Shortening the AI Learning Curve

How one online education platform aims to help veterinarians make sense of how AI will revolutionize the industry in the years to come.

Petra Harms, DVM, is passionate about responsible AI integration into the veterinary industry – the keyword being responsible.

“Responsible AI means that the technology needs to be designed and implemented in a way that minimizes harm and works to benefit the people that are affected by the technology,” she said. “In my opinion, this means that responsible AI in healthcare needs to maximize the safety and well-being of not only the patients and clients, but also of the doctors who use it.”

This summer, VetMaite, the company Dr. Harms founded and runs as its CEO, released a new online library of video lessons on AI for the veterinary community.

“AI implementation is going to happen whether we as vets learn about it or not,” she said. “There are big financial and hype drivers pushing the technology into the veterinary space. Unfortunately, the risks of using the technology will rest on the backs of the veterinarians practicing medicine. The way that AI technology gets implemented will be influenced by who brings a seat to the decision-making table – corporate groups, industry and tech, data mining entities, researchers, insurance providers etc. In order to serve the needs of practicing vets instead of further complicating their lives, we have to make sure that practicing veterinarians pull a seat up to the table. As consumers of the technology, vets have real potential power to shape how the technology is designed and applied, but they need to be enabled to realize that potential.”

In an interview with Veterinary Advantage, Dr. Harms discussed AI’s wide-ranging potential, and what safeguards are needed to safely implement it into veterinary practices.

Why do you believe there was a need to create an online education platform for the veterinary space on AI?

Dr. Harms: There is a big gap between practicing veterinarians and AI technology – there is a lack of understanding, some sense of skepticism, and almost a helplessness that comes with not knowing how to engage with it in a critical way. The majority of practicing veterinarians don’t have enough basic literacy in the topic to evaluate its risks and benefits from a confident position. It creates an uneven playing field with advancing technology and strong financial drivers on one side, and an under-informed population of veterinary end users on the other.

An online education platform is a way to democratize basic AI literacy for veterinarians, by making it affordable and accessible across the board. I can’t speak for the non-English-speaking world, but to my knowledge it is the first-of-its-kind collection of AI lessons online in the English geared toward veterinarians and vet team members. By building up their basic literacy, veterinarians can engage with the technology from a more empowered position, ask critical questions of providers, demand appropriate standards and performance benchmarks from the industry and push the industry to make their user experience a priority.

How have you seen AI evolve in recent years?

Dr. Harms: Right now, it’s a bit of a Wild West of available technologies and opportunities, with interesting new start-ups emerging frequently. There has been an increase in the number of accessible products coming to market for veterinarians. ChatGPT has really captured the attention of the general veterinary audience. When ordinary people started using it and finding out what it can do, it made AI technology more concrete to them – something that is happening today, not 100 years from now.

EMR scribe programs in various forms have been the most common product that I’ve seen brought to the veterinary market in the past six months, but I do see momentum toward implementing AI technology by diagnostic service providers as well. From what I’ve seen on the human healthcare side, this is just the tip of the iceberg. With increased product visibility come more questions about what is appropriate and safe, and end users are looking to their regulatory bodies and leadership teams for direction.

It’s interesting to watch how the veterinary and human healthcare industries are working to keep up with the pace of change. This is the most noticeable shift that I’ve seen in the past six months – larger organizations are starting to look for ways to act. The speed at which this technology has evolved is dizzying. I imagine it’s pretty tough to be someone in the veterinary regulatory field right now, wading through (sometimes weekly!) reports on new AI technology advances, with no certainty of where this will all take us, and little scientific data to base policy recommendations on. Scientific studies about the effects of various AI tools on vet space users just can’t be funded and published quickly enough to make a meaningful impact on policy decisions.

Since the beginning of 2024, I’ve seen individuals and interest groups really take a step forward to try lend some clarity to the chaos. Cornell University this past spring hosted SAVY, the world’s first symposium on AI in veterinary medicine. Its audience arrived from all over the world. SAVY featured not only presentations, but a fascinating pair of workshops canvassing both industry leaderships and general attendees to identify the current challenges facing AI in veterinary medicine.

The Veterinary Innovation Council has made addressing AI technology a current focus, and held a round table discussion on the topic during the AVMA conference this year. The non-profit Association for Veterinary Informatics has stepped in with education offerings as well.

In what ways do you see its potential in improving vet care?

Dr. Harms: A lot of the early tech has been aimed at improving time-draining pain points: writing medical records, reviewing and summarizing large medical files, writing discharge notes and correspondence, etc.

Going forward, I can see applying the tech to scheduling (patients and staff), completing referrals, prioritizing workflows for diagnostic specialists, and help with interpreting test results. Delegating these tasks to an AI model will make it possible for a vet or team member to do what they really want to do – focus on the animals and clients in front of them. This is all the low-hanging fruit, which will hopefully improve our work and the quality of personal care we can give.

Just as exciting is what AI can do to expand our abilities in the future. Machine learning is amazing at extracting subtle information from large amounts of data. I’m so excited about how that might help our food animal industries in terms of improving husbandry, quality of life for the animals, disease surveillance and yields, while minimizing environmental impacts. Disease and population surveillance of wildlife will stand to benefit, as will most other areas within the One Health framework.

Machine learning is already being used in drug discovery, disease modelling and creating digital twins. The Mayo clinic has trialed many applications, including applying machine learning to analyzing ECG signals to detect HCM and reduced fractional shortening in humans with heart disease. The study of radiomics is based on the idea that computers can register many more subtle differences in shading and color than humans, and so can extract more information from medical images than the human eye can. How cool is that? Think about the applications to pathology, radiology and advanced diagnostic imaging. Even something as simple as a photograph of a skin lesion might be able to tell us vastly more than our naked eye currently can. It will be exciting to see how AI technology will be able to take the tools that we use, right now, and extract a whole new level of interpretation from them.

What are the roadblocks to adoption of AI tools?

Dr. Harms: The first is lack of trust and uncertainty. Veterinary teams don’t want to get burned by jumping into unknown territory, and few want to be the first ones in line to test things out. It takes faith (or desperation in search of a better way) for a busy practitioner to take on a brand-new technology like an electronic scribe and try to integrate it into their already developed workflow. I think as the scribes become more common, and as they become more seamlessly integrated into the main EMR records programs, this will likely become less of an issue.

The second is lack of industry guidance, for the reasons that we’ve already touched on. We need to establish frameworks for responsible AI adoption within the veterinary industry, including guidelines for use and performance benchmarks that let users assess the safety and reliability of tools and products. This is changing as well, as leaders in the industry start to get together and create a framework for a way forward.

The third is lack of existing infrastructure, both physical (computing) and database-related (data silos). Some clinics might just not have the computer setups to run some of these more complex AI-based programs. As for data silos, a major problem that has come up behind the scenes is that databases on animal health are all stored in different systems and under different labelling rules depending on the program that was used to gather the data. It’s harder than you’d think to extract all that data, combine it all under one common system and then use it for machine learning. There are pockets of experts working on that problem as well, and I’m optimistic that it will be a problem of the past within the next decade.

The fourth is the technology itself. It’s very exciting, but far from perfect. Developers continue to struggle with hallucinations, bias and lack of interpretability. Until these can be substantially corrected, we have to continue to carefully vet information that we get from AI tools for these flaws.

The fifth is ethical implications. As an industry, we are going to have to struggle with the impact that AI technology will have on changing job descriptions. We’ll have to wrestle with the impact of AI use on the environment. We’ll have to diligently manage business transparency on where AI tools are being applied within organizations. These are all topics that need to be addressed if we want to adopt AI responsibly.

How should vet practices approach the adoption and integration of AI into their businesses? What are steps or best practices to follow?

Dr. Harms: Oh my goodness, there is so much to cover here! A Best Practices guideline is beyond the scope of this Q&A – it would take several more pages to answer in-depth, and will depend on the type of AI technology that is being integrated.

In a more general sense, a vet practice wanting to integrate AI tools into its business should tag at least one team member (or themselves!) to develop basic AI literacy. This doesn’t take long, and will allow them to act as a bridge between the technology provider and the vet clinic purchaser. AI literacy will allow the team member to ask critical questions of the AI tool provider and ensure the provider and clinic have done their due diligence when it comes to assessing quality control, data privacy, reporting errors, transparency, evaluating need for client informed consent etc. The risks to your staff and patients are often related to the function of the AI tool.

Technology that automatically filters emails will not present as many risks as technology that interprets patient ECGs for example. Once the risks to the clinic have been assessed, you can establish responsible use protocols and communicate them with the veterinary team. It might sound complex, but it doesn’t have to be – this informed, curious and prudent approach isn’t much different from what you might do when purchasing a new piece of X-ray equipment for the clinic.

When choosing an AI tool, I would carefully assess how the technology can integrate into the existing clinic flow and equipment. Many AI scribes for example will allow a trial period – take advantage of this to find out whether the scribe has a smooth and easy user interface, whether it actually improves your workflow, whether you can deploy it right away or whether you’ll have to make significant updates to your equipment in the hospital. Ask the AI tool provider about how their technology integrates into your current medical records system. Ask your current medical records system (or other equipment) provider whether they have suggestions for AI tools that have integrated well with their programs in the past. Check to see if there are major changes coming up with your medical records system that might affect the usability of your AI tool.

When you’ve committed to an AI tool, make full use of the error reporting system they provide. These technologies are still in their early years (or months) and their developers will benefit from knowing where the glitches are. On a regular basis (consider once a year or when the contract is renewed), touch base with the tool developer to see if the model needs to be retrained due to errors or drift.

Why did you want to create a hub between end users and the suppliers and developers?

Dr. Harms: VetMaite is first an education resource, and second a connection resource. Short of attending conferences or hanging out on LinkedIn, conversations about AI in veterinary medicine are hard to find. When I established VetMaite, I wanted to create a place where practicing veterinarians could speak directly with experts in the field, ask questions and discuss their concerns. Many machine learning and AI experts are also suppliers and developers of AI tech products. If I were to cut them out of the conversation, it would limit the abilities of veterinarians to get answers to their questions. When the pool of people who can answer questions in this new industry is already so small, it makes no sense to make it even smaller. Suppliers and developers also benefit, by getting feedback from potential consumers about what their concerns are about AI technology. If developers can allay these concerns by improving their products or providing education, then everyone ultimately benefits.

VetMaite’s biggest challenge is maintaining an unbiased education platform while also providing for a connection venue. If the platform strays into bias territory, then trust in the learning resource material is lost. Because of this, the message board connection hub is a strict “no sales” zone for discussion, while there is a separate section for vendors and developers to highlight their products or services.

Enter the Unknown

Since veterinary medicine is moving into entirely new territory, it’s impossible to predict how this technology is going to affect practitioners and clients. Use of AI technology opens up a whole new rats’ nest of legal liability questions if something goes wrong, Dr. Harms said. “Most members of the public have seen ‘The Terminator’ and many are wary of AI tech,” she said. Mistakes made while using AI technology will potentially attract a lot of attention. Social media can turn a single incident into a very large, very public problem, very quickly. It might only take one or two spectacular misfires to drastically slow, or even stop, the adoption of AI into veterinary medicine as trust is lost.” This is why the industry needs to be very proactive about looking out for potential complications and putting safeguards into place. This will protect technology users, patients and clients, while maintaining an environment where new ideas can flourish. “It’s a tricky balance to strike.”