Deepfakes are coming: what is our plan?
Updated: Jul 17
David C. Donald - The Chinese University of Hong Kong; editor, Machine Lawyering
-- Recently, Forbes published an article with the startling title, Deepfakes Are Going To Wreak Havoc On Society. We Are Not Prepared. I agree with this, particularly in light of the impact of Covid-19, which showed that many highly developed countries were simply not prepared to use coordinated action to shut down the spread of a devastating disease throughout society. Deepfakes are AI-generated images, audio or video files that appear to be real although they are made of whole cloth, or in this case, mere pixels. They appear to be evidence of an event or statement, but are fiction. They can be entertaining, but as they improve in quality they will destabilize an already unstable public discourse and society in so many ways that it is difficult to catalogue the dangers.
If in the context of the "new cold war" between the US and China and the coming US elections, videos air of the US or Chinese President seeking a declaration of war on the other country in light of a video showing a massive, Pearl Harbor-like attack, what are we to think? How about a video popping up on election eve of a young Joe Biden forcing himself upon an electoral aide? Or during a joint US-Japan naval exercise near the South China Sea a video appears showing a row of leading protest figures in Hong Kong kneeling and being shot execution style by a PLA officer? True or False? Shall they act on it or not? Not everyone catches Covid-19 upon contact, but some people do. Not everyone will believe these videos but some people will, and they will share them. The viral aspect of information contagion is of course why we say "go viral" for things shared rapidly on the internet.
Truth and public discussion are central focuses of law. Rules of evidence used in trials are designed to filter out misleading information and are probably the aspects of law best known by the general public from courtroom drama in film. But we don't have an equivalent for public discourse. We have post hoc legal remedies for slander and libel, but this has not stopped "fake news" and will likely not be enough to stop deepfakes - it will resemble the use of emergency room and ventilator for Covid-19.
In securities markets, the release of misleading information to sway opinion and price is very old and most rules in the securities law actually go to filtering information disclosed, requiring that specific kinds of information is released, specifying how it is to be vetted, and punishing statements that are false, misleading or manipulate market price. An additional tool available to stock exchanges is to suspend the trading of a given security if its price suffers overly from such statements. If these rules were not there, market manipulation would probably have joined pornography as an early, popular use of deepfakes.
Working from the idea of suspension of securities trading and comparing it to the technique of the lockdown used for pandemics like Covid-19, if deepfakes came to permeate the internet, would shutting down the internet and going to "hard copy" be a solution? Given the use of internet to operate essential infrastructure and conduct finance, that would seem impossible. Indeed, an expected shutdown of the internet could be incorporated into a deepfake as hard evidence to corroborate an alleged nuclear strike, as Neal Stephenson shows in the faked attack on a town called Moab, Utah, in his 2019 novel Fall: or Dodge in Hell. Stephenson postulates the faked strike on Moab (accompanied by a distributed denial of service that shuts down its internet access) and the rise of slanderbots aimed at celebrities eventually destroy the internet as an information source, leading to the rise of an entire industry of personalized data sifters to clean out the fake and abusive - so that the wealthy have more trustworthy information and the poor live in constant, violence-laced confusion.
Like a stock exchange controlling the flow of information and trading, governments and industry could also step in to filter information. China already has near total control of internet within its national borders, and could then decide what would be seen. In the US and Europe, the GAFA (Google, Apple, Facebook and Amazon) fill the equivalent role of controlling internet content and could thus exercise similar censorship. However, this would mean that in China only government-approved fakes would be possible and that in the US only industry-approved (essentially, billionaire-approved) fakes would be seen. Truth as something objectively existing would still disappear.
Organizations studying law and technology should start working through the options very seriously. As public discourse becomes less stable and tensions rise both domestically and internationally, the internet looks like dry tinder awaiting conflagration. Meanwhile there can be no doubt to the truth that as data accumulates and programs and training improve, the quality of AI fakes is steadily increasing. International contagion - whether in finance, disease or information - is rapid and can have devastating consequences on society. It would be highly advisable for public bodies to have a few publicly vetted contingency plans ready to greet a possible deepfake pandemic.