I have found a use for AI that no-one can dispute: diagnosing sick animals

AI has become a controversial topic in recent years. Not the AI behind NPCs in games, but the stuff that drives chatbots, creates generative images, and creates websites. Although there are still concerns, a team of Cambridge researchers has demonstrated the ability to detect heart murmurs in canines. I think that is pretty neat.

According to the Nvidia Developer Blog, a study was conducted using models that were "trained using PyTorch, Nvidia CUDA, and Nvidia GPUs of the GeForce 10 series, enabling efficient processing." The machine-learning algorithm used digital heartbeat data to 'listen' and diagnose heart murmurs or heart disease.

The algorithm was able to detect murmurs in any grade (that is, any intensity), at a rate 87.9%. In 57% of recordings, the algorithm matched exactly a cardiologist’s grade.

The study says that "the model is a promising instrument to enable accurate and low-cost screenings in primary care."

This is an excellent potential use-case of AI. Not only will it be cheaper for pet owners to have their pets scanned, but it could also be more efficient and accessible. You could scan your dog with a machine trained by someone who is not a vet. This could allow qualified vets to focus on other procedures.

There is one concern that should be addressed, and which the researchers were likely aware of before starting this study. A study from a few months back claimed that up to 20% local doctors in the UK may be using generative AI.

It's not a problem if AI is used to create calendars, or even drafts of letters. But it's important that you understand that AI cannot reason like humans. It can approximate reasoning through synthesising data, but a person is still needed to overcome the last hurdle of understanding.

It does not genuinely hear or see the patient's concerns or feelings and it cannot diagnose a problem. It can only compile symptoms and put them together with possible causes.

AI that might be used to diagnose a patient needs to scrape a lot of information. This is both for scientific reasons and to understand the language used by patients. LLMs can be influenced by 'hallucinations,' where false or bad information is treated like it is real, due to the amount of data they receive.

Even though they can be more accurate, it still takes a human to see the hallucination. A fake AI website fooled thousands of Halloween revelers into attending a fake event just last month.

AI can be used as a tool for diagnosis, but not to diagnose patients. In this case, the tool seems to be very accurate in detecting heart murmurs. This could be partly due to the fact that it was designed for a specific diagnosis. However, a human must verify and suggest treatment.

This could be one of the most useful uses for AI that I've seen so far, if used responsibly.