Of all the aspects differentiating lifelong learning from shorter-term, more specialized learning, perhaps none is more central than forgetting – or, to frame the issue more generally and technically, ”memory access speed deprioritization.” This extended abstract reviews some of the ideas involved in forgetting for lifelong learning systems, and briefly discusses the forgetting mechanisms used in the OpenCog integrative cognitive architecture.
...
Complete entry
Posted by
Pastor_Alex on 01/24 at 12:59 PM
The issue I see with forgetting things like “this idea didn’t work” is that the system is possibly going to go through another iteration of the same or similar idea. That suggests that most things are going to be stored in slow access rather than deleted since deletion will force the re-learning of that particular lesson. The exception might be extremely rare occurrences, but they are infrequent enough that permanent storage wouldn’t be an issue.
Posted by
Jonatas Müller on 01/24 at 05:52 PM
Interesting, it’s like a natural selection of concepts, the ones more used survive, the ones less used perish. However, I wonder if forgetting is really essential. Our memory capacity is incredibly big and “idiot savants” such as Kim Peek could remember nearly everything such as a whole library, songs and life events, without problems locating a specific memory among all of them. So the hypothesis that too much memory gets confusing seems to me to be wrong.
http://en.wikipedia.org/wiki/Kim_Peek
It seems that it is rather the form of encoding the memories that causes their idiocy (processing things too literally rather than abstractly), rather than being confused by the amount of memories already encoded. I never saw an example of someone who had significant cognitive problems because of not being able to forget, albeit the suffering problems with PTSD and traumatic memories, but it’s not clear that there would be increased susceptibility for it, and traumatic memories could be dealt with using certain memory erasing drugs (there are a few of them) and therapies.
While prioritizing certain connections in memory may be useful for behavior, I believe that forgetting is not, or at least that we could forget much less than we do and still function fine. For AI, forgetting should be even less important, as long as there is sufficient memory space outside of the prioritized-access RAM to store memories of less usefulness. When all the space has been utilized, then forgetting may come into play, otherwise prioritizing seems enough.
The justification of our flaws is sometimes used to reject transhumanism, and I find these attempts quite unfounded and wrong, though I’m sure that this post is completely different.
IEET Blog