AIs will monitor each other in a few years, says NTT vice president; AI constellation is linked by a next-generation network

SAN FRANCISCO – Nippon Telegraph and Telephone Corp. will launch a system of linked AIs in the coming years to stop the spread of false information, NTT Vice President Katsuhiko Kawazoe told The Yomiuri Shimbun. In the system, called an AI constellation, AI produces a response that contains false information or bias, other AIs will indicate there is a problem.

“By avoiding dependence on a particular AI, we want to promote the democratization of AI,” Kawazoe said in an interview with The Yomiuri Shimbun in San Francisco on Wednesday.

The AI ​​constellation will be connected via the Innovative Optical and Wireless Network (IOWN), a next-generation telecommunications network being developed by NTT.

When a question is asked to NTT’s Generative AI Tsuzumi, answers from multiple AIs are displayed simultaneously. “Existing AI cannot criticize itself and is, so to speak, a dictator. To ensure AIs are used appropriately, we let them monitor and control each other,” Kawazoe said.

IOWN is capable of high-speed, high-capacity transmissions using optical technology, and can run multiple AIs simultaneously. AIs have also raised concerns about their tendency to “hallucinate” or state things that are not true as fact, but Kawazoe says, “If we can compare multiple answers, we can avoid the biases and errors of AI.”

NTT and The Yomiuri Shimbun released their “Joint Proposal for Shaping Generative AI” on Monday in an effort to balance the control and use of AI. To ensure such balance and space for healthy discussions, the proposal states that “it is necessary to ensure that there are multiple AIs of different types and of equal rank.” The AI ​​constellation initiative is a measure to this end.