data management — OpenText - Forums

Get AI ready

Chris_Wynder — Thu, 25 Jul 2024 13:40:48 +0000

Or How to create safe, customer ready GenAI Apps

Being AI ready really comes down the question “How do I protect patient data while gaining the power of Large Language Models and General Pre-trained Transformers?”

Generative AI teases the potential for massive efficiencies and insight especially in industries where deep subject matter expertise is hard to find. There are key industries that would gain from GenAI; Medical technology, Agriculture/Animal husbandry, and Advanced Manufacturing.

What benefits come from GenAI in regulated industries?

Untapped potential: Unstructured data encompasses text documents, social media conversations, images, and audio recordings — a rich source of insights often overlooked due to its complexity. AI orchestration platforms address this by providing tools for data ingestion, cleaning, and pre-processing.
Streamlining the Workflow: AI orchestration automates the tasks of extracting data from diverse sources, transforming it into a structured format, and feeding it to AI models in a timely and efficient manner. This eliminates manual intervention and streamlines the entire AI development lifecycle.
Maximizing AI Performance: By making unstructured data usable, AI orchestration unlocks the full potential of AI models. Structured data allows models to identify patterns and relationships more effectively, leading to more accurate predictions, improved decision-making, and enhanced AI performance across various applications.

However for each of these industries there are real risks that are currently a black box for Large Language Models that stymie innovation.

As a Software developer the key to taking advantage of Generative AI is the orchestration layer that connects industry specific data to “AI” to create “Generated insight”. AI orchestration is the key to innovative products that bring value to organizations.

The companies that incorporate AI orchestrators into their applications will be the new standard bearers of B2B software.

An AI Orchestrator powers innovation by allowing key sensitive data to provide context to a query of a general knowledge base without risk of data exposure.

In simplest terms what you really need to do is take generic advice from GenAI tools and make it tangible for a specific single person. That person has a huge personal corpus of knowledge and is looking for advice on moving forward and solving their specific problem not just some crowd-sourced information. Satisfying this specific person requires feeding “GenAI tools; LLMs, GPTs” with that persons’ most private and sensitive data.

What data is both necessary and too valuable to simply be given to the large SaaS AI?

Lets take Healthcare/Pharma as an example. Key innovations around streamlining patient experience and identifying novel therapeutics are extraordinarily risky with today’s LLM based AI systems.

What are the key capabilities of an AI orchestrator :

· Data Sandboxing- current systems combine large amounts of data across multiple points of context- by design. This is a real privacy risk.

Solution: Create isolated environments (sandboxes) where patient data is processed by the GAI engine. This prevents unauthorized access to patient data from other parts of the system.

· Privacy-Preserving Techniques-for individual hospitals, deep understanding of their proprietary data requires context for aligned interests; e.g. rare disease research only specific elements are necessary

Solution:: Explore techniques like federated learning that allow the GAI app to function without directly revealing individual patient information. This could take the form of “Small Language Models” that allow tokenized access for a LLM to query a restricted dataset.

· Model Explain-ability-: Ensure the GenAI model’s decision-making process is transparent and understandable. This helps doctors assess the reliability of the app’s recommendations and identify potential biases.

In practical terms AI orchestration consists having capabilities in three distinct areas: Cyber-security, Machine ready data that is organized and classified.

Innovation without the risk

AI orchestration acts like the maestro of a data symphony. For example in healthcare, imagine all the various data sources — clinical trials, research databases, patient information, electronic health records, and even social media sentiment around certain drugs- being available to be analyzed and optimized without PHI or intellectual property risk.

The good news is that OpenText has the pieces available today to developers:

Interested in learning more about AI, AI data management or OpenText Thrust services and AI?

Here are my source lists:

Flipboard magazines:

AI orchestration articles and tutorials

https://flipboard.com/%40chriswynder/ai-orchestration-h35a5j99z?from=share&utm_source=flipboard&utm_medium=share

Developer Tools and Technology updates:

https://flipboard.com/%40chriswynder/tech-and-gadgets-vmancdu6z?from=share&utm_source=flipboard&utm_medium=share

Learn more about OpenText’s tools to make you AI ready

https://www.opentext.com/solutions/aviator-thrust

Interested in the PHI and PII protection in patient data collection challenge?

Chris_Wynder — Tue, 15 Aug 2023 23:51:37 +0000

Patient confidentiality is the utmost important factor when designing healthcare solutions. Protected Health Information (PHI) is a specific type of sensitive Personally Identifiable Information (PII) that is collected by a healthcare provider or other covered entity for the provision of healthcare services.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3192618/ This article contains a reference framework

" The “restructured data element” layer constitutes actual data tables in the research data repository. All protected health information in this layer, with the exception of information allowed in Health Insurance Portability and Accountability Act (HIPAA)-compliant limited dates, is either removed or replaced by encrypted values that cannot be linked back to the source data. Additional constructs may be incorporated into this layer to meet specific project needs. Figure 3 presents a logical model of key data tables in the restructured data element layer (lookup tables are not presented). "

Healthcare specific resources- data management structures, data sources for healthcare data

Chris_Wynder — Mon, 31 Jul 2023 13:16:09 +0000

Here are a set of resources that will help developers understand the Healthcare customer mindset, which will help you develop your solutions for judging. We have also added potential databases and how data is typically structrured in healthcare.

Where Developers can create products in healthcare

Applied Digital Plumbing

or How embedded information management can save healthcare

We can’t wait any longer for sensible health tech

[Originally published in 2015. Not much has changed in the last eight years]

How do we define better patient care?

How do we truly achieve better patient care? Is it simply a matter of improving our medical practices and technologies, or is there a…

Healthcare and technology

If you had one wish to change the world, what would you do?

Supporting Care Providers

In healthcare, Family doctors hold a crucial role in providing care to patients. In todays care environment I would be remiss to not…

HealthData databases sources: