AI engineers are technicians who extract data, design algorithms, and build machine learning models. They need to be familiar with complex logic such as math algorithms and always be ready to develop and research algorithms. AtDataHunt, we are focusing on research and development of AI algorithms to improve the quality of our services. Today, we interviewed Suho Cho, an AI engineer at the center of the field.
My name is Suho Cho, and I work as an AI Engineer at Datahunt.
You can think of my job as bridging the gap between the 'theory' and 'practical application' of AI algorithms. To do this, I repeat the process of algorithm design, analysis, optimization, and experimentation and evaluation.
Before joining Datahunt, I was working on research in the field of computer vision atLG Electronics' AI Research Center. At that time, I was in charge of developing algorithms that recognize things in the refrigerator and provide additional services.
For example, a system that checks the expiration date of products in the refrigerator and reminds you to throw them away, and if necessary, orders new ones. At that time, I was learning the data needed to recognize objects in the refrigerator.I personally selected various industrial products that could be in the refrigerator, as well as items with unspecified identities and uses, and put them in the refrigerator. Sometimes, I also had to dispose of leftover food after collecting data, which I remember being quite...... difficult.
Asa new employee, I also traveled to CES, an electronics show in Las Vegas. I remember volunteering for a really important role and working hard to develop it. I was nervous about making a mistake at an exhibition where the whole world was watching, but I still remember the thrill of a successful demonstration.
I had a strong feeling that I was working as a small cog in a large company, which is unavoidable considering the size of the company. I decided to leave because I wanted to take more initiative in my work, and to achieve that goal,I considered a startup rather than a large company. I had a lot of doubts, but Datahunt promised me a good offer and an environment where I could pursue my research.
When I worked at LG Electronics A.I. Lab, I was part of a research organization and only did research. After moving to DataHunt, the biggest difference is that I moved from research to development.
My main role so far has been to increase the convenience of applying AI in data processing. To do this properly, I read papers, do research, and aim to find the most effective ways to improve and implement them quickly. This way, I can actually apply my conclusions.
First of all, I am most satisfied that I have the opportunity to take the initiative to do a lot of things. Even people with a lot of years of experience in a big company have a hard time saying that they want to solve a problem when it arises. At Datahunt, we have a more proactive atmosphere from raising a problem to solving it.
In particular, I was very pleased with the way they were able to recognize problems and propose their own solutions. It's true that there's a lot of work, but it's very attractive that everyone is willing to do it.
The second thing I love is the work environment. I think a lot of people have a lot of concerns about the stability of a startup, and you've seen and heard a lot of horror stories.
Datahunt is a subsidiary of FiscalNote, a company valued at $150 million. So the stability is self-explanatory, and crucially, I don't know of anyone who's ever taken a lower financial offer to come here from any of their previous jobs, and I don't know of anyone who's ever taken a lower financial offer to come here, so you're treated very well.
One other thing that stood out to me was the way everyone who was involved in the interview process communicated with me, even though we were meeting as candidates and interviewers, they would say, "We'd love to have you onboard." It's a small difference, but it really shows how much they respect their employees, and it made me feel like I could play an important role in the company. It was a very positive experience.
There are a lot of things, but if I had to pick two, I think it's the diversity and the quality of our data. We have a lot of data from a lot of different projects, and when you have a lot of data and you have a lot of experience in dealing with it, you accumulate intangible values, whether it's the quality of the people, whether it's the quality of the AI models, whether it's the platform that makes it easy for us to do our work, all of those things add a lot of quality to our output.
There are many companies that process data, but I know of only a few that build data with their own platform. DataHunt has a platform that supports a variety of tasks for data labelers. This has helped us gain a significant number of customers.
The vast amount of data we collect is then used to train our AI models to upgrade their performance. The members of our A.I. team are working hard to improve our capabilities through various studies.
I've had the opportunity to experience some of our international competitors' platforms, and I've never found them to be inferior in terms of feature richness or usability, and that's saying a lot considering the development time was much shorter.
I had a lot of trial and error in the process of deploying the AI service, and since I had mostly been doing research before, I didn't have a lot of experience. I was struggling for about two months and dealing with a lot of errors. There were a lot of unexpected situations, and it was a lot of work.But I was proud that at the end of the development, we had a service that we could actually use.
However, creating an AI model is not the end of the story. If it takes an hour for the model to produce results, no one will use it, because it would be faster to do the work manually. Of course, an hour is an exaggeration, but it is the biggest battle against time to reduce every minute.
It's also frustrating when you get a quick result and it's a mess, because it's just more work to fix it. In short, our development is a constant balancing act between time and accuracy. We get feedback from practitioners along the way, and we strive to make it better. The most rewarding moment is when a model we've built after a lot of trial and error actually gets a good response. I feel good about my work when I've invested my time, thought long and hard, andDataHunt's models get better.
Currently, Datahunt has a work-from-home policy. If there were any problems, we would have ended telecommuting at any time, but so far, we haven't had any issues or problems. I'm proud that each member feels responsible and performs their tasks well enough to maintain this system.
Working as an AI developer also means battling with data all day long, and frankly, looking at just one type of data would make my job less interesting. Imagine looking at pictures of puppies all year long. But at DataHunt, we're not just talking about images, we're talking about voice, we're talking about text, we're talking about all kinds of data, and it never gets boring.
I would also like to emphasize that in the nearly one year I've been here, there hasn't been any discord among the members. From trivial questions to directly asking for help, no one is impressed or avoiding. Everyone responds like it's their job. I think it's a very complete work culture.
Even if you're not necessarily an AI engineer, I think if you can't explain why you're doing something, you should start from scratch. They say results are performance, but if you don't have a clear purpose, you're more likely to fail.
A lot of models already have code out there, and it's not a big deal to download it, learn it, and run it according to the manual, but even if you do, you have to know exactly what you're getting out of it. "I need to be able to explain what I did and why I did it," I guess that's all I'd say.
Let's take autonomous driving as an example. We're all familiar with the technological advances inLevel 4 self-driving cars, and even at this point, they're actually commercially available in the U.S. While there are occasional accidents in the news, there are plenty of accidents with human drivers. If you're going to compare autonomous driving to human driving, you're going to need statistics with a large sample size.
I think the key thing to look at here is whether or not people are convinced. If we took the steering wheel out of your car and put autonomous driving in, would you get in? I think more people would not, even though neither the steering wheel nor the AI is perfect.
To take another example, there are many companies in Korea that are trying to revolutionize healthcare with A.I. They approach it with the concept of "assisting doctors." At the end of the day, doctors are the ones who are ultimately responsible if something goes wrong because the A.I. is a bit off. It can be relatively free of ethics and bureaucracy.
To sum up, I think it's more important that people are ready to accept the technology rather than the speed at which it develops. If we can keep up with this social trend, AI will bring huge changes to the world as it is commercialized in various fields.
I'm still not good enough to create an AI that will change the world. But if there is a technology that people need, I hope I can be the person who can contribute a positive impact until it is completed.
When I want to meet the needs of users, I won't just focus on whether a feature is feasible, but will go one step further and think about how to make users happy. To be a user-centered AI engineer, that's my goal.