Separating the real risks of AI from fear-mongering

The rise of generative AI has significantly influenced the conversation around online safety and integrity. Such discussions have evolved against the backdrop of heated global geopolitical events, which have amplified the dangers of misinformation, disinformation, hate speech, and targeted online attacks.

As concerns grow about the potential for AI to drive these and other forms of online spam and abuse, it is important to identify the most effective response strategies. However, given the abundance of sensationalism around the topic, it is crucial to differentiate exaggerated fears from the actual risks presented by generative AI.

With article headlines titled “Social media has never felt more dangerous” and “AI Poses ‘Risk of Extinction’, Industry Leaders Warn”, conversations about AI-driven abuse can often take on a tone of fear-mongering rather than proving informative. Moreover, the heightened sense of terror around the topic has been met with sometimes inflammatory, other times simply poorly thought-out government reactions around the globe.

A framework to distinguish between substance and noise

A clear understanding of real dangers is important to establish in order to develop and apply effective mitigation strategies. The following considerations can be used to distinguish between baseless alarm and legitimate concerns:

1. Understand the basics. The first way to distill real risk from sensationalism is grounding oneself in an understanding of how generative AI works. Foundational knowledge equips us with understanding what is technically feasible vs exaggerated. Consulting sources produced by AI researchers such as Dr. Sasha Luccioni and organizations at the forefront of such technologies, including HuggingFace, OpenAI, and [GitHub](https://github.blog/2023-04-07-what-developers-need-to-know-about-generative-ai/, can help form that foundational knowledge.
2. Understand context. Understanding the true risk also involves understanding the context. Abuse, be it AI-generated or not, typically targets a platform. Without understanding the scope of the damage that can be done to or using a given platform lends one to overestimate or underestimate potential harm.
3. Media literacy. Finally, it is important to recognize fear-mongering tactics used in the media. Sensationalist language, overly dramatic headlines, the absence of evidence or misrepresentation of facts are all indicators that warrant a critical eye. Additionally, examining the credibility of sources cited can provide more context. It is important to consider where the article is coming from, who owns the message, and where their interests lie. If claims by “experts” are being cited, it is important to determine who these experts are, and cross-reference with other sources to gauge their reliability.

Unknown unknowns

While some fears are very clearly exaggerated, others are speculative, making it difficult to evaluate their actual impact. This makes it increasingly challenging to predict actual concerns from sensationalized ones.

Additionally, abuse evolves in an adversarial context, with malicious actors locked into an arms race with whatever defensive tactics they are trying to circumvent. Because abuse is evolving continuously, attackers adapt their methods to bypass existing safeguards, making it tough to determine whether the abuse is based on outdated information, current threats, or speculative future risks.

Staying ahead in a rapidly changing world

No one can predict the future, but maintaining an awareness of the evolving AI landscape can help inform abuse mitigation strategies. This involves committing to ongoing learning and education, including staying informed about changes to technology, regulation, and policy. Furthermore, subscribing to credible news sources can provide access to reliable news without sensationalized tactics. Finally, maintaining good security hygiene by regularly applying patches and updates can protect against vulnerabilities that may be used as vectors of attack in broader abuse campaigns.