Artificial intelligence (AI) is a field of computer science that equips machines with human-like intelligence and enables them to learn, reason, and solve problems when presented with data in various formats. The term AI encompasses a wide range of words. To define the relationship between the words, there is data analytics, AI, and robotics in data science, as well as machine learning, artificial neural networks, and deep learning. AI uses a combination of machine learning algorithms, artificial neural networks, deep learning, and other data analysis techniques to achieve intelligent capabilities (Fig. 1). The healthcare sector is experiencing a paradigm shift with the recent increasing trend of mass digitization of information. Neurosurgery is often at the forefront of innovative and disruptive technologies, which have similarly altered the course of acute and chronic diseases [22]. The purpose of this study is to gain a better understanding of AI, its role in neurosurgery, and its future development directions.

Humans have long wanted to build thinking machines. Turing [41] suggested the possibility of developing AI and devised a test method to confirm it. The Turing Machine led John von Neumann to develop a standard computer architecture. This is considered to be the beginning of AI. Rosenblatt [31] developed an artificial neural network called the perceptron using artificial neurons that mimic human neurons. Following that, AI research received a great deal of attention and a substantial research grant from the U.S. Department of Defense. However, Minsky and Papert [24] criticized the limitations of the perceptron by publishing a mathematical proof that the perceptron can be applied to linearly separable problems such as AND or OR, but not to XOR problems. It was not until the introduction of expert systems into the industry in the 1980s that AI research, which had been stagnating for many years, received attention again. A lack of computer resources, excessively difficult management, and low cost-effectiveness, have however, led to a decline in interest in the research of AI. In the 2000s, deep learning technology based on Hinton’s Deep Belief Network began to show potential for practical use [16,25]. With this as an opportunity, deep learning technology re-emerged. This demonstration of the potential of deep learning led Google to acquire DeepMind Technology in 2014, a company that specializes in reinforcement learning. Then, AlphaGo beat Lee Sedol in 2016 and Ke Jie in 2017, and AI was once again thrust into the spotlight. Since the advent of deep learning, AI has been developing rapidly, and it is changing so rapidly that it is impossible to predict which technology will receive attention in the future.

The ultimate goal of AI is to create consciousness. Thus far, humans have copied various functions of the brain, but not the consciousness. It is doubtful whether it will be possible in the future. As with any machine, AI is still a tool that produces output when input is provided. The current AI works like a slave, but does not seem to have emotions or free will. The intelligent machine can easily imitate phenomenological features of consciousness, but do not seem to be able to shape psychological features [13]. It is not possible for the machine to receive input or output by making plans on its own with subjectivity. However, human beings are aware of their existence, of the process of input and output, and of the feelings that accompany these processes. That cognitive ability is defined as consciousness. The Turing test can determine whether a machine has intelligence, but it cannot determine whether it is conscious. This has been proven with the Chinese room argument [33]. However, if AI becomes conscious, it would be very difficult to distinguish it from humans. It would also be very difficult to prove that a first-person perspective exists on the machine. The world after the realization of conscious AI will be revolutionary. Currently, the phenomenon of consciousness has been accepted as a phenomenon emerging from the complex connection of neural structures. It is advocated that consciousness can only be realized in a particular physical system because consciousness has properties that depend on physical constitution [4]. Nevertheless, proponents of artificial consciousness believe that it is possible to construct a system that can emulate this neural correlate of consciousness interoperation [12]. The mechanism of consciousness has not yet been elucidated, and the means for its research are inadequate. Computers are unable to keep up with the connectivity and complexity of the brain, but if connectivity explosively increases and complexity is overcome with the development of technologies such as quantum computing in the future, it has been carefully predicted that a consciousness phenomenon will arise in electrical systems. As neurosurgeons are privileged to have access to the human brain, they are obliged to answer these philosophical questions in a scientific manner.

People working in ICU routinely interpret large, uneven patient datasets, including data types such as physiological waveforms, continuous electroencephalograms (EEG), laboratory examinations, and radiologic examinations. This has proven to be very difficult and time consuming, and ICU workers may not be able to integrate critical information into clinical decision making in a way that could affect patient outcomes. Therefore, there is an important clinical need to support clinical decision-making by using AI to automatically interpret large amounts of data collected in ICU and improve patient care.

The majority of patients in neuro ICU are bedside patients. Therefore, EEG and functional MRI, which are used to check consciousness, are necessary for ICU patients. Claassen et al. [10] used support vector machines to detect cognitive motor dissociation to rapidly read EEGs from unresponsive healthy patients.

Continuous intracranial pressure monitoring is a routinely performed method in the neuro ICU to detect increases in intracranial pressure associated with decreased cerebral perfusion [5].

Raj et al. [29] developed a dynamic multiple regression predictive model for predicting total-cause mortality 30 days after traumatic brain injury. The model used parameters collected from continuous intracranial pressure waveform measurements obtained from intraventricular catheters or intraparenchymal probes and other physiological measurements [29].

Delayed cerebral ischemia is a usual complication after subarachnoid hemorrhage in neuro ICU. Tanioka et al. [39] attempted to construct a random forest prediction model to predict vasospasm by delayed cerebral ischemia and angiography after subarachnoid hemorrhage. This study demonstrates that machine learning can improve predictive performance by integrating large numbers of multidimensional datasets [39].

Numerous studies have demonstrated that predictions obtained from machine learning can outperform conventional predictive statistical models when applied to neurosurgery patients [11,15,17,19,21,26,34-37,40] (Table 2). Machine learning can be used not only to improve diagnosis, but also to predict patient outcomes and diagnoses.

Virtual development tools powered by AI allow for the development of new molecules or identification of new molecular targets within a virtual environment, as well as the selection of the most appropriate molecules for clinical trials. IBM Watson, an AI tool, was used to identify additional RNA binding proteins that were altered in amyotrophic lateral sclerosis [2]. AI can be used to identify patients who may be more responsive to specific medications and predict possible complications, thereby optimizing patient selection, speeding discovery, and reducing cost burden [45].

The use of AI in real-time data analysis has important clinical utility in primary and secondary prevention (i.e., disease prevention and early-stage intervention). For example, the current outbreak of COVID-19 is modeled using big data, with or without social distancing, many countries have imposed preventive measures such as tracking, quarantine, and evaluation to determine how much of an epidemic it is [47].

In the field of neurosurgery, AI is already being used in many areas, and there will be more in the future. Ultimately, it is our responsibility to keep pace with this evolution to provide meaningful outcomes and personalize each patient’s care. Rather than blindly relying on AI in the future, neurosurgeons should understand it thoroughly and use it to enhance their patient care.