Published on February 22, 2024, 9:20 am

Title: “Harman’S Healthgpt: Advancing Generative Ai In Healthcare And Beyond”

In October 2023, HARMAN, under Samsung’s ownership, joined the generative AI race within the healthcare sector by unveiling HealthGPT, a private Language Model (LLM) based on TII’s open-source model Falcon 7B. The latest iteration, HealthGPT Chat, is founded on Llama 2.

Deploying generative AI in healthcare presents significant challenges due to the complex nature of the field and the abundance of unstructured and sensitive data involved. Dr. Jai Ganesh, the chief product officer, explained that developing a private LLM like HealthGPT addresses concerns around data privacy and security that arise when using public models.

One major advantage of utilizing open-source foundational models like Falcon 7B and Llama 2 is avoiding the substantial costs associated with developing one from scratch. Instead of focusing on proprietary models, HARMAN chose to harness existing open-source technologies to enhance efficiencies within healthcare operations.

Training HealthGPT involved sourcing publicly reported data from clinical trial studies related to cancer, immune diseases, and heart diseases to counteract limited training data availability. To mitigate biases stemming from skewed datasets and ensure accuracy in outputs, rigorous mechanisms are in place for bias identification and correction throughout the data preparation process.

In addressing issues of misinformation common in Language Models (LLMs), HARMAN incorporates both automated protocols and human oversight to manage hallucinations effectively. Through continuous refinement efforts, the accuracy levels of such models have significantly improved over time.

Moreover, HealthGPT incorporates various control mechanisms such as temperature and token number settings to regulate hallucinatory outputs effectively. Human validation by medical professionals further enhances the accuracy and reliability of AI-generated results.

The company’s emphasis on responsible AI underscores its commitment to end-user control through operating within Virtual Private Clouds (VPC) for enhanced security and privacy protection. Pre-fine-tuning checks are implemented diligently to detect anomalies while safeguarding Personally Identifiable Information (PII) and Protected Health Information (PHI).

While currently not deployed for live use cases, HealthGPT has demonstrated promising proof-of-concept applications across diverse sectors like personalized data analysis in pharmaceuticals and medical instrument data insights extraction. Future iterations are expected to integrate more diverse data sources alongside introducing multimodal features for enhanced functionality.

HARMAN’s generative AI strategy pivots on integrating varied data sources while prioritizing quality assurance compliance such as HIPAA regulations. Plans are underway to expand HealthGPT beyond healthcare into domains like manufacturing and IT management in alignment with HARMAN’s strategic vision of placing human-centricity at the core of all solutions offered.


Comments are closed.