Pet Peeves and Personalities Laid Bare by AI
- By Lachlan Colquhoun
- December 10, 2018
In recent years the world has become introduced to Alexa, Siri and even came to know actress Scarlett Johansson as Samantha as the AI in the film Her.
Now, if you have an Android phone, you can download another AI called Mei, and she is a little different from the rest of her siblings.
Created by Harvard computer science graduate Es Lee, Mei claims to be the world's first mobile messaging app with built-in AI.
Today, in her beta form, she is a consumer app who helps out with personal relationships, but with further development, she could morph into a business coach and analyst, or move into the health sector as an aid for people struggling with mental health issues.
Like other AI apps, she might even come in another personality type, to be designed by the user. If you prefer a male drinking buddy type persona or even crave an authority figure in a stern CEO, one day this may be possible just by changing settings.
DISCLAIMER: These, however, are mainly flights of fancy invented by the journalist writing this article, but suggested to Mei’s founder in the course of the interview.
Personality Sensitive
In her current incarnation, once you turn on Mei and give her access to your text conversations (which have end-to-end encryption), she can analyze the exchanges and provide you with feedback on the relationship, working on at least 1,000 words of two-way conversations.
“Mei is your own personal relationship assistant,” explained Lee, who has funded the venture so far from his pocket and that of family and friends.
“She gives you intelligence about yourself and others that you may not otherwise know. The idea is that we build an AI within a messaging app which is capable of analyzing the conversation in real time, giving you insights which are specific to the conversation,” he added.
For example, Mei can suggest that the person you are texting is disengaged and disinterested and that perhaps you might be more considerate in your wording.
The AI also detects personality types. If it detects that the person you are texting is disorganized and spontaneous, it will recommend a different angle of conversation from you than it would if you were speaking with someone highly regimented and organized.
There is a scoring and classification method. Mei sorts the personalities of recipients and senders into five different personality types, with scores across 30 different personality traits that measure people against thousands of others in qualities such as introversion/extroversion and thoughtfulness.
User Training
Like all AI, Mei learns and improves on the volume of data that her algorithm can work with. In her case, this comes from users sharing data about themselves, such as their ages, gender and relationship status.
The algorithm understands people in terms of demographics. Men over 40 typically send shorter text messages, while the messages of younger people contain more emojis.
Mei’s ability to intelligently comment on conversations comes from this baseline understanding of who the users are, what typical users in this category are like, and the extent to which the conversation is under analysis.
“There are 30,000 people out there using this now, and they are telling us more and more,” said Lee.
“We are getting tens of thousands of machine learning labels each day; these help to train our models. We are almost having our users build models for us.”
If Mei detects you might be sad on a particular day, based on your previous texts, she will learn and improve simply by the user responding with a yes or a no.
New Potential
Very early on in her existence, it was clear that Mei had uses beyond the personal and in the area of mental health.
The AI, for example, will be able to identify words people use when they are depressed, or when their usage patterns change.
In this mode, AI has potential as an early intervention tool to help people fight depression.
“Right now, it's a personal consumer product, but we have talked to a lot of people, and most people say ‘wait there is a business use case here,'" said Lee.
“So, if you are able to pick up these things about the person you are talking to, wouldn’t you then be able to understand the intent of people who are contacting you for business. We are also just starting to learn about it, but we think there could be a whole range of limitless applications,” he added.