Microsoft researcher warns that deepfakes pose a growing threat

Interactive and compositional deepfakes represent two growing threats in information security. This заявил Microsoft Chief Scientist Eric Horvitz.

According to the expert, the developing capabilities of discriminative and generative AI methods are approaching a tipping point.

“The achievements provide unprecedented tools that can be used by state and non-state actors to create and disseminate convincing disinformation,” Horvitz wrote.

The scientist argues that the problem arises from the methodology of generative adversarial networks, consisting of two competing components: a generator and a discriminator. The generator is intended to create content, and the discriminator to evaluate its quality.

Horvitz added that, over time, the generator will learn to fool the discriminator.

“With this process at the heart of deepfakes, neither pattern-recognition methods nor people will be able to reliably detect forgeries,” he wrote.

The expert stressed that until now AI-forgeries have been created and distributed as one-off, autonomous creations.

“However, now we can expect the emergence of new forms of convincing deepfakes that go beyond fixed single-element productions,” he said.

Horvitz proposed several measures to prepare for and defend against the anticipated rise in deepfakes. Among them:

• stricter standards for journalism and reporting;
• increasing media literacy among the public;
• introducing new authentication protocols to verify identity;
• developing standards to verify content provenance;
• ongoing monitoring.

“It is important to stay vigilant regarding interactive and compositional deepfakes,” Horvitz concluded.

In September, scientists unveiled method for detecting audio deepfakes, measuring differences between samples of natural and synthetic speech.

In August, hackers used a deepfake for a listing scam on Binance. The attackers posed as the Chief Compliance Officer of Patrick Hillmann’s company in a series of video calls with representatives of cryptocurrency projects.

In June the FBI warned of a growing number of cases of deepfake use in online interviews.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!