Immunizing and Countering Fake AI Images Through Overstatement?

One exotic idea of countering bogus, AI-generated images which have potentially dangerous political implications is to display similar, but largely hyperbolic images which are clearly fake and are supposed to be repudiated by users, along with the ‘original’ fakes. But is this a good way of countering the effects of fabricated pictures?

Admittedly, more and more fake images, but at times already more complex content such as animated GIFs and videos, have been flooding users’ feeds. Not only are there illustrations which claim to complement news, but also fallacious illustrations as part of commercials. In the latter case, it is often possible to retrace the propagator, issue a warning, and suspend those who do not comply with the terms of service. In the fomer case, complementing news and political commentary, content moderation is the method of choice.

Regarding the effect of hyperbolic messages, it is highly doubtful whether images which look more faulty will yield the desired reject rate. It is to be assumed that users will have known or come to know about the false character of the ‘original’ manufactured fake items and accept them despite pre-bunking and countering efforts, since that groups of people possess preconceived notions against other groups of people, sometimess even state institutions.

More importantly, if it is not the public relations offices of those concerned by the fake images, there could be a whole new, but likely small, industry of commercialized design specialists who would create likely less convincing content to counter disinformation on the payroll of internet services. This is not desirable, as it would in essence be a case of running behind those who flood the internet with fake content.

There are those who opt for watermarking or IDing artificial pictures, especially in the US where freedom of speech is enshrined in the national constitution, also allowing for social media companies adapt their standards and site policies. Other regional regimes emphasize moderation more strongly, setting the threshold for take-down measures in case of slander, wrongful allegations and other elements of offense lower, in many cases.

Watermarking and moderation will continue to define reactions of social media sites and other domains and services, in the foreseeable future. Hyperbolic images as an antidote against fake content, on the other hand, risks heating up political debates and leading in the wrong direction, as far as the sketchy effects such operations can yield are concerned.

One should also mind economic considerations. It is highly unlikely that social media companies will spend a fortune overstated media to counter fake content. There can be academic experiments, but there ought not be fully-fledged experiments by social media companies with hyperbolic images.

Thorsten Koch, MA, PgDip
Policyinstitute.net
2 September 2024