Generative AI in spinal cord injury research and care: Opportunities and challenges ahead

Generative artificial intelligence (AI) with its limitless potential is here to stay. Not since the advent of the digital age has the world faced a transformation of such magnitude. With the widely available user-friendly free and low-cost products of Open AI, the US AI research laboratory, the effects are being felt throughout multiple sectors, including scientific research and clinical care.

Anyone with internet access can sign up at OpenAI.com to access a free research version of ChatGPT, a generative AI large-language model that interacts with the user in a natural language format (1). Powered by the world’s fifth largest supercomputer (2), ChatGPT gathers information using artificial neural network technology (3). Trained using large databases, it has an enormous vocabulary and is capable of understanding language and context, retrieving knowledge, generating new content in the forms of text and code, carrying on a conversation, and personalizing communications. While trained primarily in English, this chatbot supports 95 languages.

While this tool became only recently available, ChatGPT gained traction immediately and gained followers at an unprecedented pace, attracting more than 100 million users in the two months following its public release in November 2022 (3). In March 2023, OpenAI released a more advanced multi-modal version, GPT-4, available through the paid service, ChatGPT Plus (4,5).

To explore these new tools, we posed general queries to GPT-4 about the potential for generative AI to influence spinal cord research and clinical care. Selected responses are excerpted below:

• Generative AI can be used to create virtual models of the human body, including the spinal cord, the musculoskeletal system, and the nervous system. These models can then be used to simulate spinal cord injury and test the effectiveness of different treatments in a more controlled and accurate environment.

• AI algorithms can help researchers optimize the design of spinal stimulation protocols by analyzing large amounts of data, simulating the effects of stimulation, developing personalized treatment plans, and predicting outcomes.

• Generative AI can be used to design new drugs that enhance the survival and integration of stem cells into the spinal cord. AI algorithms can analyze large amounts of data to identify potential drug candidates and simulate their effects

• By optimizing the design and control algorithms of robotic exoskeletons, analyzing large amounts of data, and simulating different types of injuries, generative AI can help researchers improve the effectiveness of these devices. By analyzing medical history and imaging data, AI algorithms can recommend the most effective exoskeleton design and treatment protocol for that individual's specific injury and needs.

When asked about its potential role in recruiting research participants, it responded:

• AI algorithms can analyze electronic medical records, patient data, and social media activity to identify potential candidates for spinal cord injury studies. Once potential candidates have been identified, AI algorithms can be used to create personalized messages and content tailored to individual patients. Generative AI can be used to develop predictive models that identify factors associated with patient enrollment and retention. Generative AI can be used to engage with patients, using chatbots, virtual assistants, or other AI-powered tools to provide patients with updates, answer questions, and address concerns.

ChatGPT offered these potential applications when queried about clinical care:

• In acute spinal cord injury care, generative AI can analyze vital signs, lab results, and other patient data to predict the likelihood of pressure sores, urinary tract infections, or other complications. This can help clinicians intervene early, which can improve patient outcomes and reduce healthcare costs,

• During rehabilitation, AI algorithms can analyze patient data, such as muscle strength, range of motion, and other physical indicators, to develop individualized treatment plans, monitor patient progress, and adjust treatment plans as needed.

• ChatGPT can assist medical professionals in accurately interpreting and communicating complex medical information. By providing personalized responses tailored to each patient's needs, ChatGPT can help bridge the gap between technical medical language and the language used by patients, caregivers, and policymakers.

• ChatGPT can personalize its responses over time based on user interactions. This feature has the potential to benefit patients with varying levels of understanding and comfort with medical terminology.

All ChatGPT’s responses stress the importance of carefully evaluating the capabilities and limitations of generative AI, and using the information generated in a responsible and transparent manner. OpenAI recommends that all content generated by ChatGPT be reviewed by a human, listing this as a best practice.

Although the news is filled with articles about its potential applications in medical research and care (6), in its current state of development, ChatGPT has significant limitations. Its training does not eliminate information flawed by human error, inaccuracies, misinformation, and biases. Information sources are deficient, as the current cutoff for training is September 2021. And while information is drawn from a large body of data, mistakes can occur, and ChatGPT does not cite sources for the information it provides. Most important are concerns about ethics, privacy, and security. Because current versions of ChatGPT are not secure and are unregulated, any information of a personal, protected, confidential, or sensitive nature should not be entered as part of a chat query.

Despite these drawbacks, generative AI is now a permanent part of the landscape, steadily increasing its capabilities at a rapid pace. As one expert involved in beta testing GPT-4 predicts, “this is going to change everything about how we do everything” (7). In medicine, we are already seeing major effects on cancer care and diagnostic imaging, for example (8). Such a transformative technology will certainly impact spinal cord injury research and care, and in ways we may not anticipate.

In the next issue, we will look at the impact of generative AI on medical publishing, as authors, editors, and publishers grapple with the challenges and opportunities presented by this new age of AI technology.