AI-generated images and the unsettling questions of identity and impersonation

  • 23 Oct 2023
  • 2 Mins Read
  • 〜 by Anne Ndungu

In ‘A Tale of Two Cities’ by Charles Dickens, the author spins a gripping narrative of an English couple who have to be rescued from the guillotine during the violent turmoil of the French Revolution. 

The book starts with a trial that is dismissed on the premise that the man accused of spying resembles another man in the courtroom and therefore the witness can not possibly identify him as the perpetrator. The case is dismissed once reasonable doubt has been established regarding the identity of the accused. 

While researching AI and its biases, a thought occurred to me. Can AI develop an image that resembles me without any input of my image? After all, the AI images generated are of persons and what are the chances that it will generate images that resemble someone? Rather high. This is not even a matter of possibility. Some AI images that I have seen already resemble famous personalities. Once again, we have to talk of the fact that the AI model is fed data and it is going to create images given the data sets fed.

Many possible ethical scenarios arise from this.  If I see a billboard with an AI-generated image that resembles me, and everyone around me points it out to me and asks me if I posed for the rather realistic image, can I have any legal recourse if the image has been used in a manner that harms my reputation, my professional or my social standing? It may go against my values in a way that may be detrimental and I would have no way of explaining that it is not my image. 

This perhaps takes us back to the issue of deep fakes where recently the African Union Chairperson, Mr. Moussa Faki, was impersonated by fraudsters who faked his voice and placed several calls to several European leaders without the standard Note Verbale used in diplomatic circles. But this is a case of deliberate impersonation. 

It follows that AI can also inadvertently impersonate people, in fact, it does routinely though inadvertently impersonate people. It can be argued that every image generated by AI potentially resembles an existing person. For this reason, it is crucial to consider the potential repercussions resulting from AI-generated images that could be used in ways detrimental to someone’s reputation. 

While it might seem like a one-in-a-million risk, it is a risk nonetheless, and it is rapidly evolving into a common reality that warrants serious consideration.