In praise of stamp collecting (AI, Law, and Otter Things #25)
Welcome to the latest issue of AI, Law, and Otter Things! It is still a less-polished issue than I would like to post, but at least it is not another filler episode. This time, I speak briefly about two major items in world news before moving to a rant on research methodology and then presenting some recent posts from Digi-Con. And, as always, there are some otters at the end.
Quite a bit has happened since my last missive to you, but the most salient event has certainly been the Russian invasion of Ukraine. It is a grotesque violation of international law, and my solidarity goes to the people of Ukraine and all those affected by the conflicts. Already at this point, the invasion has produced some very important geopolitical consequences—cyberwar, the failure of Russia's information campaign, not to mention the European pivot towards defence and away from integration with Putin's regime—but I am not an expert in any of them, so I will refrain from providing any comment for the time being. Instead, I encourage you, the reader, to seek good information sources and do what you can to alleviate the situation of people being bombed and exposed to war.
In another huge piece of news, the IPCC has released a new document titled "Climate Change 2022: Impacts, Adaptation and Vulnerability". It has some shocking findings, such as that more than 3 billion people live in contexts that are highly vulnerable to climate change, and that even the "safe" 1.5°C threshold would lead to considerable and irreversible impact on the environment. I am by no means a climate expert, either, but the report suggest a need for action that creates some problems for a few trendy technologies, notably blockchain and large deep learning models.
Speaking specifically of the environmental impact of deep learning, the Stochastic Parrots paper provides a useful summary of the current literature and points out how these impacts have a disparate impact upon poor and non-white populations. It has been argued that such measurements are complicated by various types of opacity and uncertainty, and Google researchers have claimed some estimates overestimate emissions by a factor of more than a thousand. Even if one is willing to grant such a claim, however, it still refers to claims about operational emissions in training and not the entire life cycle of these models. So, we need better measurements of the actual impact of AI in our climate crisis.
In praise of stamp collecting
According to a well-known quip, usually attributed to Ernest Rutherford, all science is either physics or stamp collecting. This quote is often invoked in praise of certain properties associated with physics research—its heavy use of mathematical tools, its ability to provide predictive models, the systematic nature of its body of knowledge, and so on—which are deemed the be-all and end-all of research. Conversely, more descriptive practices, such as gathering information and making sense of it, are ascribed an ancillary role or even seen as the signs of an immature field.
This vision, as one would expect, is not exactly popular in legal scholarship. Even when scholars from Continental traditions still pay some lip service to the idea of a "science of law", the dominant view seems closer to the social sciences than to the physics envy often associated with economics. This is not just a consequence of the fact that many lawyers are maths-averse but also of the close(-ish) connection between legal scholarship and the practical reasoning needed for solving issues in legal practice. Publications in the field of law, therefore, involve a lot of stamp collecting in the sense described above.
Nevertheless, top-down theory-building is still seen as a more prestigious practice. At least in my little corner of the law, the papers that appear more often in citation lists and such are those proposing general theories for a given issue, coining concepts with some claim of generality. But I am somewhat sceptical about the value of grand theories when it comes to AI regulation. This is not to say AI regulation scholarship does not need to build theories, but rather that these theory-building instruments should focus on integrating the various disciplines relevant for regulation: various fields of the law, computer science, STS, and so on. Maybe, following Mary Midgley, we need to spend more time dealing with conceptual plumbing and less time building monuments. But this is an argument for another time and place.
For now, I would like to highlight that stamp-collecting scholarship can be a valuable instrument for building and sustaining useful theory. We need more taxonomies and attempts to systematize what has been done so far, as tech scholars (including myself here, of course) are all too prone to reinvent the wheel. This is not bad in itself, as people often stumble in details that had been missed in the previous 543289752348 attempts to introduce the same approach. But, as Claudio Lucinda often points out, reinventing the wheel means missing out on the incremental improvement to the previous version of the wheel. Stamp-collecting theory might thus help us avoid missing the insights developed by previous generations of scholars, especially as the pioneers in tech law fields are nearing or past retirement.
This systematization of thought is important, but it is not the most valuable form of stamp collecting. Any good collection involves some form of arrangement, but organization presupposes there is something to be organized in the first place. The metaphor thus points us towards another important task: the collection and description of situations related to the use of AI technologies, the social issues they create and amplify, and the regulatory approaches to these issues. While we seldom have the luxury of waiting for certainty before needing to take regulatory action, the accumulation of facts is an essential ingredient of theory-building, and it can help in revisiting stop-gap solutions once they cease being useful. Hence, we need more stamp-collecting work and better ways to value existing and future scholarly outputs of this kind within the profession. But, at least for now, I cannot do anything about the latter problem.
What's going on in Digi-Con
The past two weeks were quite eventful at The Digital Constitutionalist:
Our editor Francesca Palmiotto Ettore wrote a post on the right to contest automated decisions, proposing a transparency model that balances the need to provide information about the design of AI systems and the need to protect trade and industry secrets;
Edoardo Celeste, Pietro Dunn, and Maria Vittoria La Rosa contributed to the debate on what is digital constitutionalism and how it applies to the governance of digital platforms;
Domingos Farinho wrote on the Digital Service Act and how it handles conflicts between permitted actions in the online environment;
Renata Vaz Shimbo and yours truly published a post on humour as a challenge to automated moderation systems; and
Later today we will publish a post by Francisco de Abreu Duarte on the European approach to platform governance, the first of a three-part series.
I look forward to hearing your thoughts on these issues, and please consider submitting your blog posts on legal matters, science fiction essays, and artwork to Digi-Con!
Finally, some otters and a picture from the Badia Fiesolana
Not together, unfortunately.
[tweet https://twitter.com/In_Otter_News2/status/1497964771630628869]