「经济学人」Generation confusion

But that’s not the whole story, when it comes to AI

When it comes to disinformation, “social media took the cost of distribution to zero, and generative ai takes the cost of generation to zero,” says Renée DiResta of the Stanford Internet Observatory. Large language models such as GPT-4 make it easy to produce misleading news articles or social-media posts in huge quantities.

And ai can produce more than text. Cloning a voice using ai used to require minutes, or even hours, of sample audio. Last year, however, researchers at Microsoft unveiled vall-e, an ai model that is able to clone a person’s voice from just a three-second clip of them speaking, and make it say any given text.

OpenAI, the American company behind GPT-4, has developed a similar tool, Voice Engine, which can convincingly clone any voice from a 15-second clip. It has not yet released it, recognising “serious risks, which are especially top of mind in an election year”.

Similarly, Sora, from OpenAI, can produce surprisingly realistic synthetic videos, in response to text prompts, of up to a minute in length. OpenAI has yet to release Sora to the public, partly on the ground that it could be used to create disinformation. As well as providing new ways to discredit or misrepresent politicians, ai tools also raise the spectre of personalised disinformation, generated to appeal to small groups (think soccer moms in a specific town). It may even be possible to “microtarget” individuals with disinformation, based on knowledge of their preferences, biases and concerns.

Though all of this is worrying, it is worth remembering that not all aspects of the technology are negative. ai, it turns out, can be used for fighting disinformation as well as producing it.

This article appeared in the Science & technology section of the print edition under the headline "Generation confusion"(May 1st 2024)

我没有译文，仅供拓展阅读

LearnAndRecord

2015年2月8日

2024年5月7日

第3377天

每天持续行动学外语