. Which is faster, which is safer? And anyone could answer that. And so I think we need to build those kits. If Arduino is building a kit or Lego is building a kit, we need to teach these types of concepts at all levels. And if it’s that cheap, it doesn’t matter if you’re here in Silicon Valley, or if you’re a kid in high school in a developing country to learn these concepts from in this day and age.
Shawn Flynn 29:10
So how much harder is it to learn how the hardware interacts? Is it something that someone could learn just from a few courses? Or is this something that if you’re not learning it in tandem with your software coding, you’re never going to be able to merge the two?
Geoff Seyon 29:26
This is a little bit more on the technical side, but for people who already write good software, and they are starting to become aware of the changes in the computing environment, meaning we have access to much more parallel computation. These are three things that I would encourage people to think about. So again, almost everyone has this. If you have a MacBook today, if you have just about any current laptop purchase over the last year or two, if you’re working on a server, if you’re working on one of those desktops, they all have even CPUs that have multiple cores. And if you’re writing code, you could understand how to leverage multiple cores at the same time instead of just one. So the traditional education system teaches us how to do one at a time, because that’s kind of how we think. However, even though the recipe is written in a serial fashion, do this, do this, do this and do this. Step 1234 or 5, we all know that the experienced professional cook or not even professional, just any experienced cook can figure, okay, “I can do a couple of these things in parallel. I can pre-prep this, I can do this. I can do this while I’m doing this.” And they figure it out. But code is not that smart. Code can’t interpret what you said, and re-parallelize or re-serialize itself. You have to be explicit about how you want the machine to treat your recipe for getting something done.
Geoff Seyon 30:50
So the first thing I would think about is, first of all, is the algorithm that you’re trying to apply, is the problem that you’re trying to solve. Ultimately, something that can operate in parallel, is it parallelizable or not? But if it is, and you’re going to make the investment in time, there’s a simple local Amdahl’s law that can tell you how much faster it will be. And then you have to make the tradeoff between, do I write the code to leverage the parallel architecture? Or do I just let it run slightly less efficiently than it can be, because it’ll cost me too much to overhaul it? So that’s the first thing I would think about for any piece of software, starting 20 years ago, which we still use to 20 years into the future. That’s the first step: is the code something that is inherently parallel? And you’ll quickly find that most things are.
Geoff Seyon 31:39
The second thing is, now that you’ve implemented the code in parallel, and you’ve tested it, and you’ve done all of the usual things with it, what’s the state of the hardware that it’s being run on? Is the state of the hardware fairly constant? Or is the state of the hardware changing over time? So to give you an example. So the Apple iPhone 10 is the one that came out in 2017. There was a neural engine in that phone, which did matrix type computations in parallel. So it added, multiplied and, you know, did these complex operations in parallel. It did 800 billion to 0.8 trillion of them in one second. So 0.8 tops is what it was called. By the next year 2018, when the iPhone 10s came out, and Apple moved from the A11 to the A12 chip. That chip now had a neural engine, which did 5 trillion operations per second. So in one year, your hardware went from 800 to 5000 billion operations in one second. And it’s because they change the architecture, they put eight of the chips instead of two, and they ran it a little bit faster. And that fundamentally changes the equation. If you’re using that part of the chip for a lot of work, you have to figure out how to take advantage of that. And if you didn’t, you’re going to lose out, you’re going to be building apps, which are one fifth the speed of everyone else’s apps.
Geoff Seyon 33:09
And then the third thing I would think about is, what’s the complexity? Again, there’s a little bit on the technical side, but in machine learning, and in general, in the new era of artificial intelligence, you’re using the machine for one of two things. You’re either training a model, or you using the model that you train to predict something. So the classic example is, I have a bunch of images that can recognize a certain species of plants, I have a certain situation that I’m trying to recognize: is there traffic jam on this highway? And I take a million photos and I just stuffed it into the neural network and it trains a model. And I use a model and I don’t need to have a reporter sitting there looking at the camera saying there’s a backup on you know, 101. The system just tells me, “I can see it, I could recognize it. This is a traffic jam.” That’s a prediction or that’s an inference. So you have to think about whether or not are you training a lot, meaning the data is changing so much and you have to keep training, such as with weather prediction, or is this a pretty static problem? Both cars are going to look the same tomorrow, and probably 10 years from now, 10 years ago. And these are the types of things that would drive decision making around what happens with the hardware and what I think people need to think about from a software point of view and entering the world of parallel computation.
Shawn Flynn 34:31
Where do you see the big value in the future? Is it in artificial intelligence? Or is it machine learning? And tell us what your definition of the two are?
Geoff Seyon 34:40
That’s a very, very timely question. So let me start with an analogy, right? If you’re doing renovations on your home or office, you’re probably going to hire a general contractor. A general contractor doesn’t just show up at one category of tools. So machine learning has become very useful over the last few years since the proliferation of these low-cost parallel processes or GPUs, and the subsequent advances in the state-of-the-art software running on them. So much like the electrical power tools that revolutionized our toolkit, I’m sure you don’t use a hacksaw anymore, right? Or you prefer the drill where you just press the button then it turns the drill? The machine learning automation suite of tools is allowing the general contractor of solving problems to become much more efficient, and much more much happier as a person building something useful, that I’m much more productive than even five years ago.
Geoff Seyon 35:42
There are still things that AI has been able to do amazingly well. So I’ll give several examples because people tend to treat AI as if it’s this old school discipline from the 1960s. And I would say it is, but there’s some good things about that. You can put aside your criticism about the boomer generation for a while. But in 1967, a kid submitted a PhD thesis. He created a program that could do freshman level algebra and calculus. And his programs, which he submitted, were written in at the time. And they could differentiate, they could integrate, and they could solve differential equations. Okay? So, can your freshman do that? If you’re an investor with a kid off in school or CEO, can your freshman use deep learning to do that, even if he’s a whiz bang, deep learning architecture? I don’t think you can. But they did it. And they did it with a very, very low complexity set of tools on the order of 100 symbolic lines. It’s quite amazing. And that system is still applicable anywhere you probably as part of MATLAB. I don’t know. But it’s that straightforward. It’s just creating it was the intelligence in the first place.
Geoff Seyon 36:55
There are several places where going to that symbolic representation is useful. In healthcare, which is a feel I’ve spent, you know, a greater part of couple decades. Medicine still has protocols. So even though you might take buzzy data and assess a situation one way or the other, you as a physician still have to make an ultimate decision, right? Do you amputate? Do you not amputate? Was this person’s eyesight good enough that I give them three diopters or 2.75 diopters of strength? Which do I take? That’s the judgment call of the physician, but it’s a fairly discrete medical prognosis that you’re making. And so having the symbolic language wrapped around these continuous variables, which is really more what the world of deep learning relates to, and the variables that we can’t understand. They just say, it’s somewhere over here. Then you have to pick the discrete value that it relates to. It still falls into the realm of traditional AI and ultimately human judgment to wrap around the less useful tools on the interior, which are powerful, but they’re not the final result that you’re looking for. And there are other examples that we could go into, certain classes of discrete logic that can’t be solved very well with neural networks. I don’t know if many people who are CFOs want to deal with regulatory financial issues, trusting in your neural network to tell them which forms to file and how to check the boxes. These are specific problems that will eventually get automated, but they’re probably going to get automated by a discrete symbolic language representation of knowledge that is more similar to the type of systems that we built in the 60s, 70s, 80s, and early 90s.
Shawn Flynn 38:42
That would all fall under the AI category?
Geoff Seyon 38:45
Yeah, so think of AI as the toolkit. You have machine learning. And if you want to call it non-machine learning, that’s because humans have to code it using a symbolic representation of what they want. We could just call it code or Python. But at the end of the day, traditionally, I would refer to it as the symbolic side. And I would refer to the machine side as more the continuous predictive side. It’s hard to actually say what the machine is doing. So that’s actually, to that question, it’s one of the basic problems with machine learning. Is that we’re struggling to overcome right now, which is, why did the machine exactly predict this? And can you explain why you made the decision? There’s been some progress. But the progress has mostly been made by people with very large budgets, the Department of Defense, and a lot of these contractors. Some of the bigger AI companies have been working on it. But it’s an important problem to solve for a whole bunch of societal reasons, which I won’t go into right now. But that’s a whole separate discussion.
Geoff Seyon 39:51
And machine learning, of which deep learning is a subset of, they are even more and even more nuanced subsets inside of there. Lots of approaches that you can use the machine for. But it can do things at such a high velocity with such large amounts of data that we can’t practically sit down and code for it.
We can’t write the symbolic representation in a manner which is useful. So I’ll give you an example. If there’s a system that predicts a hurricane coming to your state, do you really care why it’s coming? Will you just care that there’s a reasonable likelihood that you need to tell a million people to evacuate, right? You can always go figure out why the machine predicted it and whether or not it was correct after the fact. But right now you need to get a million people evacuated. So that’s a good example of where machine learning trumps AI. We don’t have the time to code it. We don’t want to code it because we don’t actually know how to track 10 to the power of, I don’t know 30 molecules in the atmosphere and predict whether this thing is going to sit over there, like the most recent one over the approach Florida. Is it going to sit over the Bahamas for three days? That’s never happened before. But has it been predicted? No, but it generally predicted the path. And that’s really what matters.
Shawn Flynn 41:07
So where do you see the most value for an engineer to focus his time?
Geoff Seyon 41:11
So I think it’s a broad question. I think that most engineers should think about what they want to do with their career. There’s the caveat. Most engineering careers are relatively short lived. That’s unfortunate. Most people make it into management and then they go off quickly and do something else, go to business school, learn other general management functions. And I don’t know if they keep their engineering background as close at hand as they should, because it was probably one of the more expensive investments that they made in their life. But I think if you are dedicated to growing yourself as a professional in this day and age, I would start with getting out a book on Python, trying to solve some problems that are interesting to you. Get access to a graphic processing unit or server. Either buy one, just a couple hundred bucks for a low end one. Or rent one, it will charge you just a few bucks to rent one and start writing some, buying some things that will teach you exactly what’s happening.
Geoff Seyon 42:00
If you’re an Apple iOS developer, a lot of this is built into Xcode, right? You can teach yourself how to build a neural network with three lines of code. Classic Apple, right? They’ve taken what Nvidia and these guys give you like 12 thousand files to do. And they made it three lines of code. It’s awesome. Get involved. It’s a vast field. The PhD is a way out. Google is a way out ahead of a lot of people. Hopefully they’re all solving real world problems. The practitioners like myself, we’re trying to solve day to day business problems, which right now people are still coding things like VB script. God help us from the 1990s to solve and it’s not going to work, right? But these are everyday problems and financial services and transportation and healthcare, big problems. Try to solve them without leaning on the crutches of the past. I would say try to figure out what can you do? Go on YouTube, there’s a ton of stuff on YouTube. Go take a class in parallel architectures who will understand this language CUDA that Nvidia teachers and you will get the gist of it, you don’t have to become a master at it. Just get your head immersed in it as a software professional and start asking those questions. Is this parallelizable? If it is, okay, is the architecture going to be stable? Or do I have to keep on adapting it? And am I doing predictions more? Or am I doing training more? And start thinking about the world using that lens.
Shawn Flynn 43:49
The engineers that are right now employed by some of these huge companies in the valley, how replaceable do you think they are?
Geoff Seyon 43:56
I don’t think anyone’s going to replace them because they can’t even find the new people that they need. That was also said in the late 90s, right? And we all know what happened two years later. So I would say, just remember, there’s lots of case studies on this.
But I’ll just give my anecdotes from a high level. If you focus on only low-level stuff, without a system level view of how you’re putting the hardware and software together to solve the problems that you’re solving, the low-level view will get you in trouble because eventually, technologies are going to become obsolete, ft not all of them. Python is probably going to be around for a while. I started using Python in 2002 or 2001. Python One. One of the first companies I built in Silicon Valley was on Python Two. More recent stuff we’re building on is on Python Three, it’s probably going to be around for a while. I see Microsoft making investments in Python. When you look at the code repository, such as GitHub, Python has quickly ascended from obsolescence to number four last year to number two this year. JavaScript, which everyone seems to hate, but everyone seems to also know is number one. But we’ll see. I don’t think that next year that JavaScript will be number one. So if by next year, you’re a software professional, and you haven’t at least led with Python, I think you’re approaching obsolescence.
Geoff Seyon 45:22
And there are many other tools. Python isn’t the only tool out there. There are more sophisticated tools than Python. But Python has such an incredible ecosystem and developer library and support structure around it. Microsoft is adopting it, Apple is adopting it. Apple usually doesn’t adopt much from outside of the corporation. Apple has its own fantastic software language, which is Swift, but it’s used mostly for iOS apps. It’s not used for anything else. So if you’re building Apple applications, or you’re getting involved in an Apple environment, there’s a Python interpreter and compiler built into Xcode today. And that’s pretty amazing. As far as I know, that’s the first time Apple has adopted a language other than building it itself.
Shawn Flynn 46:06
When a founder is building a company, you’d mentioned first… one of the companies you built was on Python One and then later Python Three. How important is the chosen language that team builds the product on for when an investor looks at the company?
Geoff Seyon 46:22
That’s another terrific question. And if we could answer it consistently, you would have stacks of money sitting in the corner of your room here. So Shawn, my answer to that from where I sit is, you can’t look at software as static. Software is not like a filing cabinet, or a crate of supplies that sits in your company’s warehouse. Software is a living thing. It’s almost like water, right? People forget that water is more than just a bunch of molecules sitting in a tank or a pipeline somewhere. If you let water sit static, it grows bacteria. It gets stale. It tastes bad. It kills you, right? And water is quite a powerful thing. And if everything runs on water, and you don’t take care of the water itself and the pipes that deliver the water, you’re going to have a problem. And so software is the same thing. Software is a living thing, you have to maintain it, you have to keep it current, you have to continue to sanitize it. There’s always bugs that creep in. And in the process of doing that, you will end up refreshing the languages which get stale from time to time, because it’s really just part of your continuous investment.
Geoff Seyon 47:36
So if I were an investor, I would look at how many lines of code I have, how many of them were actually written by my team, how many visits attached to from external parties, and look at that as the whole code base, because that’s the complexity of the solution that your team has built. That’s your asset. Then ask myself, “Okay, given that I’ve spent x million dollars to build this asset, which is my product, or my platform that I sell services on top of, how much should I be investing every year? Not to add new features, but just to keep it current? How much should I spend to support it?” Most people don’t realize that that cost could be as much as 30% of what they spent to build it. And the result of that ends up like any other asset, it ends up eroding.
Shawn Flynn 48:27
When a company is building their product, should they look for existing open-source code out there to implement it if it kind of makes things go a little faster at the beginning, or should they do everything by hand in house?
Geoff Seyon 48:40
Another terrific question. If you’re a startup, absolutely beg, borrow and steal everything that you can. That’s the name of the game. When you start selling your product commercially, and you’re starting to get into your first institutional rounds of investment, i.e. series A. Let’s just say you need to figure out who owns those licenses. Some of them are public licenses. And they’re you’re free to use it as long as you credit the source. But it’s really hard to write software today without using someone else’s software, right? You’re putting it on Linux, Linux is open source, how are you going to deploy that? Well, you got to deploy to the cloud somewhere. Okay, it’s Amazon. Ultimately, you are paying for the license or you’re using the license. You just have to make sure that you’re crediting. Where people get in trouble is when they download or they find on GitHub, or some other online code repository. Someone’s cool toolset, right? We all like to find new cool toolsets and use it because it gives us a competitive advantage over the other startup or the other big company that we’re competing with.
Geoff Seyon 49:42
Just check what the licensing arrangement is. This person, he or she that wrote the software expects you to pay them, or at least give them credit if you use the software in a commercial application. And so you have to keep a log of this stuff. That’s every developer’s job and now there are tools which do that, that list them for you, go through, and document everything. But I would say think about it as well, because it’s not so obvious sometimes. Sometimes the credits get deleted, someone will pass it on to you as their code internally, unintentionally, because in the beginning, you’re just trying to demonstrate to initial clients and the investors that something can be built. What often happens is what should not have been used in production, but was just built for demo purposes, ends up getting rolled into production code. And that’s the problem.
Shawn Flynn 50:35
Geoff, what are you working on right now?
Geoff Seyon 50:37
So I’ve been working in the area of artificial intelligence companies, mostly in the digital health space. We’ve leaned now into quite a few companies in the FinTech space as well because there’s some interesting problems there that AI can solve. So what I spend most of my time focused on and thinking about is this called explainable, artificial intelligence and that’s our focus at my venture studio, STUDIO IXORA, which is spelled IXORA. It actually stands for interactive explainable orientation aware AI. And what that means is that there are very few object models, fundamental platforms that people write software on today, where it is required that you document what is the code trying to do. And in the field of AI, that’s becoming more and more important. So even if somewhere in the interior of your old toolkit of your artificial intelligence solution, you’re going to use one of these deep learning black boxes. You have to be able to explain what parameters you put on it, so that the person getting the result knows how to audit it.
Geoff Seyon 50:55
So let me give you an example of that. This example has been in the news lately. If you are going through thousands and thousands or millions of cases and trying to determine in the criminal justice system, which cases should come up for appeal, which cases should have extended sentencing? Which cases, you have to look at over again entirely, because they might have been some facts that changed that will cause the verdict to be overturned? You have to be able to audit every single change from the original decision. And you have to be able to explain what particular laws, not the laws back then, because the laws would have changed. You actually had the machine read through it because no human is going to do that in the future. Maybe they’re doing it today. But the systems that I see being built in the legal space, they’re looking and crawling and they’re creating these semantic meaning networks that adjudicate essentially the legal structure, because the way people write law, it’s very analogous to code, believe it or not. And most of the time, there is some ambiguity which can be tolerated, but it’s pretty unambiguous once you understand what the decision was. But if the decision changes, why did the change, did the circumstance change? Did someone appeal it? Was it just the facts that changed? So being able to explain that before you sentence someone is important. And then when I say sentence, then you have to say, “Okay, how much time do you get? Is it four months? Is it six months, right?”
Geoff Seyon 53:27
Classic case is this recent thing with the college scandals, right? One person got two weeks. What’s the other person going to get? What’s the standard of fairness? Well, eventually, we’re going to start using AI for some of this type of stuff. We got to know what standards we use before the ultimate decision is bubbled up to the judge, the jury, the lawyer making the decision, before they as the human hopefully make the final decision. So the AI has to be able to explain where it got its answers and not in books and books of paper, not 12,000 web blogs. It has to be human readable summary that can be parsed as if you had a paralegal filling it out for you. So that’s what explainable AI is about. And that’s where I focus my head. So I invest through a small family office in companies that fit into this space. So the venture studio model is not that we raise a fund, and then we deploy the fund. It’s that we have a small portfolio of companies that align with this thesis, will align with the direction here that we’re passionate about building. We’re passionate about partnering with them. And we have developed an ecosystem of friends, supporters, other investors here in Silicon Valley. Once the companies get to a certain stage, they go ahead and they support us in adding additional capital to the round. And then, you know, hopefully, the company is off and away.
Geoff Seyon 54:45
We’ve had quite a few successes that we’ve influenced recently when one company went public in July, we had a hand to play in bringing the commercial side of the clients in healthcare to that company. And we’re working on a few more companies right now and we’re raising a second fund. Again, these are small funds, this is order of $10 million. And because we’re using a lot of the global labor, as well, as a small team of us, it’s four people and, you know, half a dozen supporters around here, we’re able to leverage up and deploy that capital a lot more efficiently than the traditional Silicon Valley model, by using the global footprint. I obviously have to practice what I preach, right? So for us GDI, globally distributed innovation.is not just a concept, it’s a way of life. And what we build out of it is hopefully going to be a perpetual institution that will continue to pay back not only into the excitement of what we’re building in the studio, not just furthering the understanding of how explainable AI needs to work, but fostering an environment that works for women and minority leaders. These are the things that we’re passionate about.
Shawn Flynn 55:50
And Geoff, if anyone wants to find out more information about you or what you’re working on, what’s the best way to go about it?
Geoff Seyon 55:56
Best way is email, I share my calendar openly. gseyon@gmail.com, and I have a publicly available calendar which is https://gseyon.youcanbook.me/
Shawn Flynn 56:11
We’ll have that information in the show notes. And Geoff, is there any last thing you want to leave our audience?
Geoff Seyon 56:17
I think this is terrific. And I really like what you guys are doing. I hope that the journey of Silicon Valley continues. I hope that we continue to evolve. And I hope that more and more generations from all around the world get to participate because of what we’re doing here.
Shawn Flynn 56:33
Great. Geoff, thank you once again for your time today on Silicon Valley.
Geoff Seyon 56:36
Thank you, Shawn.
Outro 56:38
Thank you for listening to The Silicon Valley Podcast. To access our resources, visit us at TheSiliconValleyPodcast.com and follow our host on Twitter, Facebook, and LinkedIn @ShawnFlynnSV. This show is for entertainment purposes only and is licensed by The Investors Podcast Network. Before making any decisions, consult a professional.
