Standards needed for validating digital twins, says YouGov
Reported by Research Live and Business Insider2 sources ↓
YouGov published a validation study of its digital twins and called for shared industry standards for checking simulated research against real respondents.
Key points
- The study compared 1,008 digital twins with 693 people they were based on
- Panellists said twins’ open-ended answers reflected their views 75% of the time
- For fiction books, twins came within one point of their human counterpart 90% of the time
- Ratings of more than 4,000 entities raised person-level accuracy from 78% to 90%, YouGov says
Sources
Details
Research Live reports that YouGov conducted the project with AI platform OriginalVoices. The twins and the people they were based on received the same new questions; 69% of panellists responded within 48 hours. YouGov assigned confidence levels to the twins’ open-ended answers using OriginalVoices.
YouGov said contextual training data reduced the twins’ reliance on demographic information, and it found no pattern of biased performance by age, gender, race or education. The study did not cover all possible validation questions; further work will examine, for example, how often people repeat their own answers.
YouGov said firms using digital twins or synthetic data should test them against real respondents and publish their methods and results. It also called for shared verification and validation standards for defined use cases.
Written from coverage as of
Images
- Research Live