I
Let’s start with the current policy narrative about how intractable it is and will be to regulate the use of Generative AI, globally or not:
. . .while global regulatory solutions might [be] desirable, we also have to consider the standard or level of protection that these solutions can afford. Under the political reality of treaty negotiation, the more seats there are around the table, the more compromises will have to be sought. Such compromises often lead to lowering the level of protection. While this trade-off might be necessary to ensure that more individuals across the globe can benefit from a certain level of protection, this also results in a challenging and even paradoxical position where, simultaneously, more voices around the table push for further consensus on regulations, while many voices still remain completely excluded from the debate.
https://dataethics.eu/wp-content/uploads/2023/09/Generating-AI.pdf
Now let’s get to the reasons why this kind of reasoning is misleading.
II
Put aside the quote’s could-isms of the “can’s” and “might’s” that logically and empirically entail “it also might not”. Put aside the fact that “compromises often lead to lowering of the level of protection” actually means: Whether or not standards of protection are lowered depends on a case by case determination.
Focus instead on that assumption that it’s all about the regulator of record when it comes to Generative AI. God help us all, if that were true.
Fortunately, that cannot be true everywhere and in the same senses that matter when it comes to Generative AI.
We know, for example, that when it comes to critical infrastructures (which is and will be a major Generative AI adopter), regulatory functions must be dispersed beyond the regulator of record (e.g., infrastructure control rooms and support staff discover and correct for inevitable error by the regulator of record). [https://mess-and-reliability.blog/2023/10/01/when-it-comes-to-societys-critical-infrastructures-regulatory-functions-are-dispersed-beyond-the-regulator-of-record/]
We know that these real-time corrections for lapses by the regulator of record (i.e., the more regulators of record try to be comprehensive, the more they miss in dynamically changing conditions) are part and parcel of infrastructure control rooms as a unique organizational formation to address real-time contingencies, in this case with respect to Generative AI in actual use. In fact, the evolutionary advantage of infrastructure control rooms lies in the skills and expertise of its operators to operationally redesign in real time what is otherwise inadequate regulation and technology. [https://mess-and-reliability.blog/2023/05/24/a-national-academy-of-reliable-infrastructure-management-2/]
And in related fashion, we know that where highly reliable infrastructures matter to a society, it must be expected that the social values reflected through these infrastructures differ by staff and their duties/responsibilities (e.g., responsibilities of control room operators often go beyond their official duties). Whatever the societal values with respect to Generative AI, their differentiation through and by critical infrastructures is happening. [https://mess-and-reliability.blog/2022/04/16/critical-infrastructures-regulate-differently-than-government/]
III
So what?
Please then, more realism when it comes to regulating Generative AI! To the extent the preceding quote is today’s policy narrative, the sooner really-existing regulatory counternarratives are examined the better.
Sources. Published versions detailing the above claims and relevant cases can be found in
Re infrastructure control rooms as a unique organizational formation and the need for dispersed regulatory functions: Roe, E. and Schulman, P.R. (2016) Reliability and Risk, Stanford CA: Stanford University Press
Re evolutionary advantage in redesigning defective technology and regulation: Roe, E. (2021). “A National Academy of Reliable Infrastructure Management.” Issues in Science and Technology (accessed online at https://issues.org/national-academy-reliable-infrastructure-management-roe/; updated most recently in Roe, E. and Schulman, P.R. (2023). “An interconnectivity framework for analyzing and demarcating real-time operations across critical infrastructures and over time.” Safety Science (accessed online at https://authors.elsevier.com/a/1hl653IVV9uro7)
Re infrastructure differentiation of societal values: Roe, E. (2023) When Complex is as Simple as it Gets: Guide for Recasting Policy and Management in the Anthropocene, IDS Working Paper 589, Brighton: Institute of Development Studies, DOI: 10.19088/IDS.2023.025
WHEN COMPLEX IS AS SIMPLE AS IT GETS 