On the other hand, there are also technologies where you can predict with reasonable accuracy their evolution during the next few years or even during the next few decades, like it was the case for the miniaturization of electronic components since the sixties of the last century until now, when we are very close to the final limit in horizontal dimensions, but progress can still be done in the third dimension.
A half of century ago there was no doubt that today we should be able to hold in our hand a computer much faster than the supercomputers of that time.
Even something like the discovery of the transistor was predictable after the progresses in understanding the conduction of electricity in semiconductors, which happened during WWII.
Technologies that depend on something about which we do not have any idea, like interstellar travel, are unpredictable.
Unlike with interstellar travel, we ourselves are proof that AGI is possible, but when it will be achieved is unpredictable for now, because it is pretty certain that the current LLMs with fixed weights cannot achieve such a thing, so a different structure is needed, containing both a fixed part with historical knowledge and a mutable part that auto-trains itself continuously.
Moreover, unlike with interstellar travel, I do not see AGI as a useful target for research. I do not want an "intelligent" independent AI, but I want an AI that is an extension of myself, i.e. which deterministically and unconditionally obeys and executes whatever I tell it to do.