A MEETING OF MINDS
Yee Yee Htut is mimetic.ai’s first employee. She talks about what inspired her to get into computer engineering and her work on building Evie — your A.I. Scheduling Assistant.
You were there at the beginning, what were those early days like?
It was gruelling, but also incredibly exciting.
Back in 2014, the company consisted of Jin (Founder & CTO), myself, and nothing else — not even an office! We camped out at cafes most of the time and I was constantly afraid the owners would just come up to us and tell us to leave (laughs).
But it was great working with Jin because I was fresh out of school and had zero product development experience and he had loads from his time at places like Starhub and Yahoo. Those first few months, I was just constantly asking him questions in order to learn the things I needed to know.
A.I. assistants with NLP (natural language processing) capabilities was (and still is) an entirely new product category, and neither of us was completely sure what to expect. But we pressed on, fuelled by the rush we got every time our code responded the way we wanted it to. And over time, Evie began to take form.
So how are things different today?
Well, a couple of months ago we moved into offices so that’s quite nice (laughs). The team has also grown and we’re still hiring!
Also, in the early days, Jin and I spent a lot of time thinking about what functionalities we ought to be building for Evie. But today, we have users, most of whom are incredibly supportive and engaged, constantly providing us with feedback — meaning we have actual data driving the product roadmap and defining Evie’s next features.
Evie is an A.I. scheduling assistant. What does it take to build something like Evie?
The term ‘chatbot’ gets thrown around a lot these days. Every day somebody is announcing that they are going to build A.I. chatbots for their business. That makes it seem to the general public that this is an easy thing to do. The reality however is that it isn’t something you can knock out over a weekend with Tensor Flow.
An intelligent machine — like Evie — in the beginning is very much like a newborn child. It has no conception of our world. Performing the seemingly straightforward task of scheduling meetings requires Evie to not only have an understanding of space and time, and social structures (like relationships within organizations), but also the ability to unpack idiomatic speech.
For example, if I said “let’s meet Tuesday”, to a machine that literally means Tuesday at 12am. But humans don’t typically set up business meetings at that time.
Here’s another example: If I replied to an email at 1am on Tuesday and said “sure let’s meet tomorrow”, did I mean Wednesday or, did I mean later in the day on Tuesday? Because for most of us, tomorrow tends to be thought of as something that happens after we go to bed regardless of how late at night it is.
And there are many more such situations where humans intuitively understand another’s intent, but we’d have to break it down and teach Evie how to understand and react appropriately. This is complex because machines don’t learn the way humans do.
So how does Evie understand us?
She does two things essentially. First, she breaks down what we say into the various parts of speech (verbs, nouns, adjectives etc) to understand the structure of the sentence — or the syntax. And while understanding the syntax gives Evie the first clue about what is being said, syntax alone is not always enough to understand the speaker’s intent. For that, you require a semantic understanding of the sentence; or what it means within the the context of the interaction.
For example, if I’m being asked to meet with someone at 12pm and I respond to say “OK, but need to dash by 12:45”, Evie needs to understand that I’m saying yes to the meeting, but that I only have 45 mins max for it so she should set the duration accordingly.
Teaching an A.I. to speak human — did you ever think you’d wind up working on something like Evie?
I’ve always enjoyed learning languages. My first language is Burmese; English and French are languages I picked up later on. I taught myself Chinese because I really wanted to read 步步驚心 (Scarlet Heart by Tong Hua). And today I watch way too many Japanese and Korean dramas, mostly for the plots, but they end up teaching me the language.
Learning a new language requires very similar thought processes to building Evie. So it would seem like the dots have connected for me and building Evie combines the things that I enjoy and care about.
We often hear that there’s a dearth of female engineers. What do you think would get more girls interested in STEM fields?
Interest comes from exposure; you can’t begin to be interested in things you have no experience of.
Here’s where schools are so important in those early stages of education. I got into this because I had the opportunity to participate in a workshop on C programming in high school
You’ve been at this for awhile now. Any advice for young engineers?
Be open to learning. Focus on overcoming the hurdles in front of you, but don’t forget the big picture. Finally, don’t worry too much. Hakuna matata. :)