James Kaplan is the company’s founder and CEO Meet the Kai, an artificial intelligence assistant that makes life easier through conversation, personalization, and curation.
You originally started programming until you were six years old, what inspired you about coding and what languages do you learn?
Momentum was the Oregon Trail game built for Windows 95. I was hooked and played it every day after school. I was in 1st grade, so there wasn’t much else to do! I started thinking about all the things I would like to change in the game. When I was younger, I bought a copy of the “Game Programming for Teens” codebook from a local bookstore. I was immediately drawn to it and quickly forgot the original Oregon Trail motivations. But I always loved games, and they served as my motivation to learn new programming languages. Over the next few years, I learned Visual Basic when I was trying to write a bot for NeoPets (it worked) and then PHP when I was 12 to start creating dynamic websites. Even then, my motivation for learning PHP was to make money buying video games.
Your previous company used an AI hedge fund, what do you learn from this experience?
My biggest grip was to throw away the notion that you can’t compete with giants in a congested space. It can be very tempting to think that everyone in the economy is already being exploited and captured by a giant player. However, I quickly realized that you shouldn’t overestimate your competitors. Laziness and organizational slowness prevent new ideas. At other times, an idea or campaign may be too narrow for a larger company. Surprisingly, they may also think that something is too risky to develop, so there is no reason to go outside their comfort zone.
Could you share the genetic story behind the launch of the MeetKai AI Assistant?
I was tired of funding. Everything had to be measured by Sharpe ratio and PnL. It removed a lot of the fun from the technique, and was far from what originally brought me on the programming path. In late 2018, I spoke with Weili Dai, founder of MeetKain, about what I saw in the technology world in general. One of my main findings was that the voice assistance mode stopped. All key players adhered to the old approaches and users did not benefit from the technology. No one was willing to try new approaches, “because X doesn’t.” There was no difference. If I started from scratch and threw away all (well, most) prejudices about building a voice assistant, I could radically change the user experience. We started building a real artificial intelligence assistant unlike a voice-powered chatbot. The experience gained in my previous project combined with Weil’s mentoring led us to set up MeetKain.
What are the challenges of building an AI assistant?
There are two categories to building a true AI assistant – user expectations and technical implementation. The first problem is ignored, but it applies to MeetKai. Users are trained in the voice assistant on what is and what is not. In particular, they assume that they need to search command-centricly. We strive to train users to search in natural language. This allows for much more versatile features such as the use of the negative “Find Me Dwayne” The Rock “Johnson, which is not Moana.We can deal with it perfectly in everyday speech, but current voice assistants can’t answer us.
The second category of challenges is technical. This is reflected in two subcategories – search and comprehension. In terms of search, we are different from other virtual assistants in the sense that we maintain our content index. While this allows for all the magic that will make us the next generation, it brings with it the challenges of running and maintaining a custom voice search engine. This is an area around which we are constantly innovating. Language skills are another area where the challenges of an AI assistant are met. For most voice assistants, this would mean the ability to understand the English text. MeetKai understands and supports 16 languages. This isn’t 16 times more work because we use multilingual approaches, but it’s still a considerable amount more than “English first, English only”. However, it’s worth the time investment because it’s incredibly important to us that MeetKai is truly global.
How does MeetKai use individual artificial intelligence to differentiate itself?
We use personal artificial intelligence in two different modes – understanding and searching.
When a user says, “Can you find me something Chinese tonight?”, That could mean he wants a Chinese recipe, a Chinese restaurant, or a Chinese show to watch. With our personal, in-depth understanding, we can differentiate and provide the user with anticipation. All this is done without their knowledge ever having to leave our platform.
We bake the personalization for the search itself. One of the biggest problems that other virtual assistants face is that many searches are fed to third-party service providers. When you search for a restaurant with a regular assistant, it forwards the search to Yelp. The downside to this is that Yelp doesn’t know the user personally, and if they do, it’s privacy. Because MeetKai is a first-party application all the way to the end, we have real personalization.
What instances does MeetKai use?
MeetKain aims to be the first artificial intelligence concierge. We want to help users in their daily lives. We don’t want to be a command-line helper. We do not support features such as “volume up”, “volume down”, “30 second timer”,… you get the picture. I strongly believe that those qualities are not artificial intelligence, they are just voice-based input. If you think about the times of AskJeeves, the whole idea was that it was your butler, no one would ever anthropomorphize Google Assistant. As long as Apple Sir or Amazon Alexa would like to be, no one will think of them with anything other than an app. We are still in our infancy, but we built all the necessary technology to implement our 3-5 year roadmap toward a true AI assistant.
How can the industry prevent an AI assistant from setting prejudices or reinforcing existing bias in the user?
There is a delicate balance between providing personal results and creating impartiality. We’ve seen what happens when researchers optimize engagement measurements on social media platforms – it creates prejudices. The first step towards this industry is to think about our metrics and performance indicators. MeetKai optimizes the balance between clickthrough rate and click novelty. Artificial intelligence should be rewarded much more for finding a new result that is clicked in 30% of cases, and not just a “top result” that is clicked in 50% of cases. However, this approach has a rather obvious seizure for deeper consideration. What if the results invented by AI are only biased results for the user’s little bubble? We build our artificial intelligence to push the user’s personalization zone in both directions. If a user requests a list of articles about beef, instead of presenting them with articles that match their beliefs, we can sprinkle articles that are on the edge of their preference or slightly beyond the edge. This may include an article on animal ethics and climate change, as well as an article on the potential health effects of animal fats. Technical intuition is rooted in our approach to accepting that artificial intelligence is difficult to train to determine whether Y is biased, but it is much easier to train it to know that X and Z fall to the edges of the same zone as character Y.
Where do you see the future of AI assistants in 5 or 10 years?
Artificial intelligence assistants are increasingly moving to MeetKain’s first-party approach. Throughout their lives, Alexa and Google have been trying to develop an ecosystem of third-party extensions and skills. This causes serious privacy issues. Moreover, this does not mean any of the upper limits it sets on abilities. I expect more players in the sector to adopt the same approach that we have taken, and there are signs of this everywhere.
Thanks for a great interview, readers who want to learn more about this personal AI assistant should visit Meet the Kai.