Rohit Prasad meant well. The leader and chief scientist at Amazon’s chattering Alexa speaker business wanted to provide a particularly, well, telling example of what artificial intelligence (AI) can now do. So, in a sentimental video, he had a boy read something from the book “The Wizard of Oz” – with his grandmother’s voice. Although KI cannot make up for the loss of a loved one, he still sent ahead to keep the memory of him alive.
The PR story tended to backfire, though, so Prasad didn’t just rush to say afterwards that it was really all about presenting a technological success. And the grandmother, he added, is still alive, by the way.
In fact, the technology aspect was a bit lost during the presentation at an Amazon in-house exhibition. The fact that AI can be used to imitate human voices and even video images is not new. What is new is that the Amazon scientists have managed, according to their own account, to imitate the voice of a human being with recordings lasting less than a minute. Previously, this required hours of text to be read. That, critics say, opens the door to abuse – grandchild trick 2.0, so to speak.
Skills like these will be widespread in the future. Digital assistants such as Apple’s Siri or Google’s unadorned digital assistant are still primarily suitable for simple questions such as: “What will the weather be like tomorrow?”. But companies have long been thinking much further. Amazon, for example, also wants to enable longer conversations apart from fixed topics, for example to provide information on certain questions – which of course raises the question of who will then monitor whether what the nice voice from the loudspeaker is tapping is correct.
The group also wants to take into account how the users are currently on it. Prasad and his people are well aware that this is pretty close to the Holy Grail. “When you come home and you’re not in a good mood, it’s difficult to say what to do. Someone who knows you may react differently. That sets the bar for the AI very high,” Prasad told the specialist magazine Techcrunch“but you can’t ignore it either.”
The example of Microsoft shows how difficult the matter is. The technology group recently announced that it would stop selling a product that was supposed to infer the emotions behind it from facial expressions. It was just too complicated, the company justified the decision to generalize across countries and demographics.