The explosion of generative artificial intelligence (Gen AI) over the past 18 months has brought forth incredible new possibilities for work, personal productivity, and more. Even today, the flood of relevant technical advances continues unabated. Gen AI refers broadly to both the machine learning (ML) models known as large language models (LLMs) and to diffusion models, which are deep learning models that can generate high-quality text, images, and other content based on training data and that can be customized to user prompts.
Gen AI provides broad advantages today, but future advances—including multi-modality, efficiency improvements, and domain-specific fine-tuning—will significantly improve its performance. Through its use of both current and future tools, the Foundation seeks to improve efficiency and effectiveness to better serve the patient population and speed the clinical adoption of focused ultrasound. Efficiency gains may be realized through reducing the time required to create diverse content for both internal and external use, as well as through speeding information retrieval. Gen AI could even eventually quickly and accurately answer questions for clinicians, patients, and others without the need for active intervention by Foundation staff. It could also lead to gains in effectiveness through brainstorming and idea generation, serving as a capable “sounding board” for Foundation employees planning complex tasks and deliverables.
Practical Impacts
The Focused Ultrasound Foundation has embraced AI transformation for the reasons outlined above. The following are real-world ways in which Gen AI is improving efficiency and effectiveness at the Foundation.
Improving Efficiency. The Foundation is using Gen AI to streamline workflows and allow staff to focus on high-value tasks. Several months ago, I set out to document specific examples of a trend occurring in the healthcare ecosystem. Originally estimating that this effort might take upwards of eight hours, I turned to an internet-connected Gen AI model, receiving an initial answer in less than five minutes. Subsequent results validation—discarding one hallucinatory, nonexistent example provided by the agent—yielded reliable and useful information in approximately 30 minutes. Imagine a future where, for task after task, we can accomplish research needs, compose papers, and complete administrative responsibilities in a fraction of the expected time.
Improving Effectiveness. My initial agenda plans for a recent special event were constrained to a few topical threads—those with which I had familiarity. I asked a Gen AI agent, in effect, “I’m planning an event for these purposes, and here are the topics we will discuss. What other subject matter might be relevant?” The agent responded with a dozen possible topics for addition. After discarding those that were not pertinent, my topic list had doubled in size. We resumed planning based on the expanded catalog of possible subject matter, and this method provided a richer participant experience.
Steps to Adoption
How is the Foundation systemically leveraging Gen AI tools to help patients and make the best use of limited resources? Numerous elements have been put into place, beginning with establishing a Gen AI strategy from which all other activities flow.
In our case, strategy elements include training Foundation staff in Gen AI tool use and providing resources, such as software subscriptions. We have established best practices for responsible governance and safe tool usage, and we are systemically assessing opportunities for Gen AI use within team workflows while encouraging experimentation beyond these core areas. We’ve begun training custom agents to best serve the needs of staff and, eventually, our website visitors. And of course, we are defining and capturing progress metrics for our Gen AI journey to assess our progress over time.
Current State and Next Steps
The Foundation is still early in its Gen AI–driven evolution, as we all are, but the team is committed to responsible, safe, and effective tool use. This transformation represents a profound cultural shift, which requires full leadership support. Given that we have such support, we are coaching employees across the organization on possible ways to use Gen AI in their respective workflows. Individualized uses include the following:
- Our scientific team is finding value in leveraging LLMs to extract science-based answers from composite sets of scientific research papers. The process of combing through papers and drawing conclusions on their collective findings has become easier with tooling to assist in this process.
- Our communications team uses Gen AI to ensure a robust video presence on social media. Video repurposing software allows us to take our existing video content and cut short vertical videos optimized for social media use. Gen AI–generated content can be used as a nearly-finished product or as a storyboard to guide us in repurposing content through alternative editing software.
- Our communications team further uses Gen AI tools for audio production, including voice cloning, music and sound effect creation, and audio quality enhancement (in podcasts, for example). Similarly, Gen AI tools can generate unique imagery, produce b-roll, and edit or enhance visual content. Finally, workflow automation can be optimized with Gen AI applications, saving time and streamlining the content creation process.
- Our technical team uses Gen AI in its day-to-day coding tasks. When our team describes what we would like to do in code, Gen AI automatically creates functions to accomplish those objectives. Gen AI has similarly been useful for analyzing old, legacy code to quickly understand the code’s purpose—allowing us to spend our time moving the organization forward rather than assessing past intentions. Finally, Gen AI is being used to write simple software unit tests, saving staff time and money through intelligent automation.
- Our administrative team harnesses Gen AI to streamline data management. This technology enables the team to better manage spreadsheets and other varying data sources. Gen AI has been instrumental in comparing and cross-referencing text listed on our website with master data lists—a process that would otherwise be tedious and time-consuming. Through intelligent automation, tasks that once required an hour or more can now be accomplished in mere seconds, allowing the team to dedicate more time to strategic initiatives.
Broadly, Foundation employees are using Gen AI agents to assist with targeted research, aid in brainstorming, and create content. While Gen AI tools will not write entire papers or create other finalized products, we repeatedly find that these tools get our team half-way there, speeding our time to completion and allowing focus on higher-value tasks. Importantly, at an organizational level, our net promoter score of Gen AI tooling is high, reflecting staff confidence in our ability to accomplish more today than we could yesterday.
We are also developing custom agents, essentially modifying off-the-shelf Gen AI tools with focused ultrasound–specific data to better meet our needs. The goals of such efforts are to allow our team to more effectively find the answers they need to serve the patient population. We’ll continue to pursue such custom agents and employ third-party tools if they help us speed time to value. We don’t want to be too far ahead of ourselves, however. In each case, we consider whether the capabilities we seek may be commodities in the coming months. Accordingly, we make carefully weighted decisions about where to spend our time in fleshing out Gen AI capabilities. On top of that, this amazing technical transformation is highlighting data inconsistencies between trusted sources. For this reason, we’re working—where justified—to cleanse and harmonize our data sources to provide the most reliable and accurate answers to questions.
Conclusions
As a young engineer just out of graduate school in early 1993, I downloaded new software from the University of Illinois Urbana-Champaign—the NCSA Mosaic web browser. I’d been “online” during years of graduate study, but internet user interfaces were clunky command line matters mostly used by technical students. Now, staring into an early web browser, I glimpsed the future—egalitarian and democratic—offering new ways to work, learn, and play.
Most of us at that time could never have imagined the business model of Uber, the societal impact of Amazon, or the roughly two million apps on the Apple App Store. However, the changes that the browser brought—richly diverse content, and eventually services, from across the globe—underscored the fact that the fundamental exchange of information, and commerce itself, were being disrupted.
More than thirty years later, we are experiencing another “Mosaic moment.” We’ve spent the intervening decades roaming the magnificent, chaotic hallways of the internet, searching for data. Moving forward, we will increasingly talk with our data, and benefit from the intelligent assistants at our side. These assistants will retrieve the annotated information we need, spark our creativity with new ideas, and increasingly act on our behalf through new classes of AI agents. The underlying technology behind this—the AI/ML architecture known as transformers, and their inevitable antecedents—will tackle new, non-textual challenges and accurately predict sequences in numerous domains. These multi-modal models will change the ways in which we, as clinicians, researchers, engineers, and scientists, advance our missions.
At the Foundation, we embrace this transformation and will use it to advance patient interests. Together, we look forward to harnessing the benefits of such technological advances to effectively leverage resources, hasten breakthroughs, and deliver better outcomes to patients.
Rick Hamilton is chief technology officer and a managing director at the Foundation. He is recognized as one of the most prolific inventors in world history, with over 1,060 issued US patents.