2 ms·
Submitted a couple of days ago from the divulgative article "Creating artificial intelligence that acts more human by 'knowing that it knows'" (from TechXplore.
by mdp2021 4y ago
Submitted a couple of days ago from the divulgative article "Creating artificial intelligence that acts more human by 'knowing that it knows'" (from TechXplore.com), https://news.ycombinator.com/item?id=31598150 https://news.ycombinator.com/item?id=31598150
Annotated abridgement for that submission:
>>>
Finally, an ANN that directly tackles the matter of Intelligence (given that many systems in the past few years created a sinister confusion between the concept of Intelligence and one of its direct opposites, "parroting"), namely in the indirect implementation of one strict requisite: metamemory, as part of metacognition.
Incidentally, one may like that the researchers are Japanese - as they have often brought fresh air with their perspectives and contributions.
> Metamemory [the ability to remember things... once heard or learned] is the process by which ["]we ask ourselves whether we remember what we had for dinner yesterday and then use that memory to decide whether to eat something different tonight[" ...] It involves a person having knowledge of their own memory capabilities and adjusting their behavior accordingly
> When studying metamemory, researchers often employ a "delayed matching-to-sample task" [the use and also the evaluation of the memory of past experience, through the choice whether to have the memory tested for possible reward]
> Despite starting from random neural networks that did not even have a memory function, the model was able to evolve to the point that it performed similarly to the monkeys in previous studies. The neural network could examine its memories, keep them, and separate outputs
> Professor Takaya Arita (lead author): «The need for metamemory depends on the user's environment. Therefore, it is important for artificial intelligence to have a metamemory that adapts to its environment by learning and evolving ... The key point is that the artificial intelligence learns and evolves to create a metamemory that adapts to its environment»
Original research article: Yusuke Yamato et al, Evolution of metamemory based on self-reference to own memory in artificial neural network with neuromodulation, Scientific Reports - https://www.nature.com/articles/s41598-022-10173-4 https://www.nature.com/articles/s41598-022-10173-4
> Based on the formulation by Nelson and Narens... metamemory is regarded as a complex process in which two cognition layers at different levels of hierarchy are associated by monitoring and control [Graphically: https://media.springernature.com/full/springer-static/image/ https://media.springernature.com/full/springer-static/image/... ] [...] Their model explains that the monitoring mechanism is achieved as follows: the meta-level monitors the state of the object-level, updates the internal model including the knowledge about its own cognition based on the result of monitoring, and determines the meta-level policy based on the state of the internal model. According to the secondary representational structure and its flow of information, the evolved mechanism of metamemory can be reinterpreted as follows[: t]he part considered to store the input pattern corresponds to the object-level, the part observing the state of the memory part, including the core modulatory neuron and some connections of which weights are modulated by the core modulatory neuron, corresponds to the meta-level, and the state of the core modulatory neuron (meaning the presence of the memory content) corresponds to a model of the object-level inside the meta-level