Progress in Artificial Intelligence (AI) just doesn’t seem to stop with researchers figuring out unique ways to integrate the technology into everyday things. Now, a New York-based startup recently incorporated AI into a messaging app aimed at “improving relationships with others”.
The messaging app dubbed ‘Mei: Messaging with AI’ features a built-in AI meant to predict a user’s personality and analyze their relationships with others via their SMS messages.
Apart from analyzing relationships, the app also tells users their compatibility and closeness with the other person by studying both their personalities and providing a better picture of how the user and their friends interact – making for great AI integration-based statistical analysis.
In an exclusive interview with Business Recorder, Mei’s CEO Es Lee explains in detail what makes the app so unique, while providing interesting insights into the future of AI and its impact on life as we know it.
Q: What was the idea behind Mei and what is the target demographic?
EL: Our team has been working on the app for almost three years now. We started with another app called ‘Crushh’ that basically read your text messages and told you how much people liked you and measured the balance of a relationship. It took into consideration a lot of things within the conversation, which I call the body language of text messages, like how long somebody waits to reply, how frequently they respond or text you, how emotionally charged they are. We always knew that the initiation with Crushh would expand beyond that.
With Mei, launched only about two months ago and currently in beta version, we already have over 15,000 beta testers. We used a lot of data and feedback gathered from Crushh to find better ways to understand people. So our mission is to use AI and data to improve people’s relationships, which we think feels critical now more than ever because of the many apps out there designed to connect you to only new people or create very transient relationships.
Q: Humans can detect personalities too, so how would you say that your app has more practical uses?
EL: Everything that we do is meant to be a tool, something that augments or helps a person. There’s a lot of things that are pretty obvious, like you know your best friend better than anybody else and you don’t need our algorithms to tell you that she’s a trusting type or loyal. You probably already know these things, but our app actually looks at personalities across 30 different dimensions and I am sure no matter how well you know somebody, there are still aspects of their personality you can’t tell from salient things in your conversation such as little texting cues. For example, the number of times somebody uses the word ‘I’ instead of ‘you’ is very indicative of how they feel the power dynamic of a relationship is. It is literally impossible for humans to count how many times you mention ‘I’ vs. ‘you’ or ‘we’ within a conversation and feed that into an algorithm so mechanically there are some things that are limitations for humans. Also, humans don’t have the benefit of having the data set that we have where we analyze millions of relationships. So, Mei’s just another tool for you, even if it might reinforce a lot of things you already know.
Q: After the introduction of apps like WhatsApp, Messenger etc., people have shifted to them instead of depending on text messages. What is your view?
EL: It’s a big space and I think there is a use and a need for these different apps. That being said, we would like to consolidate – like I would love to have just one app instead of having to split between several and remembering which conversations I had on Facebook vs. WhatsApp. But, I think there is something to be said for the primitiveness of text messages that actually is very elegant.
It forces us to condense our thoughts to 140 characters, the same way Twitter does. I think with these other messaging apps that now allow you to go and type on a computer – it dilutes the interaction, making it lot more easy for people to spam, copy and paste – so we love the fact that text messages are still around, even if they may not be used as much as WhatsApp and Facebook.
Q: Don’t you think people would believe it to be an invasion of their privacy?
EL: Privacy is always on top of our mind. We realize that there is nothing more private than your collection of personal texts to people. We respect that a lot and we have done a lot of things to make sure that people will feel comfortable sharing this data with us. Before we even allow you turn on the AI, we pop up this message that we have to collect your text messages to understand you to build a profile.
A lot of people aren’t comfortable with us so we warn if you’re not, don’t go any further, this is not for you. But even if they don’t turn their AI on, they have a fully functioning text messaging app to replace their standard version. The AI will ask personal messages to figure out the things about you that you like you’re a little more laid back and more cautious. We realize what some people are thinking in their minds like ‘hey I am already afraid of Alexa listening to everything I say at home, here’s another company that could potentially abuse that’.
We are a small startup in New York. In the two and a half years that we’ve been operating, we haven’t made a single cent, we haven’t sold any data. And in fact we have put out some of this data because we think that people should really be aware of what capabilities the large firms have with data. We have actually shared our data with universities having completely de-identified and anonym-ized so no one could figure out whose messages these are, in the hopes that there will be more research dedicated to this and they would bring it all to the public and in a way that makes people comfortable.
Q: Since this app is currently available for Android users only, do you plan on expanding it to iOS?
EL: Yes, we are internally testing the iOS version and the beta should be on by the end of the year. We are focusing on getting the Android version right and then it’s much easier to port the learnings that we have from Android on to iOS. Also, iOS is a little more restricted, they don’t allow apps to access the text message history which makes it tougher for us to provide value right away.
We are releasing one feature at a time and observing how it’s received. We have a lot of features that are in the project pipeline and we want to give adequate time for those to be tested, but it will be ready within the next month or so.
Q: The app is only restricted to English for now, creating a language barrier for users who communicate in their native language. How do you propose to address this?
EL: Right now anybody with Android in the world can download the app and turn on the AI. First limitation is that it’s just English, which is a language barrier for now, but actually you are assisting in the development of different languages because our algorithms can be language agnostic, meaning if we have enough data from you in certain languages and you help us label certain things, we will be able to, overtime, use the said data to help build features. Right now, the personality tool is wrong [for you] because it’s probably looking at the majority of your messages that aren’t in English, but if you actually label – there is a setting in there that allows you to change your personality traits on what you think you are. When we have users like you that do that, we will eventually be able to do that for a non-English language. So after we launch and scale, which we are doing pretty quickly now, we will have different languages very quickly.
Q: The app is still an AI and a machine. Do you think people would be trusting enough of the app to improve their relationship with others?
EL: Yes, because what the app needs you to do is just be honest with the AI and that is a big assumption because with Crushh, we found that when AI asked a question, people are like ‘I don’t want to give away personal information’. So a lot of times they just lie. Like when there’s the option to ‘pick male or female’, often the user picks male even though they are female. We found out that people lie a lot, but once we gave them information based on what they gave us, they then immediately go back and say ‘I kind of lied about that and I am really a female, not a male’. This is meant as a tool to understand them, so only if we understand you, can we better cater our analysis and the algorithms. We really try to de-incentivize lying because there is really no benefit to the user from giving us false information. So we are being a little trusting and hopeful that people will kind of do the right thing if we get across the message that we are not doing this as an invasion of privacy and we are not looking to sell the data.
Q: What led you to base the app on the Big Five personality factor model used for psychological testing?
EL: We picked this system as there’s a lot of scientific research behind it. The Big Five has been used for decades with a lot of published research, surveys, and labeled data from psychology. We aggregated from a lot of different data sources, consulted with psychologists and sociologists and we picked it because there was just so much research and data behind it. It has always been like stress-tested for 30 years, so we thought why would we even try to challenge it or try something different?
Q: What do you want to achieve with this app? How do you see the future or Mei say in the next two to three years?
EL: First and foremost, I would like to educate people. I know people lie about stuff and I think there is so much still to be learned just about people. I think two to three years later I would love to have a lot of our findings being used in scientific studies. I would love to get this data available to the public so that people can understand people who aren’t like them. Just getting this data out for creating tools that try to give this information would be the goal.
I guess it’s probably worth mentioning that we are going to be introducing a polling eco-system and this speaks to the point that I mentioned about you owning your own data. There’s knowledge that you have that strangers might not have access to, but from your text messages let’s say I can tell you’re a reporter because you talked about ‘when are we going to press’ or mention frequently a story or an article.
If we have this information, and our users opt to be part of that polling platform, they could be connected with people that want to ask them to share their very personalized knowledge and expertise. And with the credit or token system, there’s a way you could dream that one day somebody could make a living from sitting at home and answering questions. It is something a little far out there, but I think is a possibility two or three years later.