A Primer On AI Agents In Healthcare
Understanding The Basics, Application Opportunities And Long-Term Potential Of AI Agents
It wasn’t long ago that chat-based generative AI permeated all industries and promised to transform healthcare.
As new solutions are being implemented to increase efficiency and design more effective solutions we are seeing the next phase of AI emerge.
The ‘Calm’ Before The Storm
Agents are just intelligent automations. Today platforms like Zapier allow you to set custom instructions across multiple apps that are triggered by external actions. Responding to emails or updating your CRM without you implicitly doing so has been possible for a few years now.
The jump we’re about to experience takes the brains of an LLM and integrates it across your entire workflow with the broad context of everything you give it access to.
This could be as simple as data entry and administration or competent AI Doctors with the capacity to diagnose and set treatments for patients.
Significant changes are coming and healthcare will look completely different within a decade. While there are hurdles to jump, AI, and the agents they power, will unquestionably clear them soon.
Big Tech Is Working Hard
Industry insiders have already uncovered work OpenAI, Microsoft, Google and Salesforce have been doing to make agents a reality. Their operating systems have all the data and structure to make this happen. Meanwhile, Nividia has already announced AI agents in the form of nurses. An interesting first use case given the unique complexity and human touch required in the nursing profession.
This isn’t one of those things the market will slow down itself as developers ensure a careful deployment of the technology. We’re in the midst of a big tech arms race to lead the way with AI solutions and the trillions of dollars that comes with it in the coming decades. Market forces will accelerate the move for more effective and productive AI systems and agents are a pivotal step.
It is therefore unquestionable that companies will utilise their tech to deploy agents into healthcare to tackle the huge breadth of problems that make the system so complex and broken today.
Navigating The Problem Landscape
The AI arms race has already seen some problems and challenges emerge. AI chat interfaces are prone to hallucinations. Results and text can be reviewed and verified but agents acting unchecked on incorrect information can compound problems and make mistakes more serious.
Data Privacy and Security
Patient data is highly sensitive and protected by stringent regulations like HIPAA in the United States and GDPR in Europe. Ensuring AI systems comply with these regulations while maintaining data privacy and security is a significant challenge. You better believe regulatory bodies will be quick to jump on any major failure event. That is, of course, where the error comes from.
Data Quality and Integration
Healthcare data is often fragmented across various systems and may be incomplete, inconsistent, or unstructured. Integrating and cleaning this data for AI use is a complex and resource-intensive process.
Clinical Validation and Trust
AI systems need to be clinically validated to ensure they are safe and effective. This involves rigorous testing and clinical trials, which can be time-consuming and expensive. Gaining trust from healthcare professionals and patients in AI’s decisions is crucial and challenging.
Time is what will slow down the implementation of AI agents, not technology. This has largely been the case for medical innovation in the 21st century.
Despite these important considerations I am excited to see the emergence of AI agents in healthcare.
Opportunities Today
Administration
Let’s face it, healthcare is barely at the automation stage let alone primed for AI agents like other industries are.
Nevertheless, this will be the first and biggest influence of AI in the healthcare system. Administration costs account for 26% of healthcare spending, about $265 billion a year.
Patient data management, appointment scheduling, billing and claims processes and compliance will all quickly have humans removed from the process sooner than you might expect. 90% of these jobs will go and the 10% left will simply be managing and error-correcting agents.
Operational Efficiency
AI agents will be applied in hospital settings to optimise patient flow and resource allocation. They can predict patient admission rates and suggest the most efficient use of beds and staff. This is critical in enhancing hospital operational efficiency and ensuring that patients receive timely care.
Practitioner Augmentation
Today we have AI co-pilot systems being slowly integrated into Doctors' offices around the world. Autonomous AI agents will act in the exact same way as Doctors drawing directly from patient data. The Doctor will be able to refer to their suggestions and decisions ‘on the fly’ as they provide human-centric care that will long be expected by patients. The apprentice will be smart and more capable than the master, yet will remain the apprentice for some time.
Patient Engagement and Care
AI agents in the form of virtual health assistants can engage patients, provide personalized health recommendations, and remind them to take medications. These agents are becoming increasingly sophisticated, capable of understanding natural language and interacting with patients in a more human-like manner.
This coincides with developments in healthtech. New solutions for monitoring health biomarkers enable a highly data-driven care process. The perfect situation for AI agents to succeed.
Research and Development
In clinical trials, AI agents are used to process and analyze large datasets, identifying patterns that might not be immediately apparent to human researchers. They help in optimizing trial designs and predicting outcomes, thereby accelerating the development of new treatments.
Enhanced drug discovery and development, personalised medicine and disease predictive analysis are all downstream of a more effective, agent-led research landscape.
We question whether AI can have novel thoughts. Whether it is just a dumb parrot or can actually discover solutions that our best researchers can. The reality is AI agents, fed with succinct data, will try over and over, testing all possible avenues of success. They will start by augmenting researchers until they learn sufficiently to facilitate novel outputs.
Who’s Working On It?
These are not industry-specific but are the current leaders. If you’re looking to build with AI agents this is a good place to start.
Bigger Challenges
Agent Reliability
I am optimistic that healthcare will develop its own specialized LLMs away from the general models most organisations are utilising today. This speaks to the imperative of ensuring reliable outputs by training the models of high-value data. This is why any patient-facing solutions will come later as AI models will need more effective training and rigorous testing than backend and business operations.
We’re Missing Compute
For the foreseeable future, widespread AI agents in healthcare are unlikely given the limitation that will befall every other industry — compute.
Part of the AI race is a compute race ensuring data providers have the infrastructure capacity to operate billions of actions taken every second by autonomous AI agents completing tasks that would’ve taken humans hours in mere seconds.
Whether compute scale happens before safety measures are effectively assessed and implemented is something only time will tell. Anticipate one forcing the hand of the other though.
Animus AI
Animus AI is a specialist AI interface designed for health professionals and organisations. We have crafted a focused, intuitive and reliable AI platform that transforms the way you work.
