Two researchers can look at the same progress and give dates thirty years apart without either being irrational. The divergence usually starts before any data is discussed.
The four sources of disagreement
Definition. One means human-level performance on most tasks, another means the ability to do AI research, a third means a system that can replace most workers. These arrive on different dates.
Model of progress. Believers in smooth scaling extrapolate the recent curve. Believers in paradigms argue that the current one has a ceiling and a new idea is needed, which makes forecasting meaningless until it appears.
Reference class. People who study the history of AI remember the winters. People who study recent capability curves remember the last four years.
Incentives. Fundraising, regulation and reputation all pull estimates in predictable directions.
What would reduce the disagreement
Agreed tests, published negative results, and forecasts that are scored after the fact. Very little of the public debate is structured that way.