The pace of generative AI development has been astonishing to the degree that the normative space has been unable to keep up. As governments start looking into some implementations of the technology, such as ChatGPT, more advanced techniques and products continue to emerge by the day. Society is changing in irrevocable ways, and it is paramount that the Internet Governance community turns its attention to this question. This article touches upon some points that may be of interest.

Back in November 2022, I published an article on CircleID on how generative AI was already succeeding in most creative tasks, about a month ahead of the protests staged by artists around the world concerning the usage of their art to train models. In the ensuing semester, all of the technologies mentioned in the article, be they text, code, image, or audio, have already received substantial updates and are performing much better than what was then described.

One immediately observable implication for IG is a change in the content being served by websites. In the case of looking for answers to a question in major search engines, it has become commonplace to receive pages within the top results that are written in a structure that is quite characteristic of ChatGPT. Search giant Google clarified in early 2023 that it considers AI-generated content to be as valid as any other, as long as it is of high quality. This is not restricted to websites only, as storefronts such as Amazon are seeing an influx of books written or co-written by AI as well.

Social media platforms have a self-confessed longstanding issue with bots, which already make use of more primitive forms of AI to generate posts, influencing behaviors and outcomes. New generative AI technology provides ample opportunity for this to deepen, as it has become trivial to code bots running locally on consumer-grade equipment that can find mentions of a certain subject or person and craft coherent responses for or against something based on an independently trained dataset that can contain whatever version of the facts that is desired.

To further contextualize the above, quoting The Washington Post: “Semrush, a leading digital marketing firm, recently surveyed its customers about their use of automated tools. Of the 894 who responded, 761 said they’ve at least experimented with some form of generative AI to produce online content, while 370 said they now use it to help generate most if not all of their new content.” In other words, the automated generation of content at scale is not speculative but rather a reality.

The cybersecurity sector is also seeing significant disruption. Malicious actors can combine existing methods with variations of techniques described above to, for example, create threat spaces that are more credible to victims, allowing for more tailored phishing attacks that mimic trusted sources with precision, including family and coworkers. Current technology also allows for the imitation of voice and mannerisms of a given person, fooling systems that rely on those aspects. Video generation is also quickly emerging and might eventually be achieved in real time.

Fritsch, Jaber, and Yazidi (2022) have outlined some of the current implications of generative AI for the purposes of malware creation and defense, concluding that “offensive deployment of AI within malware improves malware performance, including methods such as selection of targets, extracting authentication factors, enabling the automated and fast generation of highly efficient phishing messages, and swarm-coordinated action planning.”

The importance of the limited set of examples provided above will, however, only intensify. Recently, “autonomous agents” have entered earnest development. This implementation of AI relies much less on human intervention, instead making use of a network of AI agents to reach a pre-established goal that can be defined in human language, with no need for coding. This allows AI to have memory, performing a task and then returning to a task list that it can manage by itself, moving on to the next goal. The most popular implementation of autonomous agents right now comes from the Auto-GPT project, but there are also several others in development.

In practice, with some additional capabilities, all of these technologies can be set to work together and create self-sustaining systems that keep on generating output and spreading it around the Internet. Considering that they are scalable and are not that expensive to run (although it’s not very cheap yet), it is conceivable that in a very short time, we will be dealing with a much more complex online environment, deepening discussions such as that of credibility, conflict, identity, and so on.

How will the IG space react to this? As seen above, core questions that shape this area are seeing progressively faster changes, and while people with greater technological knowledge are finding ways to implement or avoid these to some degree, our ultimate responsibility is still to ensure that the average user has access to a global network that is useful and fruitful to them. New paradigms need to be considered in order to accommodate generative AI.

Within spaces in which Internet policy is discussed, involved parties need to get up to speed and attempt to keep up to some degree with AI development, even though this is a daunting task. Several discussions will be moot if this new age of technology is not taken into consideration, and we risk establishing norms and consensuses that will become obsolete within a short amount of time.

Instead of fighting against the inevitable spread of AI, there is a need to start building safeguards and educating non-specialists so that the impacts are mitigated, and the transitions that will be felt in many areas have a smoother curve. The beginning of this new era might be as significant as the emergence of the commercial Internet, only it will be much faster and have greater information asymmetry. We need to be prepared.