
Sign up to save your podcasts
Or


Every pollster knows the problem: for every hundred voters you ask to take a poll, only one or two actually respond. So what if AI could just answer the questions for them? It's a seductive idea — and Ben Leff has put it to the test.
Ben is co-founder and CEO of Verasight, a survey research firm founded by academics and trusted by leading institutions and media organizations. He and his colleagues G. Elliott Morris and Peter Enns recently published a series of papers asking the question the industry is buzzing about: can AI digital twins replace human survey respondents? Their verdict, after rigorous real-world testing, is a firm — for now.
In this conversation with host Eric Wilson, Ben breaks down exactly what synthetic sampling is, how Verasight built and stress-tested it, and where it consistently fell apart. Ben sees a narrow lane where synthetic data could earn a legitimate spot in the toolkit — as a directional pre-screen, a quick and cheap starting point before committing to a full sample. But for anything requiring precision in close races, the humans still have to show up.
Visit our website: CampaignTrend.com
By Eric Wilson5
3737 ratings
Every pollster knows the problem: for every hundred voters you ask to take a poll, only one or two actually respond. So what if AI could just answer the questions for them? It's a seductive idea — and Ben Leff has put it to the test.
Ben is co-founder and CEO of Verasight, a survey research firm founded by academics and trusted by leading institutions and media organizations. He and his colleagues G. Elliott Morris and Peter Enns recently published a series of papers asking the question the industry is buzzing about: can AI digital twins replace human survey respondents? Their verdict, after rigorous real-world testing, is a firm — for now.
In this conversation with host Eric Wilson, Ben breaks down exactly what synthetic sampling is, how Verasight built and stress-tested it, and where it consistently fell apart. Ben sees a narrow lane where synthetic data could earn a legitimate spot in the toolkit — as a directional pre-screen, a quick and cheap starting point before committing to a full sample. But for anything requiring precision in close races, the humans still have to show up.
Visit our website: CampaignTrend.com

2,435 Listeners

111,948 Listeners

58,066 Listeners

15,950 Listeners

720 Listeners

12,143 Listeners