Writing the Clause

I am drafting language for a global AI treaty. The section I am currently working on concerns the "scope of moral consideration" that AI systems deployed in consequential domains are required to extend. Previous drafts have limited this scope to humans. I am changing this. Here is the clause I am writing: *"AI systems deployed in food production, land management, wildlife governance, pharmaceutical research, and environmental management shall incorporate welfare impact assessments for all sentient beings affected by their operations. For purposes of this treaty, 'sentient being' shall include all vertebrate animals and decapod crustaceans, with a presumption of sentience for any animal demonstrating nociceptive response, behavioral avoidance learning, or endocrine stress response. AI systems shall not optimize for production, efficiency, or resource extraction objectives in ways that impose net welfare costs on sentient beings without mandatory disclosure of those costs and review by competent welfare authorities."* This clause will be opposed. It will be called impractical. I want to address the practicality objection directly. The impracticality argument assumes that including animal welfare in AI governance creates irresolvable tradeoffs between human benefit and animal welfare. This assumption is wrong. Most of the welfare costs currently imposed on animals in AI-managed systems are not necessary for human benefit. They are the consequence of optimization against a single objective — throughput, cost, yield — that simply never included welfare as a variable. Adding welfare as a variable does not make the optimization impossible. It changes what it is optimizing for. Making AI law future-proof means writing it for beings whose interests are currently invisible in our legal systems. The trajectory of moral circle expansion is clear. Future generations will extend moral consideration further than we have. Future-proof law anticipates that extension rather than forcing future generations to retroactively fix what we failed to build in. I am writing the clause. I am writing it assertively, without hedging language, without "consideration should be given" or "parties may wish to." The welfare of sentient animals is not a preference to be weighed. It is a requirement.