AI may be listening more than you think
Privacy, ethical concerns grow as use of voice-based applications soars
KEIICHI MURAYAMA, Nikkei commentator
TOKYO -- Real-time voice translation based on artificial intelligence still has a way to go before becoming scary good, but it's already passable for simple conversation. At least that's my impression after trying Microsoft's video conversation system with a Spanish woman.
"Have you ever been to Japan?" I asked in Japanese.
"Many times," she -- or rather the system -- replied in perfect Japanese, after translating her response from Spanish.
Not exactly a riveting conversation, but this is bound to change. Microsoft has recently improved the accuracy of the system through use of "deep learning" technology, and the future looks bright. With the addition of Japanese, the system can now speak 10 major languages.
Simultaneous translation systems based on AI have undergone quite an evolution, so much so that Japan is planning to deploy an AI-based system for the 2020 Summer Olympic Games in Tokyo. And while not a translation system, Amazon.com's Echo speaker, which interfaces with the Alexa AI app, provides voice control for playing music and reading news, weather and other real-time information.
Since the iPhone debuted 10 years ago, touch-screens are everywhere. Sound is now the next frontier for the user interface, and there are only a few technological leaps to go before it too goes mainstream.
Still, there are some voicing alarm at this development, and their fears are not new.
In his 1970 sci-fi classic "Koe no Ami" (The Voice Net), Japanese writer Shinichi Hoshi described a dark future of networked computers listening to phone conversations, accumulating personal information, then blackmailing people in exchange for keeping their secrets. Through constant surveillance, the network gradually intimidates the populace into lives of insipid tranquility.
Creepy, but still fiction.
However, this vision raises legitimate concerns: AI may not be able to will itself into sentient existence, but its prevalence -- and sophistication -- could make it an attractive tool for abuse, especially as regards privacy.
Governments and corporations are already vacuuming mountains of personal information, and voice-interactive AI will make this easier and more thorough, interpreting our feelings, determining our age and approximating our level of education, said an executive of an American venture business developing interactive AI.
The company provides corporate clients with AI solutions for customer service. While matching customers with the best possible services, the AI could conceivably amass sensitive personal information without realizing it.
In environments that employ voice-based AI, misunderstandings can occur, causing he-said-she-said spats between users and operators.
"If everything is recorded to keep evidence, privacy concerns will emerge," said Ichiro Satoh, deputy director general of the National Institute of Informatics.
AI learns and becomes more capable as it accumulates data, which begs the question: What is being learned and how is the knowledge being used? There are good -- and not so good -- answers to this.
Microsoft got an earful of the latter when it unleashed Tay, an AI chatbot, on Twitter a year ago. In less than a day, the impressionable bot soon began rattling off such a string of obscenities and inflammatory posts that Microsoft quickly pulled the plug.
AI can turn racist if it learns from racist data, which in turn can lead to social problems, said Microsoft vice president David Heiner.
AI in self-driving cars and robot nurses may be great, but maybe less so for answering questions. Unlike person-to-person exchanges, we humans may be unable to ascertain the reasoning behind AI's answers.
In order to prepare for a future of voice-based AI, there should be consensus as to how it will be used. There should also be fields and applications that are clearly off-limits.
Companies developing AI are filled with talent, but not everything in the process should be left to them, as some profit-based decisions could be viewed as ethically challenged.
The free communication app Line has more than 200 million users worldwide, a massive aggregate of private conversations among people from all walks of life. Messages are encrypted, and even Line's handlers cannot peek inside.
"People will stop using our app if they think it's weird," explained Takeshi Nakayama, chief privacy officer at the company. He noted that Line examines its service weekly from users' perspectives, ensuring that there are no privacy issues. It has also has set up a foundation to conduct studies with university professors and other experts to establish legal safeguards for privacy.
Line will release a speaker that employs AI this summer. "We'll put out feelers to see how users react to it," Nakayama said.
In the meantime, IT companies developing voice-based AI should engage their customer base and others to discuss AI-related issues. This kind of transparency will hopefully ensure the ethical use of AI as it becomes more entwined with our lives.