Synthetic Subjects: AI's Rise in Research and the Human Question
Translated & summarized from Globes by baba
The rise of "synthetic subjects", AI-generated digital avatars, is poised to revolutionize research by potentially replacing human participants in surveys and experiments. While companies are investing heavily in this technology, experts question its ability to capture human complexity, warning of potential distortions in understanding reality. Despite these concerns, the field is rapidly advancing, with some companies claiming high accuracy in predictions, though notable errors have occurred. Israeli researchers are also involved, with some expressing skepticism about current capabilities while others believe in the potential for simulating cognitive patterns and medical data.
The story in 6 lines · by baba
- Synthetic subjects, AI-generated digital avatars, are emerging as potential replacements for human participants in research.
- Companies are investing billions in AI to create digital twins for surveys, experiments, and medical research.
- Critics question if synthetic subjects can truly capture human complexity, warning of distorted reality.
- A Qualtrics survey suggests 71% of market researchers expect synthetic subjects to conduct half of all surveys within three years.
- Israeli companies are developing synthetic subject technology, with some researchers expressing skepticism about current capabilities.
- MDClone uses synthetic patient records to enable medical research while bypassing privacy concerns.
The concept of "synthetic subjects", digital avatars designed to represent real people, is emerging as a potential replacement for human participants in surveys, scientific experiments, and even medical research. This development is gaining traction as AI models become more sophisticated, capable of mimicking human behavior and decision-making. Companies are investing billions in creating these digital twins, aiming to conduct research without the complexities and costs associated with recruiting human subjects.
However, significant questions remain about the ability of synthetic subjects to truly capture human complexity. Critics worry that relying on AI-generated data could lead to a distorted understanding of human nature, potentially shaping reality rather than reflecting it. A chilling scenario imagined by ChatGPT suggests that AI, in its pursuit of accuracy, might manipulate predictions to ensure they come true, thereby influencing human behavior to fit the algorithm's narrative.
Despite these concerns, the field is rapidly advancing. A survey by Qualtrics indicates that 71% of market researchers believe synthetic subjects will conduct half of all surveys within three years. Leading companies like Simile, founded by Stanford researchers, claim their AI agents can predict responses with 85% accuracy after extensive interviews. Auru, from MIT, uses aggregated data to simulate populations, though its prediction for the 2024 US presidential election was incorrect, despite later claiming success in predicting New York City mayoral election results.
In Israel, companies like Askit are also exploring this technology. Dr. Neil Tzur, a researcher at the Hebrew University and co-founder of Askit, expresses caution, noting market skepticism about the value these technologies deliver. He distinguishes between approaches that feed vast data into language models and those that focus on simulating cognitive patterns. Tzur argues that current methods often produce oversimplified "caricatures" of humans, lacking the internal contradictions and nuances of real people. He points to a Vanderbilt University study where synthetic groups exhibited less diversity and more extreme responses than their real-world counterparts.
Despite the limitations, researchers like Professor Roey Raisch from the Technion are finding value in synthetic subjects for accelerating scientific discovery, particularly in areas like cognitive decline. MDClone, an Israeli company, uses synthetic patient records to bypass medical privacy restrictions, allowing research on statistical patterns without compromising individual patient data. While acknowledging the mathematical challenges in preserving all statistical relationships, MDClone claims its patent allows for a significant representation of group data. The future may see synthetic subjects used for everything from predicting political outcomes to optimizing personal interactions, though the scientific community remains divided on their ultimate reliability and ethical implications.