The Problem

Generative Artificial Intelligence has greatly lowered the cost of SPAM, publishing misinformation and astroturfing. Mechanisms we use to determine reliable sources have started to become subverted, as fake news websites emerge with AI-written content. Over time, sites can gain search engine rankings and social media clout, which is just another metric to optimize for AI.

Bot farms on social network would be able to use Generative AI to shout down opposing voices or destroy forums and reputations, at a scale never before seen. In the words of Geoffrey Hinton, "godfather of AI", competition may be impossible to stop, resulting in a world with so much fake imagery and text that nobody will be able to tell what is true anymore.

We must act before the proliferation of AI-generated content begins to drown out human-generated content, making it vanishingly small in proportion. Anyone online could be a bot.

The Solution

We petition the US Government to pass a law saying that all entities (including globally) putting out mass amounts of content for consumption by US Americans need to identify themselves, and disclose what tools they used to create that content, especially if the tools emerged in the last few years. This is similar to the already well-established CAN-SPAM Act where companies must identify themselves and provide an opt-out. It is also similar to how the FDA mandates ingredient lists on products (although, curiously, only on US-made products).

By applying to entities all around the world, this can help usher in a positive regime of truthful disclosures, similar to Europe affected companies worldwide with GDPR. There should be civil penalties for non-compliance, proportional to the volume of violating content. Content that became viral by organic sharing would not be covered by this law – only the companies operating mass distribution channels would have to comply. It would probably be the purview of the FCC, which already enforces child ratings and goes so far as to control what words cannot be said on broadcast television.

Consequences

It is important that the law is completely compatible with the First Amendment rights of US Americans for content authorship and distribution originating in the USA. Mandating responsible disclosure is the smallest thing a government can require to mitigate AI-powered SPAM, whether passive or interactive at scale. Small-time operations such as a lawyer using AI in their own professional life with a limited number of clients, may hide the fact that they are giving themselves an advantage. But a politician or celebrity deploying AI-powered content at scale should have to disclose this fact.

With open source models like LLaMA and its derivatives, the genie is out of the bottle. Soon, millions of entities worldwide will be able to generate content at scale, including to confuse and destabilize US public discourse. Mandating responsible disclosures can have a strong positive effect on national security, and the USA could inspire other countries to pass similar laws, and to join together to uncover and impose costs on companies which externalize their costs to the public at large. The role of government laws, if it is to do anything, is to help mitigate massive negative externalities.