What would AI regulation look like?
Last year I gave a keynote where I considered environmental metaphors for the impact of AI on the information ecosytem. By considering the harm caused by chemicals such as DDT or interventions such as the introduction of rabbits to Australia, we can reflect on how the unregulated introduction of an invasive species or agent with long lasting consequences can negatively affect our information ecosystem.
A year on and we’re seeing some of that damage – much of social media is unusable now, flooded with AI slop, the environmental damage required to power huge data centres is considerable, and the impact on a society where we can no longer easily differentiate between truth and Ai content is beginning to show serious political manifestations.
So, let’s consider, what if we were to regulate AI the way we have regulated the use of chemicals such as DDT or the intervention into natural ecosystems? One’s first reaction may be “what’s the point? The cat is out of the bag now.” That may well be true. There is too much power and money invested in AI to allow for regulation at the moment. Similarly, any regulation would be difficult, maybe even impossible to enforce, as some state or dark web alternative would always break the rules. And you may well argue that regulation is very undesirable, would stifle innovation, is authoritarian, and so on.
Even if this is true, I would argue that there are three good reasons to consider possible regulation. First, when (or you may argue if) the very real damage of AI to our real lives becomes apparent (as the damage DDT did eventually became incontrovertible despite chemical companies trying to discredit research), we will want to have some plans in place for how to start controlling the damage. Second, even if enforcement is difficult, we do still attempt to enforce regulation around things that we deem harmful, and while that is not always successful if someone is determined, it drastically reduces the everyday quantity that might arise otherwise. Lastly, consider AI regulation a thought experiment, one that will potentially make us think about the balance of AI usage, and alternative ways of framing it rather than “anything goes.”
With those three reasons in mind, what might AI regulation look like? The model I would consider is similar to ethics clearance in research. We might allow some uses of AI to continue, but other AI uses would be banned (and if you were found deliberately creating and disseminating AI content you would face a fine), but you could apply to a central board for a licence to use it. The application would need to consider the following:
- Offsetting environmental impact required for AI generation
- How the content may impact society (for example, msiinformation, depiction of women, imapct on livelihoods, etc)
- The benefit that will accrue from the usage
- Why alternative methods would not be appropriate to achieve these aims
- How its proliferation through the information ecosystem will be monitored and controlled
- Clear labelling of outputs as AI produced
Such a list looks pretty draconian, and I don’t suggest we would ever implement it. But considering the damage AI is doing, and is likely to greatly increase doing, we should not just shrug and say “whatcha gonna do?”. It doesn’t get a free ride anymore, as we are all paying for it in multiple ways. By thinking about what a regulated usage system might look like it at least gets us to ask the questions of AI and its implementation that are missing from the current political gold rush. And it allows us to envisage different future frameworks around its use that are simply not part of the discussion now.
(Featured image Photo by meisam saeb on Unsplash)
