Using AI to automate disinformation

Natural language generation (NLG) is a sub-field of AI which aims to make data more easily analysed and interpreted, thereby making the writing of data-driven reports and communications less burdensome. NLG systems also offer to bridge the gap between Big Data business intelligence tools, and low rates of data literacy in the general population: see this 2016 Forbes article.

NLG also has potential applications in cognitive science and psycholinguistics.

NLG systems use neural networks, machine learning and massive datasets of human-created prose to create a writing style, tone and structure which can be varied according to audience, context and purpose. They are presently capable of generating text which can be difficult to distinguish from the work of human authors.

However, the success of these systems has lead to a concern that they could be used to automate disinformation campaigns.

The Center for Security and Emerging Technology recently published a study considering whether an automated system, namely the GPT-3 system created by OpenAI, could be used to generate effective content for disinformation campaigns.

The study evaluated the performance of GPT-3 across six disinformation-related tasks: from generating varied short messages to a particular theme (Narrative Reiteration); to changing the views of targets including by crafting messages tailored to a particular political ideology or affiliation (Narrative Persuasion).

The authors concluded that "while GPT-3 is often quite capable on its own, it reaches new heights of capability when paired with an adept operator and editor", making it "a tool that can help them to create moderate- to high-quality messages at a scale much greater than what has come before".

Given the possible effectiveness of such campaigns, the authors surmised that the best approach to mitigation is to "focus on the infrastructure used to propagate the campaign’s messages, such as fake accounts on social media, rather than on determining the authorship of the text itself."

To find out more, read the summary, or download the full report.