Arbitration begins with a choice to leave court. Parties choose the governing law, the seat, the rules, the number of arbitrators, the scope of discovery, the form of hearing, and whether the decision must give reasons.
They may select a retired judge, an industry specialist, a religious tribunal, or a decision-maker with no legal training. The Federal Arbitration Act enforces those choices because arbitration is a matter of consent. The same rule applies when the parties expressly select an AI tribunal.
The FAA is not a human-arbitrator protection act. If competent parties knowingly agree that a defined AI system will decide a defined class of disputes under defined procedures, courts should enforce the bargain they made.
The FAA protects the bargain
Section 2 of the FAA makes a written arbitration agreement enforceable, subject to the same grounds that can revoke other contracts. Section 4 directs that arbitration proceed “in the manner provided for” in the agreement. Section 5 says the parties’ agreed method for appointing the tribunal must be followed.
The Supreme Court describes consent as foundational. In Lamps Plus, Inc. v. Varela, the Court emphasized that the parties may specify with whom they arbitrate and the rules governing the proceeding. In Morgan v. Sundance, Inc., the Court rejected an arbitration-specific rule invented by judges, even though the rule was intended to favor arbitration.
That logic cuts both ways. A court cannot invent a special anti-AI condition that appears nowhere in the agreement or the statute. It must treat the arbitration agreement like a contract and enforce the procedure the parties selected.
The statute contains no human-presence rule
Congress enacted the FAA in 1925, and Section 5 uses the pronouns “he or they.” Those words reflect the technology of the period. They do not impose a substantive requirement that an arbitrator possess a human body.
The Act supplies a direct command. Identify the agreement. Determine whether it is valid. Enforce the procedure it selects. Technology can change how a contractual function is performed without changing the function itself.
The controlling question is concrete: what tribunal did the parties choose? Express selection of an AI tribunal satisfies the consent principle. A human arbitrator’s undisclosed delegation to AI violates it.
Put the AI in the agreement, not behind the curtain
Hidden AI use presents a straightforward contractual failure. If the parties selected a human decision-maker, they must receive that person’s judgment. A machine-generated result passed through as the human’s own violates the chosen procedure.
Express AI arbitration starts with disclosure. The agreement identifies the AI role, governing rules, tribunal composition, record, procedure, form of decision, review rights, and execution boundary. It fixes those terms before the result is known. The provider proves that it used the promised process.
The parties select a complete decisional protocol. That protocol includes the governing rules, admitted record, tribunal configuration, instructions, validation, review, release, and execution controls. A model name alone cannot define the bargain. The provider must preserve the system the parties selected.
Enforce the bargain
Human dispute resolution is too slow and expensive for a vast category of real disputes. Software can apply bounded rules to structured records at a cost proportionate to those disputes. Arbitration gives parties the power to choose that procedure in advance.
The agreement should put the technology in plain view and bind it to a procedure the parties can understand, choose, test, and challenge. When competent parties make that choice, the FAA’s answer is straightforward: enforce the bargain.