Centaur learning is as much about teaching people skills as it is about giving computers skills.  Centaur learning is about balancing the strengths and weaknesses of humans and machines.  The centaur learning model suggests humans and computers should not be rivals. Instead, they are complementary. Each has something the other needs.
‍
Humans have the ability to judge when information is relevant to solving a problem and when it is not.  Computers have phenomenal memory. They store vast amounts of information in a form they can quickly access.  A computer also has superior ability to find patterns. It can spot connections between disparate bits of information.

Centaur learning implies a way to replace a reliance on narrow artificial intelligence.  Such artificial intelligence (AI) -- the kind behind IBM's Watson computer, which won the television game show "Jeopardy!" -- specializes in narrow tasks like playing chess or driving cars. Taken together, human intuition and computers' memory and processing speed combine to allow both people and machines to participate in centaur-type learning.
‍
The centaur learning model also may point the way to better learning software. Much of the software in use today, such as Web-based tutorials, assumes one-sided learning. Software tutors tend to be more like teachers than like partners. They dispense information and ask users to respond.

The centaur model also suggests how computer tutors might operate. They would incorporate elements of game software, using emotional rewards and punishments to maintain motivation. The tutors would ask questions, ranging from the mundane to the profound. The computer would assess cumulative knowledge.

The term "centaur" comes from Greek mythology.  The half-man, half-horse creature symbolizes a combination of intelligence and physical skill.The goal is to produce software systems able to learn from data.
‍
The researchers are being encouraged to develop new algorithms that can learn from mistakes.  They are also encouraged to find new ways to match the strengths of humans with the special abilities of computers.   As an example, they said computers have the ability to detect weak signals in noisy environments, while humans have the ability to interact with the human brains around them.The goal is to produce software systems able to learn from data.

We learn by interacting with people and computers. We come to form a reliance on people and machines through this process of learning.

We use computers and people to complement each other's strengths. AI speed is faster than human on complex computational problems, but less than humans on complex recognition problems.

These complimentary strengths are most valuable when leveraged in novel situations, complex tasks, and dealing with incomplete knowledge. Understanding the tradeoff between open-endedness and accuracy of results, humans can better understand usefulness and limitations of the cognitive 'black boxes' of thinking machines.