"learning deficits observed in healthy older adults are driven by a diminished capacity to represent and use uncertainty to guide learning"

From a Bayesian, or normal machine learning, perspective: older people have collected more evidence about the world, and thus would normally reduce their learning rate, or belief update amount, for best performance. In this example, though, video games represent a different environment, and the older adults have little evidence about them. However, they fail to have a higher learning rate. Looks to me like an evolutionary adaptation that does not work anymore; the environment did not change that quickly before.

To paraphrase: older people learn more slowly, and this parallels the lesser weight of new evidence when you have more accumulated evidence in Bayesian beliefs; or the reduction of learning rate in machine learning systems.