AI in the Series “Start-Up”
Warning: May contain spoilers about the series “Start-Up”.
“Follow Your Dream.” — Park Chan-ho
A story about following your dreams, overcoming hardships, putting ideas into action as well as a large dose of camaraderie and love. “Start-Up” is a South Korean television series which follows the team of Samsan Tech through their journey as a tech start-up.
This is not a series review so it doesn’t matter if you’re #TeamGoodBoy or #TeamDosan. (Although I am a #TeamGoodBoy myself)
We’re here today to explore some of the AI technology present in the series “Start-Up”. Let’s get started!
Object Detection
The first piece of AI tech we see in the series “Start-Up” is object detection. The software gave us a good laugh when it decided to detect Dosan’s dad as a … toilet.
Object detection, as the name implies, is
The process of detecting different objects in an image or video.
This is shown in the picture above where 2 bicycles, a person, and a car are detected.
So, how does this work?
First, we will have to train the software to recognize different objects. Think of the software as a small child who doesn’t know what different things in the world are and it’s your responsibility to tell them what those things are. For example, a car. You will have to point out different cars in different conditions and after many examples the child will start to be able to tell which objects are cars. Now if you want it to recognize many different objects, you’d have to give it lots of examples to learn from!
Then, how does it know where in the picture is the object?
It starts from when you give it examples to learn, you’d have to tell it where in the picture the object is. You can do this by drawing a box around it like in the picture above. That is also why when it predicts things you will see a box around the predicted object as well.
After you’ve successfully trained your software, it should be able to predict what and where different objects are in your image or video. However, it can only predict objects which you give examples of. Say, if you only gave it images of cars, it won’t know what a bird or a bike is.
AI Assistant — “Yeong-Sil”
Actually, maybe the first AI tech we saw in “Start-Up” is the AI assistant, “Yeong-Sil”. Similar to the other AI assistants we see in our daily lives, Siri, Alexa, Google Assistant, and Cortana, “Yeong-Sil” accepts commands or questions from Han Ji-Pyeong (the user) and then gives a response.
The AI technologies involved in this process are voice recognition, voice analysis and natural language processing.
When you first set-up your voice assistant, it normally asks you to say certain phrases multiple times. This is in order for it to be able to recognize your voice.
After that, whenever you ask a question, your voice will then be converted into a digital signal that can be analyzed by the software. The software will then convert the signal into words and then your question/command will be compared with the database of the software, which is kinda like its own (very large) knowledge storage, and will try to find a suitable answer for you.
The databases are located on the cloud (not in the device) which is why you normally need to be connected to the internet in order for it to work.
In 2018, Google showed off its assistant being able to make appointments for you, this can be seen here:
With more data and people using it, the AI assistant will be further improved and hopefully will be able to surprise us even more in the future!
Generative Adversarial Network (GAN)
In episode 5 of “Start-Up”, Dalmi’s sister, In-Jae pitches her team’s idea of font generation. Creating new fonts by using AI without the need of humans at all.
The AI technology used to achieve this is called, you guessed it, Generative Adversarial Network which we will now refer to it as GAN.
Now you may be pondering how does the process work? How can you create new fonts so easily? What is this wonderful thing called GAN?
Let’s start by learning how it works.
There are 2 main components of GAN, the generator and the discriminator. Using fake bills as an example, the generator or the thief in this case will try to create fake bills for the discriminator, the police, to discern fake ones from real ones. If the police manages to detect the fake bills then the thief will then learn and try to make it more real. This process repeats itself until the police cannot distinguish between what’s real and what’s fake anymore.
Similarly with the fonts, the generator will try to create new ones and the discriminator will decide whether the “fonts” are real or not.
Once the training process has finished, we will only use the generator to create new fonts for us.
And there we have it, that is the basic idea of how they work.
What other uses are there for GANs? They can be used for:
- Generating New Images
- Increasing Image Resolution
- Generating Cartoon Characters
- DeepFake (A person’s face replacing another person’s face in an image or video)
And many more! However, there are ethical issues of using GANs which you may have to be mindful of and consider when playing around with it!
Wrapping Up
Although I said this isn’t a series review, I would still recommend for you to watch “Start-up” if you haven’t already. It’s a fun series with a great cast and a beautiful OST. and also is up to date with AI and its uses nowadays.
#TeamGoodBoy
