Radically different
Researchers from the University of Pennsylvania and New York University have received a $4 million grant from the Wellcome Trust to develop an AI-driven platform to train mental health clinicians using simulations of real patients.
Within the next two years, the partners will work on the development of the STELLAR platform, which stands for Steering-Vector Enhanced LLM Agents for Realistic Digital Twins in Mental Health. The platform will create digital twins of patients that trainees can use to practice conducting clinical interviews and evaluating psychiatric symptoms.
“STELLAR brings together behavioral data, clinical expertise, and AI to ask a very practical question,” said Sharath Chandra Guntuku, PhD, associate professor of computer and information science at Penn Engineering. “Can we build training tools that better prepare clinicians for how varied and complex patients are?”
Preparing future mental health clinicians for clinical interviews can be challenging as patients will often report overlapping symptoms that shift over time and subjective experiences that can be expressed differently by each individual. STELLAR will give trainees an ethical option for trainees to simulate interviewing patients with a broad range of symptoms, backgrounds, and clinical scenarios.
“In psychiatry, the details of symptom experience matter: how someone describes distress, how symptoms overlap, how severity changes over time, and how context shapes the clinical interaction,” said Raquel E. Gur, MD, PhD, professor of psychiatry, neurology, and radiology at Penn’s Perelman School of Medicine.
Patient simulations will be created drawing from clinical data from the Philadelphia Neurodevelopmental Cohort, a repository including psychiatric assessments and clinical interviews from thousands of young people created by Penn Medicine and the Children’s Hospital of Philadelphia. Rather than copying individual patients, the simulations will create composites based on real-world data for clinicians to practice realistic conversations in the context of a clinical interview.
This will allow trainers to precisely control the symptoms students encounter, their intensity, and how they interact with each other. For instance, a trainee may practice interviewing a patient with mild anxiety and another whose anxiety overlaps with depression or psychosis to learn how to distinguish the differences in presentation between both.
Because many mental health symptoms manifest beyond formal clinical settings, the platform will also be trained using data from social media platforms, where people discuss mental health symptoms in everyday language.
“Patient simulations will only be useful for clinician training if they are grounded in real clinical speech and evaluated as clinical interactions, not just plausible AI dialogue,” said Neville Ryant, PhD, researcher at Penn’s Linguistic Data Consortium. “[Our] role is to bring speech and language science into the core of the project: adapting speech-recognition tools to clinical interviews, creating high-quality transcripts and annotations, and helping evaluate both what the simulations say and how they say it. That includes assessing the language generated by the models, the naturalness of synthetic voices, how well those voices reflect target speech patterns, and the behavior of the avatar during real trainee interactions.”
To ensure the conversations are realistic, respectful, and useful to trainees, the team will involve people with lived experience of mental health conditions as well as family members and caregivers to provide their perspective into the evaluation process. Their feedback will help researchers assess the accuracy of simulations, avoid stereotyping patients, and prepare trainees for complex and nuanced clinical conversations with real patients.
“The promise of this approach is that we can move beyond stylized and potentially biased simulations,” said João Sedoc, PhD, assistant professor of technology, operations and statistics at NYU’s Stern School of Business. “If we can create digital patients that simulate controllable plausible symptom expression and responsibly evaluate, we can augment current clinician training practices with the kinds of conversations that are essential to better mental health care.”
The post Virtual Patients Will Train Future Mental Health Clinicians appeared first on Inside Precision Medicine.
Background: Digital remote monitoring technologies, including smartphones and wearables, offer promising avenues for early detection of psychosis relapse. However, selecting devices that are acceptable to participants and produce high-quality data remains challenging. Objective: The aim of this nested pilot study was to assess the acceptability and data quality of 3 commercially available wearable devices in people with psychosis recruited to the CONNECT cohort study. Methods: Participants recruited to the CONNECT study before July 31, 2024, were included in the pilot study and selected 1 of 3 wearable devices: a Fitbit Charge 5, Samsung Galaxy Watch 5, or Apple Watch SE. Baseline demographics were compared between device groups. Acceptability of devices to participants was assessed through a Wearable Device Satisfaction Questionnaire after 3 months of use, with the proportion of positive responses to each question calculated and compared. Data completeness was also assessed by calculating the number (and percentage) of valid days of step count, heart rate, and sleep data, and comparing between groups. Data quality was assessed through summarizing the amount of troubleshooting required, additional metrics available from the wearables, and continuity of data completeness by calculating the proportion of participants with at least 3 days of heart rate data per week for the first 20 weeks of follow-up. Predefined criteria were used to determine the next steps for the wider CONNECT study: if one device was superior, this would be selected; if none were found to be superior and the Fitbit was found to be noninferior, then Fitbit would be retained. Results: Of the first 107 participants recruited to CONNECT, 105 were included in the pilot study evaluation. The Samsung Galaxy Watch was selected most frequently by participants (46/105, 43.8%), followed by the Apple Watch (27/105, 25.7%), and Fitbit Charge (23/105, 21.9%). Differences in participant demographics were observed across device groups. Self-reported acceptability after use did not differ substantially between devices. However, in terms of data completeness, the median proportion of valid heart rate data days was significantly lower for Samsung Galaxy (median 31.2%, IQR 8.5%-46.0%) compared to Fitbit (median 80.1%, IQR 26.7%-95.0%; =.003) and Apple Watch (median 49.3%, IQR 21.5%-86.0%; =.02). There was no significant difference between Fitbit and Apple Watch. Similar patterns were observed for step count and sleep data. The Samsung Galaxy Watch required more frequent troubleshooting for data flow issues and lacked additional physiological metrics, available from the other devices. Conclusions: Due to comparatively lower data quality and technical performance, the Samsung Galaxy Watch was discontinued for use in the subsequent phase of the CONNECT study. The study highlights the importance of incorporating nested evaluations of devices in long-term research.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/15a98cce4ebe0d50a261a2b9f900f482" />
Artificial intelligence is transforming what is possible in agriculture, but industry leaders should be wary of investing in AI without first laying the groundwork.
The use cases are promising, especially for an industry navigating volatile fertilizer costs, unpredictable weather, and margins that leave little room for error. Research shows AI-enabled predictive models can improve crop yield by 26%, reduce water use by 41%, and cut chemical usage by 33%.
However, what AI vendors usually won’t tell you is that these solutions are only effective if you have a clean, solid data foundation. However, at Reltio, we have experience in this area, including leading technology strategy at a major agricultural distributor and building a data platform used by enterprises worldwide–we’ve seen it first hand.
Vendor conversations in agriculture tend to follow a familiar pattern. The pitch leads with grand promises around using AI to monitor crop health in real time, optimize irrigation, and squeeze more yield from every acre.
The promise is compelling, but what rarely comes up is the question of whether the data foundation underneath those promises is accurate and complete. If not, there is a real and significant risk that AI will generate misleading outputs that seem authoritative but inspire action that is, at best, counterproductive.
For instance, a yield prediction model fed inconsistent historical data will generate imprecise forecasts. Similarly, a precision irrigation system drawing on fragmented sensor data will make watering decisions that waste resources instead of saving them.
In each case, the AI is failing because the data it was trained on was not sufficient to produce trustworthy outputs. In agriculture, every AI hallucination is a liability, and the likelihood of error is high.
The data landscape across a modern agricultural operation or a large distributor serving thousands of growers is extraordinarily complex.
Modern farming environments make extensive use of IoT devices and machinery. Irrigation systems are automated, tractors navigate fields autonomously, and drones capture field imagery at scale.
However, machine data is disparate by nature. Add in external sources, including weather feeds, U.S. Department of Agriculture data, and third-party market information, and the question of how you bring all of it together into something coherent becomes a significant undertaking.
Agricultural AI also needs to understand more than just customer attributes; it needs to understand the land: GPS coordinates, farm boundaries, field blocks, and soil variation across a single property. Where do you apply fertilizer, and at what rate, and in which specific area of the farm? Not all parts of a field are the same, and an AI system that treats them as if they are will produce recommendations that are at best imprecise and at worst damaging.
There is also a compliance dimension due to the chemicals and the responsibility involved. Operational AI in agriculture needs significantly more checks and governance than it might in a lower-stakes environment. When a flawed recommendation gets acted upon in the field, the consequences can be severe.
Data readiness is the difference between AI delivering on its promise vs. a “garbage in, garbage out” scenario. Fundamentally, being ready for AI means having a data model that accurately reflects how the business operates.
For a company like Wilbur-Ellis, a 104-year-old, family-owned agricultural distributor, that means understanding who your customers are, which fields they farm, which inputs they need, which suppliers those inputs come from, what they paid last season, and how all of that connects to margin. That information needs to be current, consistent, and accessible across the organization, rather than locked in separate systems that were never designed to talk to each other.
Similarly, for farming operations themselves, data readiness means having a reliable, connected picture of what is happening across every field: soil health records, input application histories, yield data from previous seasons, equipment performance, and real-time sensor readings from irrigation systems.
Governance matters just as much as structure. Prices change, relationships evolve, and suppliers come and go. An AI system drawing on data that was accurate six months ago but has not been maintained will make recommendations based on a version of the business that no longer exists.
The good news is that the path to data readiness is feasible. It starts with a strong data model: a single, governed source of truth that connects customers, suppliers, products, pricing, orders, and margins in a way that reflects how the organization operates.
From there, it requires data pipelines fast enough to deliver insights when decisions need to be made, governance frameworks that keep that data trustworthy over time, and security controls that ensure sensitive commercial information is accessible to the right people under the right conditions.
This is precisely the challenge that Reltio, an SAP company, was built to solve. Reltio enables companies to unify their fragmented data so AI agents and systems can operate from a complete picture of the business. Reltio builds a trusted system of context, known as the context intelligence layer, that brings all entities, relationships, rules together under one roof and makes business data easy to access and interpret.
For Wilbur-Ellis, building that trustworthy data foundation has meant being able to ask more complex questions and trust the answers, which is the precondition for any AI system to be genuinely useful.
The question worth asking before the next AI conversation is not whether the use case is promising. It almost certainly is. The question is whether the underlying data foundation is strong enough to make the output trustworthy.
Agriculture has always required its leaders to make high-stakes decisions under uncertainty, and AI offers the genuine prospect of making those decisions faster and better informed. That prospect is only achievable for organizations that have done the foundational work first, and the businesses that will get the most from AI are the ones investing in that foundation now.
This content was produced by Reltio. It was not written by MIT Technology Review’s editorial staff.
Nature Medicine, Published online: 30 June 2026; doi:10.1038/s41591-026-04540-1
Author Correction: Digital AVATAR therapy for distressing voices in psychosis: the phase 2/3 AVATAR2 trial
Nature Neuroscience, Published online: 26 June 2026; doi:10.1038/s41593-026-02353-6
Amygdalar acetylcholine signals dynamically represent salience. Chen et al. show that striatal D1 and D2 neurons oppositely control these signals via basal forebrain cholinergic pathways to shape salience-associated learning.