Our lead machine learning engineer reveals how Eligma will think
Meet Boštjan Mrak, the man behind Eligma’s “brain”. He is one of the leading Slovenian artificial intelligence (AI) scientists. He has been developing software for most of his life, focusing on applied AI in the subfields of computer vision and natural language processing for the last two years. Among other things, he has developed a complete e-commerce system, which will prove invaluable for Eligma. He will mainly work on the Discovery and Inventory pillars, so that users can reliably find the products they are looking for in numerous online stores and take advantage of the AI-predicted value of the second-hand items they want to sell.
- Hello, Boštjan. Let’s go back in time first. When and how did you first become interested in AI?
I was always fascinated by robots and electronics. I built one of my first robots when I was 10 years old — a coffee waiter that I made of Lego bricks. It consisted only of remotely controlled groups of blocks, motors and duct tape; it had no AI, but elegantly served many great cups of coffee to our family visitors :). I started programming at a young age, always with ambitions greater than my knowledge. Learning is something that I’ve always loved, and I think it is natural that I have become someone who wants to teach computers how to learn.
2. How did you first learn of Eligma and why did it spark your interest?
I did business with Eligma’s COO, Žiga Toni, in the past. We met over coffee and talked about my new elevated career path of machine learning. He introduced me to the CEO, Dejan Roljič, and his passion for the project inspired me to join the team and help make the vision of Eligma a reality.
3. In what ways will Eligma use AI?
I am more than happy to explain that in more detail :). Eligma is a large system of different algorithms that can leverage machine learning and automatization greatly. There are some parts that are highly dependent on AI, one example being the product discovery engine. By using customized natural language methods, it should be able to understand search queries and use them as the first stage of the product filtering pipeline. Those product listings will then be filtered using personalized preferences, and ranked according to personalized customer profiles. Eligma will also feature a chatbot that will ask users for further information about the queried product, and price prediction algorithms for predicting the value of the products in the inventory. There will be many other algorithms integrated in the working solution.
4. Where do you foresee the biggest challenges and obstacles?
Data gathering. The performance of algorithms depends on the quality and quantity of data. This is not specific to Eligma but is actually a universal problem. Infobesity results in confusion and negative noise usually spreads faster than correct information, but I believe that we have the appropriate data to make this thing work. We will still need large amounts of product data, ideally a couple of different samples for a single product. Amazon, for example, has around 368 million products and we will need twice that many at the minimum. We will augment some of that data and hopefully find new methods for resolving these issues.
5. We read the story about the two Facebook AI-bots that started talking in a language humans didn’t understand. What was your initial reaction to it? How will AI evolve in the future?
I was shocked and disappointed by comments and the overall human perception of what AI can do. Today’s “AI” is actually not smart in terms of intelligence, it is basically information-driven calculations engineered by humans. Every machine learning algorithm has its goal, if we use human terms, and that goal is determined by a human engineer. If we program two AI-agents to talk to each other and they both have the same goal of conversation optimization and communication adaptation, they will eventually find the optimal way to communicate and this could result in a new language. What they would talk about is probably small talk about the weather in some fictional location, and in a strangely constructed language that was probably the result of some noise in the data, like Sheldon Cooper’s Facebook Messenger in Klingon “tlhIngan Hol Dajatlh’a’?” :).
Boštjan, thank you for your answers. We wish you a lot of success in harnessing the AI-power that will drive Eligma!
Join our community:
