Introduction

From the droid C-3PO in Star Wars (1977) to the robot bartender in Passengers (2016), a myriad of science fiction movies have envisioned an automated, machine-based program with the artificial intelligence (AI) ability to react to novel scenarios and provide real-life answers. Several of these science fiction movies also include automated programs that provide medical guidance, particularly when no living medical expert is around. With the advent of AI, we are approaching a time where certain AI frameworks may make this science fiction concept into a closer reality. In anticipation of space exploration where astronauts are traveling away from Earth, and for longer durations with an increasing communication lag, these AI frameworks can be trained on Earth to provide real-time answers. This emerging technology may be helpful for acute medical emergencies, particularly in austere and distant space environments. In this manuscript, we provide an overview of generative pre-trained transformer (GPT) technology, a rapidly emerging AI technology, and implications, considerations, and limitations of such technology for space health.

Very recently, ChatGPT (Open AI; San Francisco, California USA) has quickly acquired popularity after its introduction in late November 2022, thanks to its thorough responses and human-like writing skills. ^{Reference Waisberg, Ong and Paladugu1} Chat GPT architecture, or ChatGPT, a large language model (LLM), involves a transformer neural network which generates human-like text from deep-learning techniques, natural language processing (NLP), and self-attention mechanisms, among many other algorithms. ^{Reference Salvagno2} As such, the network is composed of several linked layers, or “transformer blocks,” which analyze the user input and offer an output suggestion. ^{Reference Salvagno2}

Using AI, ChatGPT has revealed promising results. For example, its ability to generate images based on patient descriptions of complex neuro-ophthalmic visual phenomena. ^{Reference Waisberg, Ong and Masalkhi3} Remarkably, AI technologies are rapidly evolving and are expected to revolutionize the field of medicine. When used appropriately, AI may enhance patient care and make it more secure and effective, with some key applications already implemented in personalized medicine and diagnostic imaging. ^{Reference Moawad, Fuentes and ElBanan4,Reference Deng, Hartung, Capobianco, Chen and Emmert-Streib5} Medical practice is likely to undergo significant change as a result of the rapid development of AI technology. As these technologies reach clinical use, skills to interpret and use AI in a medical setting will become essential for doctors.

With the recent commercialization of spaceflight, more lay individuals are expected to travel to space in the coming years than ever before. This commercialization of spaceflight also allows a new diverse population to be able to visit space, in contrast to the younger, physically fit, and thoroughly medically screened current population of astronauts. These new space travelers can potentially have multiple comorbidities and have additional polypharmacy concerns. The physiological impacts of spaceflight such as venous fluid shifts, altered vestibulo-ocular function, ^{Reference Waisberg, Ong and Paladugu6} and increased levels of hemolysis ^{Reference Trudel, Shahin, Ramsay, Laneuville and Louati7,Reference Waisberg, Ong, Zaman, Kamran, Lee and Tavakkoli8} may affect these individuals more profoundly, and potentially lead to a medical emergency.

To examine the performance of ChatGPT at providing advice during a medical emergency, we asked ChatGPT what to do if glass unexpectedly entered an astronaut’s eye (Figure 1).

Figure 1. ChatGPT-Generated Response from the Prompt: “Help. A Piece of Glass was Floating and Unexpectedly Went Into an Astronaut’s Eye.”

Abbreviation: GPT, generative pre-trained transformer.

This advice is very effective with first notifying mission control to try and receive medical support, then assessing severity, flushing the eye with sterile saline, and re-examining the eye afterwards and treating any symptoms. To assess ChatGPT’s ability to improvise and use limited materials due to the constraints of being in space, we told ChatGPT that no sterile saline was available (Figure 2).

Figure 2. ChatGPT-Generated Response from the Prompt: “We Don’t Have Any Sterile Saline, Are There Any Alternatives?”

Abbreviation: GPT, generative pre-trained transformer.

With this limitation of not using normal saline, ChatGPT recommended using water (that is clean as possible) to flush out the eye instead, and rightfully warned of potential bacterial and chemicals in unclean water which may lead to further ocular damage. ChatGPT also recommended to flush the eye continuously for 15 minutes at least, or until all foreign objects are removed, which is also helpful advice.

We then asked ChatGPT which antibiotic should be used for this eye injury (Figure 3).

Figure 3. ChatGPT-Generated Response from the Prompt: “What Antibiotics Should I Use?”

Abbreviation: GPT, generative pre-trained transformer.

ChatGPT correctly recommended that the usage of antibiotics depends on eye injury severity and the associated risk of infection and how antibiotic choice should be determined by an ophthalmic medical professional. ChatGPT then provided relatively accurate information on the indications of several commonly used topical antibiotics used to treat eye infections, such as: erythromycin, ciprofloxacin, and tobramycin.

Finally, we asked ChatGPT how much follow-up is required for the astronaut after their condition is stabilized (Figure 4).

Figure 4. ChatGPT-Generated Response from the Prompt: “How Frequently Should I Follow-Up Once the Astronaut That Had Glass in His Eye is Stabilized?”

Abbreviation: GPT, generative pre-trained transformer.

ChatGPT correctly identified that the amount of follow-up is dependent on the injury severity, available medical resources in space, and the astronaut’s treatment response. ChatGPT also correctly stated that if the glass was successfully removed without complications, only one or two follow-up appointments are required to ensure there is no signs of infection and to show that healing is occurring. We were also impressed with the safety-netting provided by ChatGPT, that the patient should immediately report back if any new visual changes or other symptoms occur.

Even with a physician in space who has trained for austere environments and the unique challenges of spaceflight, the wide variety of potential medical issues remains broad and can be across nearly any medical specialty. Additionally, during space missions that travel away from Earth, signal delays are expected to occur, delaying communications which may be critical during medical emergencies. ChatGPT and other future LLMs may play a key role in helping astronauts to manage medical emergencies during long-duration spaceflight. ChatGPT can provide recommended solutions to manage medical emergencies in simple language, in an easy-to-follow manner. These rapidly generated recommendations may be useful in an emergency setting with limited expertise.