Seventeen current postgraduate nursing students aged between 23 and 45 years participated in the interviews. The sample comprised 13 females and 4 males. Four participants were current PhD students, six had academic degrees as their degree category, and four participants were part-time postgraduate students. The duration of participants’ AI exposure ranged from a minimum of two months to a maximum of two years, with natural language processing being the most prevalent AI technology employed. The research interests of the seventeen participants were diverse, encompassing infectious disease care, neonatal care, smart healthcare, emergency medicine, and geriatric care, among others. The detailed demographic and professional characteristics of the participants are shown in Table 1. Three themes were identified: cognitive perceptions of AI, attitudes towards AI, and practical dilemmas in using AI(Supplementary Table 2).
Theme 1: Cognitive perceptions of AI
The theme includes two subthemes: understanding of AI and awareness of AI products.
Subtheme 1.1: Understanding of AI
A limited understanding
In regard to AI technology, respondents have a more one-sided perception and tend to understand only generative AI. The majority of the respondents said that they did not know much about AI and found it too difficult to answer the “understanding of AI” question. This finding is indicative of the limited and incomplete understanding of AI among postgraduate nursing students.
“I do not truly know much about his [the AI’s] information sources and collection channels; I started using the AI when it was recommended to me by someone else who said it was convenient. “(N1).
“Generative AI is what I use a lot; I do not know much about the others. “(N10).
“With the question you just asked, suddenly I realised that I truly know quite a bit less about this AI, and then maybe it is more one-sided as well. “(N16).
B layman’s understanding of AI
AI is frequently perceived by some respondents as something of “advanced” nature, whereas others contend that any entity possessing a degree of “intelligence” may be classified as AI. The respondents indicated that AI represents a computer simulation of human intelligence designed to assist in addressing diverse issues encountered in life, work, and academic pursuits. This technology facilitates convenience for individuals, enhances productivity, and makes people “lazy”.
“I think AI is more about using a big data platform, I guess, to be able to solve many of our real-life problems…. It can facilitate us to retrieve something and then also come up with some more complete answers. “(N10).
“I will put it in a more generic way like I understood lazy before, lazy equates to sort of like that productivity thing where you’re lazy and you’re letting the AI take care of those things that you do not want to do. “(N4).
C AI applications are promising
The respondents asserted that human learning capacity is inherently limited, whereas AI possesses sophisticated algorithmic programs that demonstrate exceptional prowess in various domains. Second, the disciplinary scope of nursing encompasses a multitude of fields, such as medicine, psychology, law, education, research, and interdisciplinary collaboration, indicating that AI holds the potential for a diverse range of applications within nursing, which are likely to integrate well with AI.
“I think the future of nursing has to go to AI; it is a very important direction; it is going to be very widely used…. Anyway, I can put it this way; I think it can be applied to every aspect of nursing. “(N13).
“In the field of clinical nursing, nursing teaching and research, all of these areas should be covered in some way because of the wide range of applications of AI. “(N17).
D functional understanding of AI
The respondents identified the following functions of AI: problem-solving, information gathering and processing and analysing data, language processing, image and video processing, and the ability to perform some clinical work.
“He’s able to catch many things based on this big model, and there may be times when we want to research content; then, I go on this AI and ask, and he may give me many things. “(N16).
“AI can help us with a lot of data; for example, now we will collect a variety of some data in the clinic. Thus, the data are rather cumbersome, and it is possible to rely on AI to organise and summarise the data. “(N1).
“As far as I understand now, AI can already be applied to certain surgeries…nursing, I think he can replace us in the ward to deliver or issue medicine. Patients can also be picked up at the door of the ward for admission and so on. “(N6).
Subtheme 1.2: Awareness of AI products
A ChatGPT and other large language models
The respondents perceived the big language models are natural language processing models trained on large amounts of textual data that are capable of understanding and generating human language to capture linguistic information for problem solving. Respondents indicated that the AI technology that they currently understand and use more is big language modelling, such as products such as ChatGPT, Wenxin Yiyan, Doubao, Kimi, and Wisdom Spectrum Clear Speech.
“There are some AI things that are being utilised in daily life, such as ChatGPT, and then there are some AI products in China, such as Doubao and so on. “(N1).
“I mainly use Wenxin Yiyan and Kimi now. “(N11).
B AI intelligent robots
Respondents perceived the AI intelligent robots integrate all the technologies of AI to sense the environment, learn and process information and perform tasks. The respondents indicated that they had some familiarity with or exposure to AI robotics, albeit not to a deep extent. The primary application areas they were aware of included surgical robots, health education robots, and medication delivery robots.
“I know that surgical robots are now more widely used in the clinic, replacing us nurses to do some of the more delicate manoeuvres. “(N14).
“Let us say a patient is in the hospital, and then he will put a robot next to him, and the robot will do a daily health promotion for that patient. “(N4).
“What about that logistics cart or something like that, ah, that is also AI,… that is that smart cart for nursing, not having to have a caregiver to get the medication, I think it is just going to lighten their workload. “(N9).
C knowledge map
Respondents perceived the knowledge graph as a dynamic three-dimensional structured knowledge base that has been widely used in nursing, especially in the fields of nursing education and geriatric care. The respondents indicated that knowledge mapping can develop a personalised knowledge base for different patients and students, which can be tailored to provide advice and assistance.
“We’re now working on a knowledge map…… it is a dynamic, even three-dimensional, you can click on it to go and see more in it…… for this kind of dementia and disability of elderly individuals, made a knowledge map to give him targeted guidance and so on. “(N16).
D predictive model
Respondents perceived predictive model as a tool that can forecast future events on the basis of historical data. The respondents indicated that predictive models are widely utilised in clinical settings, primarily for forecasting the onset and progression of diseases, thereby alleviating patient suffering.
“An application inside a clinical decision-making knowledge system, where predictive models are used to predict the risk of a disease and then presented in a visual form. “(N5).
“The predictive model is very much applied, where we predict the risk of occurrence of the outcome of a nursing event. “(N12).
E virtual reality (VR) and augmented reality (AR)
The respondents believe that VR and AR are extensively applied in nursing education and clinical practice. These technologies enable students or patients to engage in immersive experiences, which are beneficial for both academic learning and health education.
“And then there’s the virtual simulation that is included in our classes. “(N14).
“Now there will be this like AI digital person ah. This digital person, he just grabs an avatar of me, and he just renders the whole of this in its form using my voice. “(N16).
“What I have been exposed to is giving health education to patients on the ward…. It is more with the help of a VR scenario, like simulating an environment, and then getting the family to go and participate in the care of the newborn. “(N10).
Theme 2: Attitudes towards AI
The theme includes two subthemes: positive attitudes of support and encouragement and neutral attitudes of concern and scepticism.
Subtheme 2.1: Positive attitudes toward support and encouragement
A machine learning is universal and precise
The respondents believe that predictive models constructed through AI algorithms offer greater generalizability than conventional predictive models do. Unlike traditional search engines, AI is capable of providing precise answers and proposing effective solutions.
“The previous predictive models are through the form of collecting data, then it may lead to incomplete, and then the bias is more serious, then through the artificial deep machine learning then it is more pervasive predictive models. “(N3).
“The big model can give you a detailed answer to this point you specifically asked him, and you will get the answer you want quickly without all the effort and time. “(N4).
B AI applications are initially troublesome but subsequently convenient
Some respondents mentioned that the development of AI models can be complex and challenging; however, once AI models are established, they significantly facilitate subsequent applications. Despite initial difficulties in the implementation of AI, the overall attitude towards its adoption remains positive.
“There is some trouble at first, but if every student he uses it, the first trouble goes to the second, and the third he does not have any trouble with it, and it becomes more and more complete and perfect. “(N16).
C AI is emotionally responsive and more emotionally stable
Respondents believe that AI has the potential to elicit emotional responses, either currently or in the future, and that its emotional stability surpasses that of humans. AI is perceived to be more patient with patients and unaffected by negative emotions.
“The kind of emotional companionship he can actually give is that his language is inherently infused with a lot of human language. “(N17).
“And people have seven emotions and six desires, and when they come into contact with patients, they definitely have emotions ah…… However, robots are different, if you set him up as a program, then no matter how this patient gets, he will not get angry. “(N15).
D expanding the direction of nursing research
Nursing, as an interdisciplinary field, encompasses knowledge from various domains. The respondents believe that integrating AI with nursing could pave the way for numerous emerging research directions in the future, facilitate a deeper understanding of interdisciplinary advancements, and reduce information disparities.
“Such a research paradigm and research direction of our original nursing is relatively narrow ha; by making a kind of cross-combination with AI, we are also able to expand the scope of our nursing bar…. This will produce some new directions. “(N13).
E increased efficiency
The respondents highlighted that AI offers numerous conveniences. In clinical settings, AI-driven patient education can enhance patient compliance; the use of AI for medication dispensing can reduce nursing errors; leveraging AI to streamline cumbersome documentation tasks saves time and improves work efficiency; and AI can also assist in scientific research, academic writing, and more.
“For example, if you do scientific research ha-ha, you use it to assist you in language improvement, I think he can truly improve your productivity. “(N17).
“The application of that clinical decision-making system mentioned earlier…. can reduce a lot of your work. If you’re on top of the clinic and you use AI, it can reduce your workload to a certain extent. “(N5).
F strong willingness to learn and use AI technology
The respondents demonstrated a positive attitude and strong willingness to learn and utilise AI. The majority of the respondents considered it essential to acquire knowledge about AI, believing that it would be highly beneficial for future development. Additionally, some mentioned their limited understanding of AI and the challenges in discerning the authenticity of AI-generated data, emphasising the need for systematic training on how to use AI correctly.
“Whether it is graduate students or undergraduate students, the use of this area is particularly high, and then we have a characteristic of student use, regardless of the authenticity…. Therefore, I feel that centralised training is very necessary. “(N10).
“I’m sure in the future I will be exploring the use of that and ways of applying AI and then integrating it with what we do in nursing to do some of that work. “(N13).
Subtheme 2.2: Neutral attitudes of concern and scepticism
A reliability, authenticity and security concerns
Some respondents expressed scepticism regarding the data and information provided by AI, questioning its reliability and authenticity. Additionally, they raised concerns about the safety of implementing AI technologies in clinical settings, particularly the potential risk of harm to patients.
“Whether the content developed by the AI technology is truly able to report the real situation or the information provided is completely correct. If he provides some wrong information, it can lead to some potential damage to the patient instead. “(N13).
“I asked him (AI) to find me a reference, but the references he found me, he had some that were virtual or fictionalised. “(N16).
B concerns about the overdevelopment of AI in certain areas
Some respondents argued that the development of AI should be subject to limitations, as not all fields require extensive AI advancement. They expressed concerns that the proliferation of AI could disrupt certain industries and potentially increase unemployment rates.
“I think it can be used appropriately, but there should be some limitations, that is, in some areas it can be very developed, let us say this military area ah whatever, but for example, in the nursing area, you cannot develop it too much particularly advanced, and I do not think if it is not a threat to people as well. “(N15).
C AI lacks a human touch
Respondents perceived nursing is a discipline that emphasises humanistic care and prioritises patient-centred approaches, mental health, and effective nurse–patient communication. The respondents expressed concerns that integrating AI into nursing could result in a lack of humanistic care and diminish emotional interactions between patients and nurses.
“I’m neutral, I guess, because I think our nursing industry is actually an industry with humanistic care…. because AI may run errands and those should be able to help us, but humanistic care ah whatever, I think it is still lacking. “(N6).
“For example, one of the biggest problems with the digital human lectures he has is that he has no emotion; he does not have any change in expression. He also cannot promote this better understanding of each other through this eye contact or other forms of communication. “(N16).
D difficulty in presetting and accumulating data
Respondents mentioned that the use of AI requires a learning and accumulation process and may encounter many obstacles and difficulties in the early stage of application. Respondents indicated that the process of applying AI is difficult and requires considerable time and effort and that this process may be “discouraging”.
“When you go through this assistant coach to communicate, he’s also going to be before you have to preset some big databases, like we, for example, the big models must preset something, he’s caught enough of it that he’s able to go and form that answer. “(N16).
Theme 3: Practical dilemmas in using AI
The theme includes three subthemes: the ethical and legal challenges of AI, the challenges of AI, and the external challenges of AI.
Subtheme 3.1: Ethical and legal challenges of AI
A invasion of privacy and data breaches
Privacy rights and data protection in AI technology have consistently remained focal points of public concern. The respondents expressed significant apprehensions regarding potential data breaches and compromises in personal privacy when utilising AI systems, which may lead to detrimental consequences and legal implications.
“The patient is having an interaction where he may know some of your personal information. AI has the potential risk of compromising personal privacy. “(N13).
“I think information leakage is an issue I’m more concerned about. I’m worried about what you’re telling him. This information is definitely stored. “(N17).
B the legal responsibilities of AI are poorly defined
The legal implications of AI encompass multiple dimensions, with the delineation of legal liability being one of the key concerns raised by respondents. Respondents raised two critical liability concerns: (1) accountability distribution among developers, manufacturers, and users for AI system malfunctions causing harm, and (2) responsibility attribution for errors occurring during autonomous decision-making processes. The respondents indicated that the ambiguity surrounding the determination of legal responsibilities presents a significant challenge to the broader adoption of AI technologies.
“Because of our personal use of new technology, you have a kind of that kind of medical risk, and who is truly responsible for that? “(N4).
“If AI immaturity causes damage to this patient, who is responsible for that damage? This is the main question. “(N2).
C definition of author contributions
The attribution of authorship to AI systems remains a contentious issue in academic discourse. Some respondents maintained that when AI technologies are employed in research, the specific contributions of AI should be explicitly delineated, with clear declarations regarding which aspects of the work are assisted by AI. However, considerable debate persists regarding both the methodology for such declarations and the fundamental question of whether AI can legitimately be recognised as a contributor to authorship. This ongoing controversy represents a significant challenge in the ethical application of AI within academic research.
“From the scientific field then it is this question of author contribution ha-ha, it is definitely something that should be labelled to indicate in which part of it how you’re using AI. “(N17).
“There was a discussion about whether GPT could be published as an author on this article, and it has been and still is highly debated. “(N4).
D academic misconduct
Respondents acknowledged AI technology’s growing adoption and convenience, but expressed significant concerns about its potential for facilitating academic misconduct. The respondents emphasised the necessity of establishing clear boundaries for the use of AI to prevent issues such as breaches of academic integrity.
“However, this also has some flaws ah, for example, before posting articles well, those scholars Chat GPT generated content directly on their own articles, there is an academic misconduct and so on. “(N4).
Subtheme 3.2: AI’s own challenges
A poor quality AI products
Respondents perceived that AI is currently in a phase of rapid development, with many technologies and products still lacking maturity and refinement. The respondents expressed concerns regarding the suboptimal quality of AI-related products, which they believe hinders their effective application and broader adoption.
“Now there are some under this banner of virtual simulation, right, and then some of the things that come to be pushed in schools, and we tried it out, and we just think that he that thing is truly more of a paediatric thing, it is kind of thing that kindergarten dolls play with currently. “(N14).
B lack of generalisation of predictive models
The respondents indicated that the widespread adoption of AI technology faces limitations, as decision-making systems or predictive models designed for specific regions may not be directly applicable to other areas.
“After building the predictive model, it still has some limitations in terms of generalizability; for example, maybe this hospital can take this indicator, but in another hospital, he may not be able to collect these indicators, so there may still be some limitations in terms of its generalizability. “(N12).
C side effects of generative AI use
Respondents perceived that the formidable capabilities of AI offer substantial convenience to users but simultaneously introduce certain adverse effects in its application. The respondents observed that human nature tends towards dependency, and as AI facilitates time and effort savings, users may become increasingly reliant on these systems, potentially leading to a gradual decline in independent critical thinking.
“We have a lot of people now, after he uses AI more, he gets a taste of that sweetness and he keeps going to rely on AI, and then he loses one of his self-judgments and ability to think. “(N1).
“If students are so dependent on this AI, their motivation, their learning behaviours will definitely be affected…. because this dependency is truly strong in human beings. “(N15).
Subtheme 3.3: External challenges of AI
A user acceptance of AI
The respondents indicated that the potential integration of AI into nursing education or clinical practice is contingent upon its acceptance by both students and patients. A prevailing perception among many individuals is that AI represents an impersonal, machine-like entity, which may contribute to a lack of trust in its applications within healthcare settings.
“In terms of knowledge acquisition, we’re used to having a teacher imparting knowledge and then maybe a new form of use people will have a process of adapting to. “(N10).
“The challenge is that it should come down to the willingness of the individual to embrace this one new thing because it may not have been realised that this one thing is important and the positive effect that it can have on them as a person in nursing. “(N4).
B applications that lack fairness
The respondents expressed concerns regarding the inequitable accessibility of AI applications, citing examples such as the limited availability or paid subscription models of products such as ChatGPT. They further highlighted that regional economic disparities and variations in students’ learning capabilities contribute to unequal levels of AI adoption and utilisation across different demographics.
“For example, that question about licenced versus pirated versions…. You need that key if you’re licenced, and that key is something you have to buy. “(N4).
“In this kind of (inland China) area then…… has little exposure to this kind of about care robots. “(N9).
“The average institution is the student his kind of learning ability; he is actually still very different from the students of the prestigious schools. “(N15).
C age challenge
The respondents identified age and educational background as significant challenges in the adoption of AI. They noted that older individuals and those with lower educational qualifications may exhibit reduced receptiveness and learning capacity toward AI technologies, consequently diminishing their motivation to engage with such systems.
“If we invoke AI technology now, inevitably there are some relatively older nurses in the clinic who cannot operate as smoothly as just slightly younger ones. “(N5).
D funding challenge
The respondents widely identified financial constraints as a significant barrier to AI implementation. They emphasised that the development and application of AI technologies entail substantial costs, and without adequate financial support, the practical deployment of these technologies becomes unfeasible.
“The first thing is that if you are on the economic side, this use of AI technology, then for his algorithm optimisation and the design of the model, it is also an enormous expense. “(N7).
“The first challenge, if you introduce the practice, is definitely the financial aspect ah. I personally feel that one of the bigger things about AI is that it is very costly. “(N10).
“We had a member of our group who wanted to do something in this area. However, owing to a lack of funds, he would not be able to implement it. “(N9).
E nursing lacks AI-related education
The respondents indicated that nursing education lacks a training board for AI-related knowledge and that this aspect is missing, leaving nursing professionals with a lack of AI knowledge background and hindering the integration of the disciplines of nursing and AI.
“We lack systematic learning, so I feel that as a school then, I feel that I can offer lectures or trainings about it. “(N10).
“For us researchers, maybe we have not learned about AI in particular depth, just because we do not have a background in this area; well, this background in AI may lead to an understanding that will be a little bit shallower and not deep enough. “(N13).
