Popular science fiction allows people to believe that artificial intelligence springs forth fully-formed, capable of grandiose feats of calculated evil, dictated by inhuman logic. Although, for science fiction writers and movie producers, artificial intelligence is something entirely alien from normal thoughts and feelings, but they couldn't be farther from the truth.
Research into the development of artificial intelligence can have profound implications on what people know about human intelligence and learning, according to George Konidaris, a graduate student in the Autonomous Learning Laboratory, or ALL. Building a robot that can think and learn like a human requires a profound understanding of human processes of learning but can also yield insight into how the human mind works.
"Sometimes we have a stronger connection to neuroscience than we do to the rest of computer science," Konidaris said. His research has concentrated on using "reinforcement learning" to build artificial intelligence for robots. Reinforcement learning - one of the primary areas of interest at the Autonomous Learning Laboratory - is essentially building up skills and behaviors by creating a reward for certain behaviors, like when a child learns to pick up a pot by its handles after learning that picking it up by the base will burn.
"You tell it what's good and what's bad so it learns. It's an agent interacting with an environment," Konidaris said. Konidaris has worked with robots in a number of contexts, including building a robot that used reinforcement learning to figure out how to navigate a maze while at the University of Edinburgh. For Konidaris, creating autonomous robots has a much higher purpose than vacuuming floors, like the Roomba. It is about "understanding the nature of intelligence by attempting to synthesize it."
Reinforcement learning has several advantages over other forms of machine learning; primarily it allows artificial intelligences to develop a greater degree of flexibility and autonomy than simply programming fixed skills. Research into this area has led to a number of interesting questions, and as much as researchers at the ALL turn to neuroscience and psychology for inspiration, they have answered questions and brought new insights into those fields through the trial and error inherent in building new intelligences.
"Thought is very complex from the outside," said Konidaris. "And by starting from the inside the ALL isn't just teaching computers to learn but learning how our own brains function."
Be the first to comment on this article!