U.S. technology company Nvidia, the next generation microchip maker and AI expert, has revealed how much its facial and other image generation technology has improved over the past four years.
The images below are not real people but AI-generated ‘deepfakes’ produced by the company’s algorithms. Hyper-realistic generations of animals and objects such as cars have also been published. The impressive new technology the images represent is provoking both optimism and concern around potential applications.
Source: Nvidia
Nvidia’s image generation technology works on the basis of a GAN approach. GAN stands for General Adversarial Network and is comprised of two agents that can be compared to a counterfeiter and detective of fakes. An Nvidia blog post describes the AI-process as follows:
The typical GAN setup comprises two agents:
- a Generator G that produces samples, and
- a Discriminator D that receives samples from both G and the dataset.
G and D have competing goals (hence the term “adversarial” in Generative Adversarial Networks): D must learn to distinguish between its two sources while Gmust learn to make D believe that the samples it generates are from the dataset.
A commonly used analogy for GANs is that of a forger (G) that must learn to manufacture counterfeit goods and an expert (D) trying to identify them as such. But in this case—and this is a very important GAN feature— forgers have insider information into the police department: they have ways to know when and why their products are marked as fake (though they never get to see any of the real goods). Likewise, the police have access to a “higher authority” that lets them know whether their guesses are correct. During training G and D keep playing this game, getting better as they play, until they both become so good that the samples produced by G are indistinguishable from the real stuff.
The positive or neutral ways that the kinds of deepfakes that Nvidia engineers have succeeded in producing could be used are for stock photography or any other kind of business model that is image based. Deepfakes could be expected to significantly bring costs down in these niches in future.
The fear is that hyper realistic deepfakes will mean it becomes almost impossible to trust any kind of content in the future. How will we know if any image or video is real or has been conjured up by an AI algorithm? Deepfakes could be used to manipulate politics or anything else for which public opinion is important.
The good news is the level of sophistication achieved by Nvidia researchers is still something that would be very hard for anyone else to achieve. Their project involved using eight of the super-powered Tesla GPUs the company manufactures as well as some of the world’s most advanced Deep Learning experts dedicating countless hours. However, the technology will undoubtedly become more accessible.
Part of Nvidia’s motivation for publishing its research, other than showcasing the latest technology in the world of powerful processors and the expertise of its research teams, is to raise awareness around the potential dangers of deepfake technology.
It is something we have to be aware of in the contemporary world and start discussing safety measures to protect against. Some suggestions include ‘real’ images all having some kind of tag incorporated into them to authenticate the location and time at which they were taken. The greatest value of the Nvidia research is certainly how it highlights the need to now question what we may have once simply accepted as reality.