Healthcare & care
Dr. Kristine Lee wants AI to help clinicians hear more of the patient’s story.
A physician’s account of documentation technology points toward care support that goes beyond a shorter note.
When a patient describes difficulty reaching an appointment or getting enough food, the care team needs a way to record the concern and connect it to follow-up. Documentation is one step in that work.
The need: useful information that reaches the person who can act
Healthcare includes a great deal of communication: what a patient wants to explain, what a professional needs to record and what another team needs to understand before the next step. Information can be accurately captured and still fail to lead to useful action if nobody is responsible for following it through.
AI research needs to distinguish question preparation, administrative assistance and clinical documentation. Each task has its own purpose, data environment and review process.
Treating these as separate tasks makes the opportunity clearer. The goal is to support understanding and coordination while preserving the patient’s voice and the responsibilities of the professionals delivering care.
A physician's next question for AI
In an August 27, 2026 account, Permanente Medicine highlighted Dr. Kristine Lee's work on ambient AI documentation. Lee, a physician leader at The Permanente Medical Group, described the potential for deliberately designed systems to capture social needs raised during a visit. The institution frames this as a next test for the technology. It is not evidence that a documentation tool has resolved those needs or improved patient outcomes. Read the health system's account.
That distinction matters: a better record can support care, but a recorded transportation problem does not create a ride.
Follow the information through the workflow
Before an appointment, a general-information tool would help someone organize questions they intend to ask. It would distinguish the person’s own observations from a generated suggestion. The person would decide what to bring to the appointment, and clinical questions would remain with the care team.
Within an approved institutional system, a documentation tool would prepare a draft from authorized material. The responsible clinician would check what was captured, what was omitted and whether the wording changed the meaning. The system would need a correction process that preserves the distinction between source information and model inference.
For an administrative application, a team would use assistance to locate a current internal procedure or prepare a reviewed explanation of a service. The source and version would matter. A plausible but outdated answer could send someone toward a process that no longer exists.
A better note is one step in a larger service
If a record identifies a practical need, follow-through requires an appropriate person, an available service and a way to determine whether help was actually received. Model output cannot supply those resources by itself.
Evaluation would follow the particular workflow. A documentation study would examine accuracy and correction effort. A service-coordination study would examine whether an identified need leads to an appropriate referral and whether help is received. These outcomes would be reported separately.
Staff would also need professional instruction specific to their roles, including how to question an output and report a recurring problem.
the program’s proposed public infrastructure would broaden access to research capacity and supporting tools. It would not remove the need for staffed services, local resources or clinical accountability. American Super Intelligence has no announced relationship with Lee or Permanente Medicine.
How Liberty will offer useful foundations for care support
LibertyGPT will support reviewed general information and preparation for a conversation with a qualified professional. Institutional applications built with Liberty OSS will provide a separate route for research and approved internal work. Liberty Gov will support authorized government health missions.
Open model access will let qualified teams examine performance on their actual tasks and adapt models within appropriate data controls. It will not, by itself, validate a clinical use or establish the permissions to process patient records. Those decisions will remain part of the responsible institution’s evaluation and authorization.
the program’s education program will prepare users and technical teams to work within those boundaries. Liberty will help through accessible services, adaptable models and shared learning, while the people and organizations delivering care will retain responsibility for connecting useful information to real support.