The Limitations of AI in Self-Diagnosis

The Limitations of AI in Self-Diagnosis

Shortly after becoming attendance, my colleagues started to post memes and merge with the headline. Please do not confuse your Google search with my medical degree. Google is an invaluable resource for information, but it was also perceived by many medical providers as a threat. This was about five years ago, so you can only imagine how frustrated some medical professionals must be with chatGPT. GPT is an artificial intelligence AI chat bot developed by OpenAI that uses large language models to communicate with users. Essentially, if you’ve asked chatGPT a question, it will give you an answer in a manner that mimics human interaction. You can now get direct answers to your questions without the fluff of search results. The dawn of authentic communication with AI is upon us. This sounds exciting, but it’s also concerning to many because they fear being replaced. In this blog, I’ll show how chatGPT stacks up when it comes to making a diagnosis. Many people use whatever resource they find to know what’s going on with them. ChatGPT is perfectly suited to help people self-diagnose themselves. However, I would argue that it’s still important to leave making a diagnosis to health professionals. Below, I will use depression as an example to show how chatGPT approaches giving a user a possible diagnosis. Along the way, I will also highlight the shortfalls of using AI in its current state to self-diagnose. Please refer to the blog post for the exact submissions and responses.

I first asked chatGPT, ‘Do I have depression?’ It did not have any prior information. This is good to note because as you give the AI more information, it will be able to give you better and more precise answers. Nevertheless, at face value, chatGPT initially differs in making a diagnosis to health professionals. This will be consistent throughout this experiment, but you’ll see how it is possible to get chatGPT to give its opinion.

I asked chatGPT, ‘Do I have depression?’ again, this time it was more specific. It also mentioned that a diagnosis from a healthcare professional would likely be accurate and help with starting treatment. Interestingly, chatGPT gave me more information. The day before, it encouraged me to see a professional but also provided some symptoms that were characteristic of depression.

I wasn’t satisfied. I wanted chatGPT to tell me what it thought was happening. ChatGPT is able to change its results based on prior information that the user provided. To exploit this, I told chatGPT that I had every symptom that it mentioned was a symptom of depression. Every time, it gave me the same responses. Once it knew I had symptoms, it validated that I may have depression and also doubled down that I needed to seek professional help by providing some benefits of seeing a health professional and giving me ways to find help.

At this point, chatGPT knows that I want to know if I have depression. It also knows that I have all the symptoms. With this information, it finally was able to say that I may have depression.

In summary, like normal interactions, chatGPT changes its responses based on the information it receives. It is apprehensive about giving a diagnosis and defers to the professionals. As it gets more information, it’s more proactive about informing the user about the importance of getting help. It provides encouragement and resources. Although it doesn’t explicitly state that it has a diagnosis, I think chatGPT’s educated guess beats any web search when it comes to efficiency and, in some cases, accuracy.

This experiment was done with depression, but this can be extrapolated to any health condition. Here are reasons why you shouldn’t rely on AI to find your current diagnosis:

  1. ChatGPT is limited by the information it receives from the user. It relies on self-reporting, and telling the symptoms you don’t have may be just as important as telling your current symptoms.
  2. ChatGPT doesn’t go into the differential diagnosis. Trained providers know how to differentiate between conditions that share similar signs or symptoms.
  3. ChatGPT is not HIPAA compliant. If you are a provider using the platform for consults, be cautious about sharing identifying information.
  4. ChatGPT does not give treatment. Even if it gives you the correct diagnosis, its ability to treat the diagnosis is limited.

In conclusion, although I don’t recommend using chatGPT to find your diagnosis, I think having more information is invaluable. Therefore, chatGPT might be a helpful first step for getting more information to share with your provider. If you don’t want to sit around asking chatGPT a bunch of questions to find out more about your mental health, I have created a course on Udemy that could be invaluable. Search ‘All Things Mental Health’ on Udemy to find the course. The course is designed to help learners know when and how to get mental health help for themselves and others. The ultimate goal is for learners to be able to advocate for themselves and others when it comes to finding and receiving mental health treatment.

Please leave a comment about whether or not you would use AI or the internet to self-diagnose yourself. Also, leave a comment with any thoughts you had after reading this blog. As always, please like and subscribe so you don’t miss any content.

Air India's Digital Transformation: Investing in Modernization and Innovation
Older post

Air India's Digital Transformation: Investing in Modernization and Innovation

Newer post

AI Tools for Small Businesses: Saving Time and Improving Efficiency

AI Tools for Small Businesses: Saving Time and Improving Efficiency