Science & Technology, UK (Commonwealth Union) – The possible applications of Artificial Intelligence (AI) continue to be in much focus with the possibility of many jobs particularly in the service sector likely to be lost and the possibility of a lesser number of new jobs being created. Experts have encouraged those with concerns about losing their jobs to learn new skills and make additional income-generating financial investments such as stocks and real estate.
Neuroscientists have found ways that exploratory actions make it possible for animals to gain knowledge of their spatial environment with greater effectiveness. The results may pave the way for the formation of improved AI agents that are quicker needing a lower amount of experience.
Scientists from the Sainsbury Wellcome Centre along with the Gatsby Computational Neuroscience Unit at University College London (UCL) discovered that the instinctual exploratory runs that animals conduct are not random. These purposeful measures permit mice to learn a global map effectively. The findings that appeared in Neuron defined the way neuroscientists evaluated their hypothesis that the specific exploratory, measures conducted by the animals like darting fast towards objects, have a significant role in assisting them to gain knowledge on ways to navigate their atmosphere.
“There are a lot of theories in psychology about how performing certain actions facilitates learning. In this study, we tested whether simply observing obstacles in an environment was enough to learn about them, or if purposeful, sensory-guided actions help animals build a cognitive map of the world,” explained Professor Tiago Branco, Group Leader at the Sainsbury Wellcome Centre at UCL who was also the corresponding author for the paper.
The prior studies had seen researchers at SWC note a correlation between the effectiveness of animals gaining knowledge on how to go around an obstacle as well as the number of times they had gone to the object. For this research, Philip Shamash, SWC Ph.D. student who is also the 1st author of the paper, conducted experiments to test the effect of stopping animals from conducting exploratory runs. With the expression of a light-activated protein known as channelrhodopsin in 1part of the motor cortex, Philip had the ability to apply optogenetic tools in halting animals from starting exploratory runs toward obstacles.
Researchers discovered that even though mice took up a long time to observe and sniff obstacles if they were blocked from going toward them, they were unable to learn. This demonstrated that the instinctive exploratory measures themselves could assist the animals know the map of their atmosphere.
In evaluating further the algorithms that the brain may be applying to learn, the scientists engaged with Sebastian Lee, a Ph.D. student from the Andrew Saxe lab at SWC, to conduct various models of reinforcement learning that individuals have formed for artificial agents, and notice which one most closely replicates the mouse behavior.
There were 2 main classes of reinforcement learning models: model-free and model-based. Researchers discovered that under certain conditions mice engage in a model-free method but under other conditions, it seemed like they had a world model. The scientists then implemented an agent with the ability to arbitrate between model-free and model-based. This was not exactly how the mouse brain functioned; however, it assisted them to find out what is needed in a learning algorithm to give details on the behavior.
“One of the problems with artificial intelligence is that agents need a lot of experience in order to learn something. They have to explore the environment thousands of times, whereas a real animal can learn about the environment in less than ten minutes. We think this is in part because, unlike artificial agents, animals’ exploration is not random and instead focuses on salient objects. This kind of directed exploration makes the learning more efficient and so they need less experience to learn,” added Professor Branco.
