Using Sequences of Life-events to Predict Human Lives

• Germans Savcisens, Tina Eliassi-Rad, Lars Kai Hansen, Laust Hvas Mortensen, Lau Lilleholt, Anna Rogers, Ingo Zettler, and Sune Lehmann

Questions

Model

How about using the encoder-decoder architecture or just GPT (decoder) architecture? I think there are some interesting things that this generative model will allow.

Personality questionnaire

Unclear how life2vec model is used to predict this. Given a natural language question, how do you connect this to the life2vec model?

Fig 4

Simpler semantic axes? For instance, can you use the axis from low-income to high-income to project all the jobs or health conditions? This can be compared to the actual correlation between jobs and income or common health conditions and income. Or this can be even calculated with locations.

e.g. - http://yongyeol.com/2018/06/19/paper-semaxis.html - http://yongyeol.com/2021/04/23/paper-periodicalembedding.html - https://journals.sagepub.com/doi/pdf/10.1177/0003122419877135

Analogies?

Random idea

Will it be possible to “simulate” some RCT experiments and compare them with the actual RCTs? (maybe you need a generative, GPT-like model)