The prediction has a two‑year horizon from the podcast date of 31 May 2025, i.e., until around 31 May 2027. Today is 30 November 2025, so we are only about six months into that two‑year window. It is therefore too early to say definitively whether the prediction will hold.

Empirically, as of 2025 the U.S. unemployment rate is in the 4–4.4% range, not remotely close to 20%:

• Monthly data for 2025 show unemployment at 4.0–4.4% from January through September 2025, with 4.2% in May (the podcast month). (theglobalstatistics.com)
• BLS‑based series and Fed/other summaries likewise report headline unemployment around 4.0–4.3% in early and mid‑2025. (economy.fedprimerate.com)
• Forecasts from professional forecasters and economists project unemployment in the mid‑single digits (around 4–4.5%) through 2026, not anywhere near 20%. (philadelphiafed.org)

So so far, the scenario of ~20% unemployment (overall, let alone explicitly “due to AI”) has not materialized. However, because the prediction explicitly concerns the state of unemployment two years after the podcast (by May 2027), we cannot yet confirm it as correct. A sharp unemployment spike between now and May 2027—whether or not attributed to AI—could still falsify the claim.

Given that the evaluation date precedes the end of the forecast window, the prediction’s ultimate accuracy is inconclusive (too early to tell).