July 15th, 2026
Aging clocks can be built from any sufficiently complex set of biological data measured in a sufficiently large number of people across a sufficiently large range of different ages. Machine learning techniques are used to find algorithmic combinations of data points that predict age to some sufficient threshold of accuracy. The algorithm is then applied to people who were not in the original sample populations, and most such clock algorithms do an acceptably good job of hitting the mark when considered over groups of people. Unfortunately they are not all that useful for an individual; in part the variance is a problem, but the main challenge is that it is entirely unclear in most clocks as to what the results actually mean. It is also unclear as to how we should expect any given clock to react to any given intervention used to treat aging.
The best path forward to making aging clocks useful for individuals, and for the assessment of novel therapies to treat aging, is probably to collect as much data as possible and observe the emerging patterns. Classes of therapy will have to be assessed in parallel with clocks. Different populations and different strategies for clock use will have to be assessed against actual outcomes, such as mortality rate and disease incidence years later. This won't be a fast process.
