Creating artificial intelligence that acts more human by knowing that it knows
Researchers from Nagoya University have made significant progress towards creating neural networks with metamemory capabilities. Through an evolutionary computer experiment, they developed a model that displayed meta-cognitive behaviors similar to that seen in monkeys.
Metamemory refers to the ability to evaluate one's own memory and knowledge. It allows adjusting behaviors based on understanding memory limitations. Mastering metamemory is key to developing human-like artificial intelligence that could be interacted with like a family member.
The research team focused on a 'delayed matching-to-sample task' commonly used to study metamemory. In this task, a subject sees an object like a red circle. After a delay, they must select the previously seen object from similar options. Correct choices get rewarded while mistakes get punished. However, the subject can opt out of the choice to earn a smaller guaranteed reward.
Humans leverage metamemory in this task. If they remember the initial object, they choose to take the test for a bigger reward. If uncertain, they avoid the risk of penalty and instead collect the smaller reward. Prior research found monkeys capable of similar metacognitive decisions.
Using neural network models, the Nagoya group simulated this same scenario. Despite starting with random networks lacking memory, the models evolved over time to perform like the monkeys. They developed abilities to examine memories, retain them separately, and route outputs correctly. This emergent metamemory occurred automatically without human guidance.
Lead researcher Professor Arita explains, “The key point is that the artificial intelligence learns and evolves to create a metamemory that adapts to its environment.” Making metamemory flexible to environmental needs enables broader applications.
This research provides a blueprint for developing adaptable AI with human-like metacognition and memory. The results constitute major progress in the quest for artificial general intelligence and even consciousness. There are still challenges ahead, but the future looks promising.
Source:
Materials provided by Nagoya University. Note: Content may be edited for style and length.