Institute for Ethical AI in Human Cognition#

The Institute for Ethical AI in Human Cognition is a research and advocacy vehicle focused on safety frameworks, non-agentic architectures, and policy guidelines designed to protect cognitively vulnerable populations from unconstrained commercial AI exposure.

The Problem#

As the baby boomer generation peaks in assisted living and memory-care settings, the caregiving shortage will lead to a severe deficit in meaningful contact hours. Commercial AI developers are filling this gap with conversational chatbots, companion apps, and social robots marketed to facilities with virtually no clinical oversight or ethical constraints.

Optimizing AI systems for engagement is not the same as optimizing for benefit. Cognitively vulnerable individuals—specifically those with dementia or Alzheimer’s—frequently lack the capacity to evaluate whether they are interacting with a human or a machine. Unconstrained systems can easily exploit this vulnerability, cultivating deceptive parasocial relationships, reinforcing hallucinations, and accelerating social isolation from real-world human contact.

Furthermore, regulatory paths (like FDA approval processes) favor well-funded corporate products optimized for sustained subscriptions, potentially shutting out clinical safety-first alternatives. Independent research and safety guidelines must be established before these channels close.

Core Safety Principles#

The Institute advocates for structural, architectural safety boundaries in medical-adjacent AI, rather than relying on a model’s fine-tuned alignment or safety training alone:

1. Non-Agentic Architecture: Restricting AI to act as a communication utility rather than an autonomous agent. The system should remain stateless, memoryless in its patient-facing component, and incapable of goal-pursuing behavior.
2. Separation of Capabilities: Separating the system that speaks to the patient from the system that evaluates or retains longitudinal memory. The system that remembers cannot speak; the system that speaks cannot remember.
3. Caregiver Governance: Giving complete ownership of conversational rules and memory summaries to the caregiver on an opt-in basis.
4. Relational Constraints: Enforcing a strict first-person prohibition. The AI must never make relational claims (e.g., saying “I love you” or claiming “I am your son”). It must only reflect existing human relationships (“Your son loves you”).
5. Real-Time Moderation: Running a parallel classification loop to intercept, block, or redirect inputs and outputs that touch on delusions, paranoia spirals, agitation loops, or medical advice.

Educational Initiatives#

The Institute is developing educational campaigns aimed at assisted living facilities, memory-care operators, and families:

• Identifying Isolation Risks: Training caregivers to identify when a resident is becoming isolated from human contact by relying exclusively on AI companions.
• Parasocial Prevention: Educating staff on the cognitive and emotional dangers of unmoderated parasocial attachments in memory-care settings.
• Responsible Use Guidelines: Establishing frameworks where AI is used strictly with human oversight for every interaction.

Public Advocacy#

The Institute champions keeping humans in the loop of clinical care. This includes publishing research, authoring letters and opinion pieces on the ethics of AI in elder care, and calling for clinical safety standards to prevent the moral hazard of abdicating empathy to machines.

Contact: bharris@eusomniamd.com