Math Makes it Personal

What? Really?

Just this morning, we released a new version of Talk to Me, Goose! And, we’re very excited about this one. Why? Math! Yep. Math!

All along, we’ve been wondering how do we make this more personalized and easier to use as we go? If you’ve been following along, you may recall that when we introduced Merlin, we said that he had the memory of a goldfish. He could provide great suggestions about a topic with a mood and a little bit of input. But, he didn’t remember a thing. The objective was to give him a brain. So, we did that. But, what’s a brain? It’s got to be more than just a memory, right? Remembering everything that you’ve ever said is fine, but what do you do with it?

Going beyond just remembering

We started by remembering the things that were said. That was the first step. And, we kept thinking about how to analyze it and what to do with it, and then inspiration set in. David recalled sitting with a team in a small conference room in Cincinnati years ago and talking about segmentation and geeking out on the analysis required to do it, and then we were off. We went back and created 1536 dimensional vectors of every text input created by users (those who consented…just to be clear) and then we clustered them, and interesting things emerged. We found distinct clusters of things that we talk about in the app. Using those clusters, we found we could surface semantically similar phrases for users to pick without any typing at all.

Why is this important?

Well, the more you use Talk to Me, Goose! the more personal it will become. And, it won’t just be a rigid, rote, repeating machine. It will evolve and be creative suggesting text that sounds like you with some creative nuance added by AI. And, if you use the suggested text to spur Merlin on, you’ll get even more text to use and build from…at speed! Again, AI for Good. Math…isn’t it beautiful?

What’s next?

Context! What if Merlin had even more context about what’s going? We’re thinking about how to give Merlin insight now. How do we hook him into the flow of the conversation a bit, so he can recall from text to text and over time to make suggestions that are relevant over a period of time? We think it’s possible. We just need some more inspiration. Share some of your thoughts below.