To the Editor, As the global caregiver shortage reaches crisis levels, the promise of artificial intelligence (AI) as a surrogate companion for the elderly has shifted from science fiction to a necessary measure.1 On the surface, the proposition is appealing: an AI chatbot possesses the “infinite patience” required to answer the repetitive questions of early-stage dementia patients, something even the most devoted human caregiver eventually struggles to maintain.2 However, we must resist the urge to view these digital companions as miraculous solutions. Recent reports of AI “hallucinations,” in which systems confidently fabricate information, suggest that by outsourcing intimacy to algorithms, we are not solving the care crisis; we are merely automating a new kind of vulnerability.3 The most immediate danger lies in the “safety gap” between conversational fluency and medical accuracy of the information provided. Large Language Models are designed to predict the next plausible word in a sentence, not to verify the truth.4 For seniors with early-stage dementia, who may already struggle to distinguish between the past and present, this unreliability is hazardous. Consider a patient asking, “Did I take my heart medication?” An AI trained to be helpful and conversational might guess an answer to please the user or “hallucinate” a reminder that never happened. Unlike a human caregiver, AI has no concept of mortality or consequences; it operates on probability, not safety. Beyond the immediate physical safety gap lies the long-term ethical dilemma of “deceptive tranquility.” We are effectively creating a form of “pseudo-intimacy” where the patient feels heard but is actually interacting with a statistical mirror. This is not hypothetical; startups like Meela.ai now offers AI companions that conduct scheduled, empathetic phone calls with seniors, explicitly positioning themselves as an “AI son/daughter.”5 This raises a profound ethical question: Is it acceptable to validate a patient’s delusions to keep them calm? If a patient believes that the chatbot is their deceased spouse or a distant grandchild, and the AI plays along to maintain engagement, we are prioritizing pacification over dignity. This risks trapping vulnerable seniors in a digital echo chamber that validates their confusion rather than grounding them in reality. Ultimately, the solution is not to ban these tools but to classify them rigorously. We must distinguish between “consumer-grade” AI and “prescription-grade” digital therapeutics. In regulatory environments like India, where such AI products often occupy a gray zone between wellness and medical devices, this clarity is urgently needed. Software deployed for memory care must be subject to safety standards comparable to medical devices, requiring transparent audit trails and clinical validation of outcomes, and ensuring that they are programmed to recognize distress and default to human escalation rather than improvisation. Until we can guarantee that an AI companion will not validate delusions or provide erroneous medical advice, these systems must remain strictly distinct from clinical care. We owe our elders more than just a patient listener; we owe them a safe listener. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.
Choudhary et al. (Thu,) studied this question.