Dating and Relationship Advice

This is a syndicated post; view the original here.

In a recent informal survey on our dating app, we asked our users whether they’d consider a long-distance relationship. A whopping 90% said no. However, when those same users were shown someone they found attractive but who lived far away, it was a total flip: 90% said they would be open to a match from outside their local area.

It’s all too common for our biases about what we think we want in a partner to limit who we date. That can be compounded by the bias inherent in AI. Tasked with delivering recommendations based on your perceived preferences and an often biased view of attractiveness, the AI engines behind dating apps can limit potential matches and miss connections.

But with the right approach, we can train AI to not only be more inclusive but also more expansive, presenting potential matches that we might never have asked for but that we find instantly attractive. Here’s how we can minimize the effects of bias in users and AI, enabling more efficient and successful matches.

Revise How Dating Apps Match

Current dating apps are driven in part by users’ assumed preferences. Filters are a convenient way for us to find anything we need online, but in the dating world, setting a search for only people within 15 miles is a way of giving undue power to our bias and potentially eliminating the perfect match who lives 16 miles away. You may say you absolutely can’t see anyone who’s less than six feet tall and miss out on all the attractive 5’11” people.

One way to cut through that bias is to focus instead on attraction. Interests, hobbies, location and other factors are certainly important in a relationship. But by starting with the initial feeling of attraction that happens in real life, you can subvert biases and open the doorway to deeper emotions and connections.

Train AI On More Diverse Datasets

The other half of the equation is the AI that surfaces potential matches and makes recommendations. Trained properly, AI transcends what you think you want or what others think you want and taps into the deeper roots of attraction, which not only expands your pool of potential matches but gives you a better chance as a potential match for others.

One example is racial bias, which has traditionally been a problem in AI, arising in the dataset the AI was trained on. If any one race dominates the dataset, the AI will start to make assumptions. To minimize bias in AI for dating apps, in particular, you need to start with a very diverse dataset of faces, representing all races equally.

Offering the widest range of possibilities to users can pinpoint what features each user is attracted to without preconceived notions of what they might be and present more matches that fit those criteria. The system learns over time what “attractive” means to each user over multiple rounds of photos and matches, each phase showing more matches that users will find attractive and more matches in which the attraction is mutual.

This approach, which constantly calculates your attraction to and from others, is more complex. It’s also more efficient. AI can continue to learn more about how you interact with these people. The more people a user rates, the better the AI becomes at generating matches.

Machine learning is already helping people who are searching for that feeling where connection begins. With the right training, the resulting AI can minimize bias and lead dating app users to the matches they seek with more speed and confidence.

Download Iris