Voiko tekoälyn käyttäminen välitystuomion laatimisessa muodostaa perusteen välitystuomion kumoamiselle?

This thesis asks whether the use of artificial intelligence in drafting an arbitral award can constitute grounds for setting the award aside. The use of AI by arbitrators is no longer hypothetical, and the first court challenges have already appeared, yet the existing arbitration framework does not regulate it directly. The thesis examines how the established rules of international commercial arbitration apply to this new practice. To answer the main question, three further questions are addressed. The first is whether AI-assisted drafting amounts to a delegation of the arbitrator’s mandate. The second is where the line falls, that is how much assistance is permitted before the arbitrator’s personal mandate is breached, and whether errors in the award are themselves enough. The third, addressed more briefly, is how the use of AI should be disclosed and regulated. The hypothesis is that the limits developed for arbitral secretaries, who may assist the arbitrator but may not take over the decision, provide the measure for distinguishing permissible assistance from impermissible delegation. The thesis adopts a doctrinal method. It works mainly at the transnational level, taking the New York Convention and the UNCITRAL Model Law as a common framework, and draws on national arbitration laws, institutional rules, the EU AI Act, the first reported cases, and the soft law guidelines that arbitral institutions have recently issued. Because arbitral proceedings are confidential and awards are rarely published, the analysis relies on these sources rather than on empirical data about how arbitrators actually use AI. The thesis first sets out the foundations of arbitration, in particular the arbitrator’s adjudicative powers and the established limits on delegating them to a secretary. It then examines the grounds for setting aside an award under the New York Convention, focusing on the right to be heard, the composition of the tribunal and the arbitral procedure, and the public policy exception. It then turns to AI itself, including its limitations such as bias and hallucination and the framework of the EU AI Act, before analysing the first cases and the institutional guidelines. The thesis concludes that the use of AI in drafting an award can be a ground for setting it aside, but not in every case and not by reason of the use of AI alone. What matters is whether the arbitrator has crossed the line from using AI as support into delegating the decision itself. Where AI is used only to assist, the position is no different from the use of a tribunal secretary. Where the reasoning or the decision is in substance produced by the AI system, the arbitrator has not exercised the mandate, and the award becomes open to challenge. The final decision, and the reasoning behind it, must remain with the human arbitrator.