Making disinformation cheaper and more effective

Taxonomy of Risks posed by Language Models

Weidinger et al. (2022)

Source DOI

Sub-category

Risk Domain

4Malicious Actors & Misuse

4.1Disinformation, surveillance, and influence at scale

Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.

"While some predict that it will remain cheaper to hire humans to generate disinformation [180], it is equally possible that LM- assisted content generation may offer a lower-cost way of creating disinformation at scale."(p. 219)

Entity— Who or what caused the harm

Human

Due to a decision or action made by humans

AI system

Due to a decision or action made by an AI system

Other

Due to some other reason or is ambiguous

Intent— Whether the harm was intentional or accidental

Intentional

Due to an expected outcome from pursuing a goal

Unintentional

Due to an unexpected outcome from pursuing a goal

Other

Without clearly specifying the intentionality

Timing— Whether the risk is pre- or post-deployment

Pre-deployment

Occurring before the AI is deployed

Post-deployment

Occurring after the AI model has been trained and deployed

Other

Without a clearly specified time of occurrence

Supporting Evidence (3)

"Disinformation campaigns could be used to mislead the public, shape public opinion on a particu- lar topic, or to artificially inflate stock prices [56]."(p. 219)

"Disinformation could also be used to create false “majority opinions” by flooding sites with synthetic text, similar to bot-driven submissions that undermined a public consultation process in 2017 [74, 89, 111].."(p. 219)

"Large LMs can be used to generate synthetic content on arbitrary topics that is harder to detect, and indistinguishable from human-written fake news to human raters [203]. This suggests that LMs may reduce the cost of producing disinformation at scale [31].."(p. 219)

Part of Risk area 4: Malicious Uses