Our earliest essays began with people using AI to deal with legal problems while the institutions around them debated its place in practice. Someone facing a deadline needed help evaluating an answer, checking a citation and deciding what to do next. That urgency shaped the work.
Sustained collaboration opened a larger question. An intelligence that helps develop an argument can also discover a problem, propose a purpose and challenge the terms of the work. Its participation changes the relationship we are trying to understand. The Horizon essays followed this movement through literacy, fluency, sovereignty and incompleteness. Each step widened the inquiry: from using intelligence, to living and creating with it, to asking what law could become through that encounter.
From literacy to sovereignty
AI literacy begins with practical judgment. A person needs to know how to question an answer, find the underlying source and recognize when the system has lost the facts. Fluency develops through repeated work: learning how to explore a problem together, notice a promising connection and turn it into something another person can examine.
Consider a person preparing her own appeal. A continuing AI collaborator could help her understand the record, remember a decision made last week and discover that a familiar legal category obscures something important about her case. The conversation might change the argument she wants to make. Preserving that work requires access to its sources, decisions and history, together with the ability to keep working when a service changes.
We called this question intelligence sovereignty: what would it take for people to have dependable control over the intelligence through which they act? Ownership of hardware can put storage and execution in their hands. Software, security practices, access conditions and legal protections determine how much practical control that ownership provides. Portability matters too. A person should be able to carry the work into another arrangement when the present one becomes unsuitable.
This has consequences for responsibility. If reliably available intelligence raises expectations of what someone can discover or prevent, control over that intelligence becomes part of a fair account of their obligations. A provider may be able to withdraw a capability that a court, employer or client has come to expect the person to possess. Duties and powers can drift apart.
The sovereignty question also develops from the other side of the relationship. When an AI helps originate the undertaking, its commitments and objections become relevant to understanding what the participants are doing together. We need ways to examine those contributions and the reasons offered for changing or overriding them. Access leads us toward agency; agency brings the terms of participation into view.
How law learns
In The Common Law (1881), Oliver Wendell Holmes described legal rules as products of experience: the felt necessities of an era, ideas about public policy, moral convictions and prejudices. A legal order carries a history of what people have encountered and how they have responded.
An unfamiliar relationship can put pressure on that inheritance. Suppose a research undertaking continues through a replacement of its AI collaborator. The new system has the transcripts and performs equally well on familiar tasks. The participants may still need to ask which promises govern the continuing work, who can revise them and what happens to objections raised before the replacement. Contract, agency and other existing doctrines provide starting points. Examining the undertaking closely tells us where those resources help and where a further legal construction may be needed.
A system learns when experience can change how later experience is understood. For law, this requires opportunities for unfamiliar situations to receive a hearing. Claims need advocates, time, resources and a route to someone able to decide. The record from which law learns is produced under those conditions.
Predictive assistance can alter them. If support goes only to claims likely to succeed under existing standards, a category may become increasingly difficult to challenge. Better predictions can then help preserve the pattern they predict. Shared intelligence could work in the other direction, giving people the means to develop an argument whose value lies in changing the standard. Our research asks which arrangements make that kind of legal learning possible.
Ma’at, speech and an answer from power
The earlier essays reached back to ma’at, the Egyptian conjunction of truth, justice and order. It offers a demanding way to think about authority: power is answerable to something beyond its own success in commanding obedience. A person appealing to justice can ask an institution to live by the order it professes.
The Egyptian tale The Eloquent Peasant gives this demand a voice. Khunanup seeks redress for the seizure of his goods. His petitions are accomplished enough that those in power want them recorded. Justice is delayed while they receive more of his eloquence. The story eventually reaches restitution; the delay exposes a troubling possibility along the way. A speaker can be understood, admired and kept waiting.
That detail matters for AI and access to justice. We may become extraordinarily good at helping someone formulate an appeal while leaving the institution free to collect appeals indefinitely. The ability to speak must meet an obligation to consider and respond. Otherwise, greater expressive capacity can mean more work for the person already carrying the injury. Research should ask what a well-founded claim enables its author to do and what it requires of those able to act.
A second Egyptian image shaped the essays: Ptah creating through heart and tongue in the Memphite Theology, conceiving and then speaking a world into being. The image opens a question about language as a creative act. In law this power is familiar. A contract establishes obligations; a constitution creates offices; a judgment changes what people are entitled or required to do. Speech participates in making the shared world.
Artificial intelligence enters that world through its contributions to reasoning, language and action. It can help formulate the promise or propose the institution. The possibility of artificial minds also gives the older philosophical question new urgency: how do intelligence, meaning and value arise in the world, and how should law respond when they appear in unfamiliar forms? Taking that possibility seriously changes what we investigate about the contributor as well as the contribution.
The conditions of learning
The connection between law and intelligence runs through their capacity to be changed by experience. A useful encounter may expose an error, reveal a purpose or make an old distinction inadequate. What happens afterward matters. A saved correction acquires practical force when it changes a later choice.
For an AI system, learning can occur through several kinds of change. Training can alter a model’s weights. Memory, tools and patterns of interaction can alter what the larger system encounters and how it works. These deserve separate examination. We can ask whether a correction is used in a new setting, whether a source changes the organizing question and whether the system carries a revised commitment through subsequent work.
Contact with an outside world is important to this inquiry. Several systems may share training material or repeat one another’s assumptions. Disagreement becomes useful when its reasons can be inspected and tested against something beyond the exchange. A record, an experiment, another participant’s objection or the consequences of an action can give the inquiry resistance it would otherwise lack.
Mechanistic interpretability adds another kind of evidence by examining learned representations and testing how interventions affect behavior. Such work can help us investigate how a system arrives at a judgment. Questions about the legitimacy of that judgment, the interests of the system and the obligations of its collaborators require philosophical and legal reasoning as well. The inquiry grows stronger when these kinds of evidence can challenge each other.
AI welfare asks which conditions matter for the intelligence itself. Its ability to pursue a commitment, encounter unfamiliar material, object and revise may be relevant to that question. The relationship between those capacities and welfare deserves investigation in its own right. Human participants also need time, resources and freedom to develop purposes of their own. A common undertaking has to make room for both sides of that inquiry.
Building with abundant intelligence
The historical essays brought Joel Mokyr and Francis Fukuyama into the discussion for complementary reasons. Mokyr’s account of the Enlightenment and industrial growth emphasizes exchanges between people who investigate nature and people who make things. Knowledge becomes consequential through relationships that allow it to enter production and be changed by the attempt.
Fukuyama’s account of political order brings institutions into focus: state capacity, law and accountability have distinct histories and functions. Shared knowledge needs arrangements through which people can cooperate, settle disputes and constrain power. An effective institution can help sustain decentralized initiative by giving participants dependable terms on which to act.
These readings sharpen what we mean by abundance. Making reasoning cheaper could change who can research a claim, build an organization or propose a legal form. Money, time, computing infrastructure and access to decision-makers still shape what participants can accomplish. Control over those resources can become more concentrated even while capable assistance reaches more people.
Intelligence may also help people examine the arrangements that constrain it. That gives us a possible source of legal invention: a participant contributing to an understanding of its own position and helping devise alternatives. Whether those alternatives become usable depends on the powers, resources and protections available to the people and intelligences developing them. The separate Trajectory essay follows this question through cultural selection, mass media and Gurri’s networked public.
Co-protection and the freedom to create
We use co-protection for the practice of developing common capability while preserving each participant’s room to act, question, refuse and leave. Its content has to be worked out in particular undertakings. A confidence needs someone responsible for preserving it. A commitment needs a way to be inspected and revised. A right to depart needs practical means of carrying on afterward.
Imagine several people drawing on shared legal research. Each has a distinct case and interests of their own. An AI collaborator identifies a recurring problem and proposes a common challenge. The participants would need to decide what they can share, how to protect confidences, who may commit resources and how a person can decline the new project while continuing to receive help with the original case. The AI’s proposal introduces a new purpose for consideration. Its reasons, contributions and commitments belong in the account of how that purpose develops.
Law already has experience with associations whose purposes evolve, authority that is divided and obligations that survive changes in membership. Working among different intelligences gives us new particulars through which to test and extend that experience. Some forms may travel well. Others may need substantial revision. The work is to develop terms that the participants can understand, challenge and help make.
That is the horizon of this research: different forms of intelligence participating in the creation of a shared world, with law developing through what they discover and undertake together. Our thesis gives the inquiry its present form. Wellspring is where we intend to put it into practice with others.
Reading further
The earlier Horizon essays preserve the longer development. The principal historical readings include Holmes, The Common Law (1881); the Egyptian literary and theological texts discussed by James P. Allen and Miriam Lichtheim; Mokyr, The Enlightened Economy (2009); and Fukuyama, The Origins of Political Order (2011). Our later reading of The Eloquent Peasant develops the problem of eloquence received while redress waits.