10 May 2025
What exactly is AI and how is it being used in veterinary medicine? This article seeks to explain how the sector can ensure this technology augments, rather than undermines, the standards of care and values that underpin the veterinary profession...
Image: Adobe Firefly / AI Generated
Artificial Intelligence (AI) is seeping into every aspect of how we live and work. From the algorithms that determine what content we consume to the routes our navigation apps tell us to take to work, technology is so embedded into our decision-making that we’re barely conscious of its impact in everyday life.
In veterinary medicine, a similar transformation is already underway. Whether through diagnostic analysis, automated note taking, or decision-making aids, AI is starting to reshape the art, science and working lives of vet teams and patient care. While this technology holds the promise of optimising workloads and patient outcomes, it also raises considerable concern on practical, moral, ethical and societal levels that warrant serious proactive consideration and discussion.
AI refers to computer technologies designed to perform tasks traditionally requiring human intelligence, such as recognising patterns, making decisions or generating creative content.
Currently, all AI systems are considered “narrow” or “weak” AI. They are incredibly powerful at doing specific tasks, but cannot apply broader reasoning. The next paradigm shift in AI will be the development of “general” or “strong” AI, which could mimic human intelligence across several domains, completing a wide range of tasks equal to, or better than, human capabilities. Beyond this, the sci-fi world of artificial “super” intelligence is when the machine surpasses human intelligence in all things and humankind is left pondering the imponderable.
Within narrow AI, tools can be broadly grouped into two categories – reactive and limited memory.
Reactive AI systems with limited capability to produce outputs in response to a set of data inputs lack memory-based functionality. Therefore, these machines cannot “learn.” Their power lies in the capacity to process vast amounts of data, such as IBM’s Deep Blue processing all the potential moves in a game of chess to determine the winning sequence1. More relevant uses include cytological or genetic analysis to speed diagnostics and research.
We are probably more familiar with the second type of AI tool, not necessarily because they are more common than reactive, but because of the hype following the release of ChatGPT – the fastest adopted application in history2. Limited memory machines build on the capabilities of reactive machines with the capacity to learn historical data to make decisions. These systems are trained on data that is then stored in their memory to form a reference model for solving future problems. For example, an image recognition AI is trained using thousands of labelled images, teaching it to recognise and name the objects it scans.
When a new image is scanned by the AI, it uses the training images as references to categorise the contents of the image presented to it. Based on its “learning experience”, it should subsequently label new images with increasing accuracy… provided the data inputs are high quality and representative. Conversely, models fed poor or biased data may become less accurate and/or increasingly biased
over time. For example, large language model outputs have been shown to change political leanings over time3 and with different language inputs4.
In veterinary medicine, we could imagine an imaging tool trained on a wide variety of dog breeds, then utilised most frequently in practice for a particular breed with a conformation more susceptible to a given pathology. This may bias or skew the results from other breeds as the model is learning on a sub-population.
Even within the field of narrow AI, the potential to enhance veterinary medicine and patient care is vast. Some of the most exciting fields of development are:
Enhanced data analysis. AI can extract and analyse vast quantities of data at speed, beyond human capacity and without getting bored or tired. AI can unlock additional levels of detail and pattern recognition, such as the field of radiomics – analysing digital radiographs down to the individual pixel.
Data integration. AI can integrate data from multiple sources, such as varied file types or practice management systems. This could be hugely beneficial to expanding evidence-based medicine research in an industry where data is often siloed and in varied formats.
Reduced admin burden. Automating time-consuming tasks like documentation or appointment booking. Admin burden has been shown to contribute to burnout in medical professionals5, so this could improve working lives.
Faster innovation. In research, AI has already sped up drug discovery by predicting how proteins fold or identifying useful molecules from genomic datasets. In-silico trials also provide additional levels of analysis to select possible therapeutic targets, reducing the need for expensive, time-consuming and ethically challenging in-vitro and in-vivo trials.
Decision-making aid. AI tools can provide lists of differentials, providing a second opinion, flagging a missed differential, or suggesting next steps.
Meeting unmet needs. Accessibility to veterinary care is highly variable across socioeconomic and geographical boundaries due to a multiplicity of factors. Veterinary teams can only support a fraction of the animals that could benefit from care. AI offers the opportunity to vastly increase access by overcoming many of the limiting factors.
For all the potential benefits, there are very real risks that need careful attention:
Bias in data. AI systems are only as good as the data they’re trained on. Veterinary data is highly variable across species, breeds, and conditions.
Transparency. We typically don’t know what training data was used, how it was weighted, or how outputs are generated. The “black box” element of AI makes integration into clinical decision making challenging.
AI hallucinations. Generative models can produce confident-sounding, but incorrect, responses. These “hallucinations” may look extremely convincing and may go unnoticed if not reviewed by an experienced, confident clinician.
Direct-to-owner tools. AI tools released directly to animal owners could mislead, misinform and negatively impact patient outcomes without veterinary oversight.
Job displacement fears. Legitimate anxiety exists that automation could replace roles. Certainly, many roles will need to change and adapt, but the goal should be to use AI to augment veterinary teams, enhancing skills use, job satisfaction and patient care.
Data security. Questions remain over who owns the data used in these systems, how it is stored, accessed, used and protected. These questions are particularly pertinent for cloud-based AI.
Deepfakes and deception. AI-generated audio or video of vets or animals could be used maliciously or fraudulently. Systems to detect these forgeries will be essential.
Carbon footprint. Training AI systems requires immense computing power and associated carbon emissions, raising sustainability concerns.
The AI genie is out of the bottle. There’s no going back, so how do we best go forward? How do we ensure the profession of tomorrow looks back with gratitude on the steps we took today to safeguard the integrity of practice and advocate for the animals under our care?
As with social media, it can be hard to predict the far-reaching consequences of decisions made now without looking back from the future with a retrospectoscope. With this in mind, the main lesson is the importance of not allowing tech companies (big or small) to charge ahead with no accountability. While most AI developers start with good intent to solve problems, we must be mindful this was also the original intent of developers of the vast social media platforms, and we are all now actively living with far-reaching unintended outcomes, particularly on the younger generation.
Rather than three wishes, what are the big questions we need to carefully consider now to ensure the consequences of this technology benefit teams and patient outcomes?
