This randomized controlled trial examines the mechanisms of conversational AI-based interventions for early dementia prevention in socially isolated older adults.

Social isolation is a well-established and modifiable risk factor for cognitive decline and dementia. The completed I-CONECT trial (NCT02871921) showed that frequent, one-to-one, semi-structured conversational engagement delivered remotely by trained human interviewers improved cognitive outcomes among socially isolated older adults with mild cognitive impairment. Two limitations have constrained translating that approach into a scalable preventive intervention: its reliance on trained human interviewers limits scalability, and the behavioral mechanisms responsible for its cognitive benefits remain incompletely understood.

AI-CONECT addresses both limitations by delivering the conversational intervention through a conversational artificial intelligence (AI) voice chatbot rather than a human interviewer, and by testing a hypothesized behavioral mechanism: that the intervention increases participants' social self-efficacy (confidence in their ability to engage socially), which in turn increases real-world social engagement.

Inclusion Criteria:

• Aged 75+
• Clinical diagnosis of MCI, or MMSE ≥ 24 or MoCA ≥ 18 (i.e., early MCI)
• Identified as socially isolated by at least one of the two criteria: (1)score ≤ 12 on the 6-item Lubben Social Network Scale (LSNS-6); (2) engages in conversations lasting 30 minutes or longer, no more than twice per week, per subject self-report
• GDS-15 (15-item Geriatric Depression Scale) at/below 9 (not severely depressed).
• Social self-efficacy mean ≤ 8 (on the 0-10 scale) across the 6-item Social Self-Efficacy Scale at screening, indicating low confidence in social interaction.
• Ability to understand the research consent form.
• Ability to connect a tablet to the Internet.

Exclusion Criteria:

• Diagnosed with dementia such as AD, ischemic vascular dementia, normal pressure hydrocephalus, or Parkinson's disease.
• Schizophrenia, or other major psychiatric disorder defined by DSM-IV criteria.
• Medications: Frequent use of high doses of analgesics; Use of sedative medications except for those used occasionally for sleep (≤2 per week); Use of unstable dosing of Cholinesterase inhibitors (need to be stable dosing for 2 months).

De-identified data will be shared with external researchers under a Data Use Agreement (DUA). The de-identified data may include datasets (outcome scores, adherence rates, chatbot-usage metrics), full conversational transcripts, and audio and video recordings of sessions; all are de-identified (direct identifiers removed) before sharing. De-identified data, analytic code, and summary reports are also available to the NIH/NIA Program Official on request. Data are stored on HIPAA-compliant servers (e.g., MGB Business Dropbox, MGB Azure) and encrypted drives. No identifiable data or Limited Data Sets are shared; the participant linkage key is kept within Mass General Brigham. No individual-level data are deposited in public or controlled-access repositories under the current protocol; future repository sharing would require additional IRB review and approval.

De-identified data and related materials are available upon request; the current protocol does not specify a fixed availability start or end date. Availability and retention follow Mass General Brigham and NIH record-retention requirements.

Requests are reviewed by the study team. Sharing occurs only under an executed Data Use Agreement (DUA) or equivalent research agreement approved by the Mass General Brigham Contracting Office, is limited to de-identified or aggregated data for the analyses described in the protocol, and contractually prohibits re-identification and onward sharing. Data are transferred only via Mass General Brigham-approved secure, encrypted methods.