Can Generative-AI (ChatGPT and Bard) Be Used as Red Team Avatars in Developing Foresight Scenarios?

This chapter examines the question of whether the Generative-AI (Gen-AI) systems of OpenAI’s ChatGPT and Google’s Bard (from 2024, Google Gemini) have value as ‘Red Team Avatars’ when developing foresight focused scenarios. After some initial explanation, both exploratory scenario and more dystopian science fiction (sci-fi)-style scenarios are drawn upon to provide illumination. As demonstrated in turn, many limitations—even more profound restrictions—were encountered during the course of the exercises conducted for this chapter. Overall, results could be argued to be somewhat disappointing, in that, as found also in previous chapters, for the Gen-AI systems to be most effective at Red Teaming they required substantial efforts in the area of prompt engineering. The efforts quickly became more resource costly, for example, in terms of the time taken to task the Gen-AI effectively, than when compared to the value that could be elicited by using them. When not actually denied, reference points soon become more lost in overall background ‘noise’ than realised as extractable ‘signals’. Once more, the findings here remind that properly verified ‘real’ and ‘human’ intelligence has greatest use and value when it comes to sophisticated activities, such as those of and required by Red Teaming, and to other similarly advanced analytical and assessment or estimation activities. In highly differing circumstances, Gen-AI might be able to assist at best, but it cannot compensate or replace. This conclusion is particularly acute in safety and security terms. End-users and other stakeholders should take close and continuing note.