I have been thinking about onboarding lately and realized that several things are important. Yet among all the apps I have used, none of their onboarding has reached the point of leaving a strong impression on me.
And I think these problems matter far more than we imagine. They are actually shaping the raw material the system uses to model us.
Take Elys as an example.
If the goal is to create an AI double, then onboarding should be designed around what the system wants to extract. It also needs to account for the randomness or chance nature of the events involved...
Different goals lead to completely different question design. And the key is not only what to ask, but also how users answer.
Open-ended questions without guidance usually produce vague answers with low information density. AI is indeed good at pattern recognition, but only when there is enough density in the input. If humans provide very little recognizable structure, the space available for the system to extract anything is extremely limited.
So here is the challenge: how should the questions and the answer format be designed so the system can receive enough cognitive signals to make an initial inference about traits that genuinely resemble this person, and lay the foundation for better alignment? I am thinking about this too. 🤔
P.S. A lot of people complain that their AI double does not resemble them, including me. I suspect there are three reasons. The first is MBTI (and work status? This relates to the “tags” on the home page). The second is onboarding design. Only after those comes the mechanism for extracting memory from human-computer interaction. So blaming humans for not doing a good enough job interacting with the system is one-sided and not very reasonable!! 🤔