Basic introduction to Artificial Intelligence

https://doi.org/10.58974/bjss/azbc010

---

Knowledge of the basics and fundamentals is essential to understand the potential benefits and risks of Artificial Intelligence (AI). AI refers to the field of science aiming to provide machines with the capacity of replicate human cognitive functions such as reasoning, learning from experience and self-correction. AI means enabling computers do things that would require intelligence if done by humans.

The term AI was first mentioned by Claude Shannon and Nathan Rochester, two scientists from IBM, at Dartmouth Conference in 1956 (New Hampshire, USA). They presented a computer able to solve problems and learn to speak English, and predicted a completely intelligent machine would be available in the next 20 years. We are not there yet, but AI has already been incorporated into many areas such as renewable energy systems, weather prediction, manufacturing and, of course, medicine.

AI eliminates human error, so in medicine it is expected to reduce the number of misdiagnoses, errors in treatment etc. AI can easily maintain up-to-date medical information from journals and textbooks, and put it into practice for patient care. Another advantage of AI is its speed, reducing the time needed to perform a task.

AI includes three major subfields, machine learning (ML), artificial neural networks (NN), and natural language processing. All of them are connected, and they all have applications in robotic surgery.¹

1. Machine learning

ML offers the ability to learn and solve a problem. Computers can learn from data and experience, so they have the ability to improve their performance on specific tasks. They use algorithms to learn from data and discover patterns, which can be used for future predictions. It allows them to complete tasks that are too difficult to solve with programmes designed by human beings.²

ML can be classified as supervised or unsupervised. The range between these two categories includes semi-supervised learning and reinforcement learning.

In supervised learning, human-labelled training data are given to the ML algorithm to teach the computer a function (i.e. to recognize a pancreas on a CT scan).¹ Humans encode the features into algorithms that are processed by AI.³

In unsupervised learning, the dataset consists of unlabelled examples given to the machine. There is no predefined outcome, so the machine has to explore and learn.

With reinforcement learning, the machine has different attempts to accomplish a task. Through its own mistakes and successes, the machine has a negative or a positive reward and learns how to perform a task (.e.g. playing games or driving).¹

The application of ML into an autonomous robotic system is the future of robotic surgery. The objective is to give the robot the ability to see, think and act without human intervention. Although some autonomous devices already exist, ML is only experimental and will require a considerable amount of work before finding a role in real practice.

2. Artificial neural network (ANN) and deep learning

The name ANN reflects the similarity to biological neurones in the human brain. The ability to learn by example is a human brain function.

ANN is the part of ML that uses this artificial neural network to achieve speech recognition and language translation. It gives the computer the ability to analyze vast amounts of data, and improve in performance with more data. Every time data are processed, ANN uses the results to develop more expertise, discovering complex relationships between data. ML and ANN learning are the essentials to enable machines with intelligence.

ANNs are based on layers of connected neurones. Each connection transmits signals to the next one, similar to the human brain. There are three main layers of learning from the data (Fig. 1).

• Input layer: Consists of digitized inputs that will go through the next layers. It is designed by humans.
• Hidden layer/s: these are responsible for the analysis. There could be more than one hidden layer, which increases the complexity of the analysis. The accuracy of the output depends on the number of hidden layers.⁴
• Output layer: This is the end product, the final part of the network. It is the result of the calculation of previous hidden layers.³

When the model employs more than 20 hidden layers, the technique is then called deep learning.³ The main limitation of deep learning is the large amount of data required for training and the difficulty of collecting relevant data without bias. That is the reason why there is so much expectation about big data and its potential to deliver a precise and efficient database.⁶ Most clinical data are unstructured, such as clinical free text, and need to be extracted using natural language processing.

3. Natural language processing (NLP)

NLP is the ability to programme computers that can understand and process human language. A vast amount of clinical information is in the form of narrative text, which is unstructured and incomprehensible for a computer programme. AI recognizes these words and sentences and understands their meaning, extracting information from unstructured data such as clinical notes, journals, etc.⁵ It converts this free clinical data into analysable, structured data. NLP is currently one of the most widely used AI technologies, and it will be incorporated in to digital medical records in the near future. For example, NLP has been used to identify predetermined words in surgical notes that predict the risk of anastomotic leak after colorectal surgery.¹

This is the first of a series on AI, aiming to describe how it works and its importance to healthcare. It is expected clinicians will work increasingly with AI systems in daily clinical practice. AI will dominate data science due to its amazing learning capability, and its ability to process complex data. It has the potential to improve healthcare delivery.

There are still some concerns regarding AI, such as the cost needed for its development, and some legal and ethical issues, referring to the interaction of AI with the world. Work is still being done on how to ensure AI is used with respect to the approved laws and policies, and which aspects of it will require legal assessment.