0:00
41:24
Artificial Intelligence makes a lot of people nervous. That's understandable.
Today we sit down with Ken Fujiwara of Hacarus to discuss why that is, and what this startup is doing to fix it.
As in so many other fields, when comparing AI in Japan and the West, we find that the technology is fundamentally the same, but the social attitudes and business strategies are very different.
Ken is a serial entrepreneur, but running an AI startup was never part of his original plan. He had bigger goals in mind, and we talk about how he plans to pivot back to them someday.
We also discuss Kyoto's booming startup ecosystem and why one CEO has publically stated he wants to destroy it.
It's a great conversation, and I think you'll enjoy it.
Show Notes
The problem with Deep Learning and how Hacarus is unique
The importance of founder's hidden failures
Why Ken left Sony to start a startup
How to know when you need to pivot
Why pivoting is hard in Japan
The integrator business model and why it works in Japan
Pivoting a startup to back to your dreams
The importance of explainable AI
Why you need to know about Kyoto startups
Why one company wants to destroy Kyoto's startup ecosystem
The reason you see so many interesting IoT startups coming out of Japan now
Links from the Founder
Everything you ever wanted to know about Hacarus
Follow them on Facebook
Connect on LinkedIn
Get in touch by email: [email protected]
Transcript
Welcome to Disrupting Japan, straight talk from Japan's most successful entrepreneurs.
I'm Tim Romero and thanks for joining me.
As you can imagine, I get asked a lot about how the Japanese startup ecosystem is different from others and I love that question.
The problem is that people usually aren't really happy with my answers. It seems that everyone wants to hear stories about anime or strange gadgets, or cool trends in gaming, and yeah, there's plenty of that in Japan too, but the things that are really unique and interesting like evocative machines and the integrator model, and the role enterprise has to play in supporting startups, those things take a lot of time to explain to anyone who doesn't already understand Japan, at least a little bit, but they're important.
Today, we sit down with Ken Fujiwara of Hacarus and we're going to look at how Hacarus is using the integrator model to jointly develop AI products with large enterprises. Ken also explains how he had to pivot Hacarus away from his original vision and how he might be able to pivot back to it in the future. We talk about the challenges of pivoting and staying true to your mission, cover a few very good reasons why people don't trust AI, and we talk about one CEO who has made it his mission to destroy a startup ecosystem.
Oh, and near the end of the show, we have a really interesting discussion about the startup ecosystem in Kyoto. There really are some amazing things going on in Kansai, but you know, Ken tells that story much better than I can, so let's get right to the interview.
Interview
Tim: So, I'm sitting here with Ken Fujiwara of Hacarus, and thanks for sitting down with me today.
Ken: Thanks for having me.
Tim: Hacarus is a collection of AI platforms that's targeted both at medical and industrial use but you can probably explain this a lot better than I can, so what exactly does Hacarus do?
Ken: Alright, so Hacarus is basically AI startups and provide AI desk applications for medical, such as AI-enabled diagnosis solutions and for manufacturing industry, we provide digital inspection services, and one of the core differences of our company is that we don't use a mainstream AI technology called deep learning. We use something else.
Tim: I've noticed that, so you've talked a lot about your ability to create AI models based on very small data sets. How does that work? I mean, what exactly are you guys doing, if you don't mind me asking what the "secret sauce" is.
Ken: Sure, yeah, I don't mind talking about the "secret sauces." So, in machine learning, in general, the basic assumption is that you need a lot of data or what we call training data, and these days, people, they use technology called deep learning. How deep learning works is that basically, you feed it tons of data and it can abstract the futures from that data set and it can create the model. Our technology called the sparse modeling is quite different, so it can do the same thing but it's from small data sets. It's been a while in academia run, like, year 2000, we had the one person who incorporated that technology and commercialized it, so that's our core strength.
Tim: Okay, I can see how being able to operate on smaller data sets really opens up a lot of broader commercial possibilities because a lot of people just don't have that much data. So, tell me about your customers. It seems like you're dealing with quite a few different applications.
Ken: We focus on two industries, medical and manufacturing, and most of the time, as you said, they don't have access to a big amount of data. For applications like autonomous driving, there's an infinite amount of data because you let the car drive in the city. However, for applications like AI-based diagnosis for rare disease, basically, you don't have a lot of data. For certain diseases, there are only 100-plus unique data, that's it, that's all we're talking about. So, our customers are pharmaceutical companies who want to make AI-based diagnosis solutions using this small data set. For the manufacturing industry, most manufacturing companies, they do have a lot of data for non-defective products. However, it comes down to defective data, they don't have access to a lot of data, so 1/10,000 production units, there's only one defective, so again, if you want to do defect detection, you need a lot of defect data or non-okay images.
Tim: So, Hacarus is not creating a specific product. It's mainly consulting and helping your clients use this sparse modeling to solve their problems?
Ken: At the moment, yes, so the majority of our revenue is based on what we call contract work or consulting work. So, we come in and we listen to the particular problem that the customer has and we provide a tailor-made solution to that customer. We've been doing more than 100 projects that we are trying to productize our knowledge and package them as sort of a license, so that we can sell it to another customer.
Tim: I think it's interesting looking at Hacarus over the last six years and actually, I want to talk about some of your really interesting pivots you've done, but before that, I want to talk a little bit about you. You founded Hacarus more than six years ago now in 2014, but before that, you've been involved with quite a few startups before, haven't you?
Ken: Yeah, actually, Hacarus is my fourth startup, that's why I look so old right now, so yeah, so before this, I've launched three tech startups. This is not the first time doing a startup thing in my life.
Tim: It's your fourth real startup, but you've also had a lot of other projects you've worked on that never quite got traction, right?
Ken: Yeah. I mean, like, my win rate is almost like 10%, so only one win out of 10 failures, that's my success rate.
Tim: No, see, I don't think people really appreciate that, so I've started four companies as well, but the thing is, I've also had probably 12 other projects that never quite got to financing and full-time employees, and I don't know, I think people overlook the importance of having those.
Ken: Yeah, I mean, there's a known saying that failure is a good teacher to success without failure. You don't get success.
Tim: Yeah, I mean, that's the only way you learn.
Ken: Yeah, of course.
Tim: So, what led you to start founding companies? Because before, you were at Sony and apparently thriving as a productive member of society, so why leave to start a startup?
Ken: To be honest, I never wanted to be an entrepreneur in my life. I was a computer geek when I was a teenager, but luckily, my father was a computer engineer as well, and basically, I was trained by him, so by the time I became 18 years old, I was pretty good at writing code or programming, so I decided not to join Japanese university, and instead, I went to the United States and I stayed in the US from 1995 until 1999. That was the dotcom bubble and everyone's talking about starting their own company, like the next Google or next Yahoo, or next ~
Tim: What city were you in?
Ken: In Los Angeles. You understand the crazy atmosphere back then, and none of my classmates were actually trying to get a job at a big corporation, I was the only exception. I felt sane because I was the only person getting a nice job at a big corporation like Sony and I was considered a complete failure from their perspective.
Tim: I mean, but after you get into Sony, Sony is a great company to work for. Was it just sort of missing that college atmosphere and made you decide now, I want to go back and start a company after all?
Ken: Yeah, so I knew that I was going to launch my own startup in 20s but I had no idea how to run a company, just like everyone else, so I joined a big corporation with the only purpose to learn how to run a company, so yeah, so I stayed there for three years how a big company is operated.
Tim: Then went out on your own.
Ken: Yeah, exactly.
Tim: Your first couple of startups were B2B software, they didn't have anything to do with AI, so what drew you to AI?
Ken: Let me say, AI in the early 2000s wasn't usable. Before deep learning, there were two prior waves but they didn't work successfully, so deep learning is a successful application of AI into the industry.
Tim: It's interesting you say that because I mean, you were kind of hinting at this before but there's very little in AI that's genuinely new in terms of the algorithms and the research, right?
Flere episoder fra "Disrupting Japan"
Gå ikke glip af nogen episoder af “Disrupting Japan” - abonnér på podcasten med gratisapp GetPodcast.
