Autonomous Organizations, by Shawn Bayern

Rating: 👍

Robot, dreaming, courtroom, freedom and liberty --ar 3:2 --sref 1008910165 1416658095 4649175912

What is Legal Personhood?

Legal personhood for software systems will likely be one of the defining civic issues of our time. What rights are we required to afford them? What rights should we afford them? Shawn Bayern, author of Autonomous Organizations, doesn’t tackle these questions directly in a moral sense, but does argue that not only should Autonomous Organizations (AOs) have some level of legal personhood but also that they effectively already do under contemporary law. Notably, this book doesn’t really deal with DAOs in a crypto sense, nor does it directly deal with LLM-based AI systems, as it was written in 2021. It only concerns itself with an abstract ‘algorithm’ that makes decisions and enters contracts, regardless of whether that algorithm is deterministic or not.

Can Algorithms Enter Agreements or Contracts Today?

Bayern points out that algorithms already can engage in contracts, both with people and one another, with minimal to no human oversight. If you apply for a credit card online, an algorithm can autonomously assess your credit risk, make you a custom offer, send you a contract, and receive your signature - all without a human manually checking or approving any of the above steps. Algorithms can also enter contracts with one another - when one quant trading firm’s algorithm decides to sell a security and another’s decides to buy the same security at the same venue at the same price, they are implicitly contracting with each other on behalf of their representative firms. Under the Uniform Electronic Transactions Act (UETA), “A contract may be formed by the interaction of electronic agents of the parties, even if no individual was aware of or reviewed the electronic agents’ actions or the resulting terms and agreements.”

Can Algorithms Do So Autonomously?

We’re clearly fine with algorithms acting on behalf of firms or organizations or people with minimal oversight - Bayern takes this a step further and asks what if those orgs had no people and only algorithms? Would they still be permissible? Bayern argues that this is also already permissible under contemporary law, mainly by establishing an LLC with 0 members. In effect, you create an LLC (as you normally would), adopt some operating agreement or charter (in this case a piece of software that runs your firm), and then have all the humans leave the LLC. Bayern claims that this gives the algorithm ‘legal personhood’ in effectively the same way a corporation can achieve ‘legal personhood’ - corporations can’t marry each other or practice religion per se, but they can now have their interested represented in a court setting. Furthermore, many states already allow for LLCs to exist indefinitely, provided their initial charter says so. These ‘personless’ LLCs can exist for effectively any purpose, not solely creating shareholder value. Per the Revised Uniform Limited Liability Company Act (RULLCA), “An LLC may have any lawful purpose, regardless of whether for profit.”

Potential Frameworks for Holistic Regulation of AOs

Maybe this feels like a loophole to you, and maybe this feels like it will be closed by future legal frameworks. Bayern is skeptical, based on an analysis of what those future frameworks for granting (or prohibiting) legal personhood might be.

The Denialist Framework: This would be if legislation came out that said “only entities with XYZ characteristics can be considered legal entities, and all others are denied such rights.” The difficulty with this approach is that it is very reductive and brittle - imagine if we had such rights that only gave civic liberties to men, or people of a certain class, or race, or age, etc. Over time, we clearly minimize the parties denied certain rights, and the modern corporation is already, in effect, a synthetic organism (though not yet autonomous).

The Regulatory Framework: This would be if there were no hard and fast rules for granting legal personhood, but instead some public body had to reason about a candidate person, and decide on an ad-hoc basis. Imagine if we had some body that could say that ABC org is allowed to have a nonprofit status for certain reasons, but XYZ cannot have the same status because they are shady or something. This is a possible future for autonomous organization regulation, though it does imply a lot of state overhead.

The Grantable Framework: The most permissive framework Bayern discusses is one where ‘personhood’ is a status that can be bestowed upon by anyone who already has that status. I am a person because my parents were people; my friend and I can establish a corporation with legal personhood because we are people, that corporation can spawn 3 other subcorporations because it is a corporation. As Bayern says, “legal personhood is like fire: its capabilities can be granted by anyone who already has it.”

Maybe this still feels wrong, because regardless of how an entity is granted personhood, it just feels weird for a software system cloaked by a corporate shell to have no human oversight. Here, too, Bayern notes that most operating agreements for LLCs are not publicized.

What’s the Use?

An important, but unaddressed question, is what the point of that autonomous organization would be? Bayern notes that fear of the uncertain or possible negative externalities shouldn’t prevent us from allowing this legal technology to exist (even the LLC only really dates back to the 70s), as we can always strictly regulate that specific negative externality. I’m sympathetic to this argument, but Bayern doesn’t fully articulate what the potential positive externalities are for autonomous organizations. He highlights some higher-level categories where maybe you could make use of autonomous organizations (autonomous charities or inheritances that disburse based on an algo), but is somewhat light on examples of what they would do, or why being autonomous entities would be critical to their success, or even why this would be useful socially. We’ll have to cut him some slack, given this was all written prior to modern LLMs and ‘agents’ (who tf knows what that even means).