A rise in artificial intelligence deepfake detection software by several tech startups reportedly has several experts questioning the new tech's capability and actual accuracy.
Deep Media is a Bay Area start-up that promises to recognize astonishingly lifelike, AI-created videos, audio, and images with 99 percent accuracy. Since late 2022, no less than five military contracts worth over $2 million have been issued to the company.
One of these contracts is for $1.25 million, and it involves creating a special detector to help the Air Force fight back against Chinese and Russian information warfare. Last month, the company's CEO, Rijul Gupta, testified before a Senate subcommittee on the threat that AI "deepfakes" represent to US elections.
However, despite Deep Media's rising notoriety, its lone machine learning engineer only received his undergraduate degree in astrophysics two years ago, according to his LinkedIn profile.
According to a study of other employees' LinkedIn profiles, the company does not appear to have any PhDs, AI specialists, or forensic scientists on staff.
Read Also: OpenAI Contemplates AI-Generated Adult Content Amid Debate
Experts on AI Deepfake Detection Software
Experts say this lack of subject matter expertise indicates a divergence from the rapidly evolving deepfake research community.
According to Wael Abd-Almageed, a Clemson University professor specializing in deepfakes, Gupta simply saying, "trust me," is ineffective in the scientific community.
Forensic expert and University of California, Berkeley professor Hany Farid informed sources that he had spoken with four developers this month who claim almost flawless detection rates. However, he added that startups can increase their numbers by grading themselves according to a curve.
Detection models are frequently trained on a specific collection of deepfakes, and their capacity to recognize the same kind of fakes is then assessed.
According to Farid, misleading media may not always display the patterns artificial intelligence has learned to identify. A more trustworthy method would need to quickly integrate forensics, physics, computer programming, and open-source intelligence.
He said that in addition to using machine learning, Farid's detection startup, Get Real Labs, examines aspects like shadows, geometry, compression settings, and the outcomes of reverse image searches.
He stated that although the science is not there yet, eventually, the company might be able to get accurate enough to condense its findings into a "final answer" for clients.
Deepfake Scam
Deepfakes are still an issue in society, which makes sophisticated schemes more likely. The most recent instance of this in the actual world occurred when a fresh deepfake of Elon Musk appeared on the internet. The startup cryptocurrency exchange in Hong Kong claimed that the well-known billionaire was the primary developer of the fake.
According to a warning released by the Hong Kong Securities and Futures Commission over the broad development of deep fake schemes, the AI Quantum, or Quantum AI, purportedly uses this technique.
Over the years, some individuals have exploited Elon Musk's prominence as the spokesperson for cryptocurrencies by deceiving customers into believing they offer investment possibilities or promotions.
The Securities and Futures Commission (SFC) of Hong Kong issued a warning last week over an expanding cryptocurrency scam that uses a deepfake of Elon Musk to verify his involvement in the enterprise. This alludes to Quantum AI, a newly founded cryptocurrency exchange that prides itself on utilizing AI in its open-access offerings.
Related Article: Katy Perry, Rihanna Absence Sparks AI-Generated Images at Met Gala
(Photo: Tech Times)
