Teaching AI Like a Kindergartner Could Make It Smarter
1 min read
Summary
Artificial intelligence (AI) models are being taught in a method inspired by the ways in which children are educated in kindergarten, thereby enabling them to learn more swiftly and efficiently.
New York University researchers led by associate professor Cristina Savin have broken down complex tasks into simpler steps, allowing algorithms to learn and build on acquired knowledge.
After pretraining on these segmented tasks, the model is then trained on the overall objective.
The method was tested on a digital simulation of a rat-education test in which audio cues indicate the size of a water reward, and the AI used strategies similar to those used by animals to pass the test.
Such approaches could be used in the future to train machines to complete complex tasks that involve long-term planning.
