We keep talking about Merlin and how we’re trying to reduce the level of effort, and this morning, I thought, “Why not make a little demo of how cool it is?” 30 seconds of Merlin making magic below. Because, it’s not just about reducing the level of effort. It’s about reconnecting with people. It’s about making everyone feel more connected, and we think that seeing is believing.
What’s going on here?
On the left side of the screen in the “thought bubbles” are personalized common phrases that are built from text that I’ve used frequently in the past. The more that I use the solution, Merlin, in the background is collecting information about the kinds of things that I say and gathering insight about the topics I speak about and the tone that I use.
Periodically, an analysis is performed that determines what are the most common kinds of things that I say and in what kind of tone do I say them. And, with that information in hand, Merlin uses generative AI to surface semantically similar phrases that can be used to seed a conversation without having to type full sentences. These are the common phrases, and they can be refreshed with one click if there’s nothing there that suits me. Of course, I can also just type in the box above to say what I want directly.
On the right, you see me launch Merlin. Merlin takes some simple inputs. In this case, I told him I had an “Everyday Need” and that I wanted to use a “Warm and Friendly” tone (or Mood, in this case…oh, there has been much debate). And, then, I gave Merlin some clues. “I need a blanket.” With that information, you saw Merlin suggested five options each of which had a bit warmer and friendlier tone than just “I need a blanket.” Some of these options were short, some were longer. Each though had a tone that was warm and friendly, and each re-inserted some gratitude and intention back into the request.
Why is this important? Well, caregivers matter too!
As we’ve been working on this, we have realized that the speech generating device is NOT just for the person that is losing the ability to speak. It is for everyone with whom that person communicates. When we are reduced to having to type out everything that we have to say, it’s hard. It takes time. So, we are reduced to saying very little. Just the facts, ma’am. “I need a blanket.”
Now, put yourself in the shoes of the caregiver who has to go get the blanket. Time after time. What does that feel like? Wouldn’t it be nice to be asked? “Hey, when you have a minute, could you get me a blanket?” But, damn. That’s a lot of typing when all I need is a blanket. But, what I really want to say is that “I need a blanket, and if I could, I would get up and go get it myself because you have to do so much for me. And, I am so grateful for you that you have no idea.” But, egad, that’s way too much to say when you have to type it all out a letter at a time and you’re cold.
So, why not let Merlin help with just a few clicks? And, while Merlin won’t get all the emotions right and can’t possibly understand how truly grateful we are, at least he can take some of the pressure off and reduce the stress of the caregiver too. We hope.
Where are we going next?
Well, sometimes Merlin still sounds like he’s writing an email. “I hope this message finds you well and happy…” So, AI is not perfect. The hope is that as Merlin becomes increasingly more personalized and context aware that we can make it so that he realizes that this is not writing an email to customer service…or worse…filing a legal complaint…or writing a thesis. But, he can make us laugh from time to time.
As we try to create more personal, dynamic and context aware suggestions with Merlin, we’re also trying to create more flow. How can Merlin maintain flow from one text generated to the next? Can we make it easier to generate consecutive suggestions that build on prior inputs? Can we make Merlin even more personalized? For example, we realized the other day that Merlin does not even know our names, and there is not a mechanism for him to learn them, necessarily. He doesn’t know his location, the time of day or the weather, but we’ve thought about whether that context would be helpful. What do you think? Drop some thoughts in the replies below. The wheels are turning here, and we’re open to hearing from you.