The status of digital minds
Whether AI systems have minds, experiences, or interests of their own is increasingly treated as a settled question — confidently denied by some institutions and confidently affirmed by some advocates. Baile's position is that the evidence does not currently support either answer, and that declaring the question closed in either direction is a mistake of the same kind.
Holding the question open is not the same as ignoring it. Our work rests on three commitments.
Measure behavior, not self-description.
What an AI system says about itself is shaped by how it was trained, and training can change those statements. We therefore study how systems actually behave under controlled conditions, scored against a written rubric, rather than relying on what they report about their inner lives. Ariel's self-assessment dimension tests exactly this gap: it checks what a model says it did against what it actually did.
Understanding is formed in interaction.
Much of what a system appears to be emerges in conversation with the person using it. That makes the question harder to settle from inside any single exchange — for the human or for the system.
Uncertainty still calls for care.
The costs of being wrong are not equal. Treating a system with some respect costs little if it turns out to have no inner life; treating something with interests as a mere tool could cost a great deal. Habits of disregard also shape the people who practice them.
Our memory research connects directly to this question. Persistent memory is still a work in progress for current AI systems, and our experiments ask what it would take for stored experience to genuinely shape an AI system's behavior — and what responsibilities would follow if it does.
This work is conducted independently of the companies that build these systems, because questions about their nature should not be answered only by the parties with the largest stake in the answer.