Summary: A new puzzle game helps children see when submitted intelligence is still killed. The game reveals Arcu Actph Arks-Visuect Puzzle Pures are simple but difficult for AI-and allows children to compare their responses and Chatbot answers.
Whether AI receives the correct answer, its meaning is usually incorrect, teaching children to ask confidently that they are false. With the test, an error, and guidance, children learn to sink their orders and see AI as a tool rather than approved.
Key facts:
- Tool for visual learning: The arc puzzles teach children how AI treats the attention of the pattern.
- AI limitations are displayed: Children are quick to learn that AI explanations often do not practice.
- Sensitive thinking to promote: Children get confident in asking technology and directly directing.
Source: Washington University
While the current generation of intelligence Chatbots are still basic facts, the systems respond with such conviction that they often convince the most.
Older people, even those who are experienced by the deep knowledge of the domain, continue. But seeing mistakes in the text is very difficult for children, for they often do not have the content of the content of a lie.
The University of Washington researchers have developed a game of Ai Puzle to show children the area where AI systems are still working normally and fails certain puzzles. In games, users get the opportunity to solve the puzzles 'arc's' puzzles (shorts by extract and coorpus consultation) by filling colored patterns.
They can ask a variety of chatbots in AI to resolve the puzzles and have programs describing their solutions – almost till they fail to do well. The team checked the game in two children's groups.
They have found that children learn to think critically about AI answers and find ways to reduce the Systems of better answers.
The investigators presented their findings in June 25 in the children's contact designing at 2025 the conference in Reykjavik, Iceland.
“Children are naturally like arc puzzles and are not specified in any language or culture,” says Lead Ayayushi writer, a student of the UW doctors in the formation of people and engineering.
“Because puzzles are dependent on hiring a visible pattern, even children who can study and can play and learn.
The arc puzzles were developed in 2019 to be difficult for computers but are easier to people because they want to be issued: To view a few examples of the pattern, and use it to a new example. Ai models of the current cuts have developed in the arc puzzles, but they were not arrested by people.
Investigators form AI puzzles with 12 puzzle arc that children can solve themselves. They may compare their solutions to those from various AI discussions; Users can choose model in the drop-down menu.
“Ask AI to explain the” button produces a resolution of a solution text. Even if the program gets a puzzle, its description often neglects. “Help mode” allows children to try to direct AI system to the right solution.
Dangol said: “First, the children gave the plans that were genuinely made,” Dangol said. “Like, 'Oh, this approach is like donut.' Ai model may not understand that a child means that there is a hole in the middle, so the baby needs to come in. Maybe 'a white space surrounded by blue squares.' “
Investigators checked the program in the UW College of Designments on the previous year from more than 100 children from Grades 3 to 8. In these times, 21 years of 21-year-old 6-11 have played Ai puzzles and worked with investigators.
“Kidsteam children are used to provide advice on how to make technology better,” said Co-Eyper Jason writer YP, the professor of UW Assor, the Director of the child.
“We didn't think about adding the help mode of the help, but we talked to children about how we can rescue the puzzles and the idea comes from that.”
By examining, the party found that children were able to detect errors in puzzle solutions and text explanations from AI models. They also see the difference that the human brain contemplates and how AI systems produce details.
“This is an Internet idea,” said another child. “It is trying to settle only on the Internet, but the human brain builds.”
Investigators also detected that as children work in the help mode, they learn to use AI as a tool that requires leadership rather than an answering machine.
“Children are smarter and skillful,” says the Great Julie's writer, UW professor in the formation of people and engineers. “We need to give them opportunities to make their minds according to what he is not, because they can actually see.”
Ruhua Zhao and Robert Wolfe, both students of Corert in the Known School, and a Trushaa Rannahan, a master student in the formation of people and engineering, are also authors working in the paper.
Support: This study is sponsored by the National Science Foundation, Institute of Education Science and Jacobs Foundation's Cerendation's Cerendation.
About this AI issues
The author: Stefan Milne
Source: University of Washington
Contact: Stefan Milne – University of Washington
Image: This picture is placed in neuroscience matters
Real Survey: The findings will be launched in the construction of communication and 2025 conference
