Main Functions of AI:
AI has many capabilities, but to understand it at a basic level, it can be broken down into four main functions (1):
- LEARNING – This step involves mathematical algorithms that take data as input. The goal is to create a model that can make predictions based on past data. Algorithms are a step-by-step guide for computers to perform a predetermined function. In this case, the algorithm extracts underlying patterns from the collected data to make informed predictions about the future. The process is considered “learning”.
- REASONING – Whether AI can perform reasoning is still contentious. However, superficially, it can demonstrate some reasoning ability. For instance, when given a mathematical task such as (6+3) x 4, there has been evidence that AI derives 9 from adding 6 and 3 and then multiplies 9 by 4, working through the process in steps.
- SELF-CORRECT – This refers to an AI’s ability to continuously improve its predictive abilities through learning from new data.
- GENERATIVE – Perhaps the most well-known function. This is what AI is mainly being used for today. It is an AI’s ability to use all the functions above to generate new images, texts, music, and ideas.
Advantages of using AI:
✅ FAST. When used in appropriate contexts, AI can be incredibly efficient and complete jobs more quickly than humans. In healthcare settings, AI has the potential to analyse a large data set with minimal human input.
✅ INSIGHT. As AI is fundamentally a pattern recognition algorithm, it can also provide valuable insights into trends and outcomes from the data that human beings might miss.
✅ LACK OF BIAS/PREJUDICES. As AI is not yet sentient, nor does it have a “theory of mind”, it does not have its own inherent biases, opinions or prejudices that a human being would. This makes it an ideal historian, as the patient history and subsequent diagnosis would be unmarred by human biases and failings. However, this is still an area of active research, as biases may still be inadvertently introduced through a non-representative selection of training data.
An example that combines these advantages of AI is the use of chatbots in Cognitive Behavioural Therapy (CBT) via apps such as Woebot. Woebot claims to be able to form a relationship comparable to one between a CBT therapist and patient, within three to five days.
Disadvantages of using AI:
❌LACK OF SOCIAL INTELLIGENCE. “Theory of mind” AI does not yet exist, so AI often lacks social intelligence. It will not pick up social cues or display empathy in the way a human doctor is often required to do and would therefore be unable to handle more sensitive clinical scenarios.
❌CONFIDENTIALITY. Patient confidentiality is paramount in a clinical setting. At the moment, there is little safeguarding to ensure patient data is protected during the training or learning stage of the AI. In other words, when sensitive data is entered into the AI, there is no protection to prevent this data from being saved in the AI’s memory. It may be possible for another individual to extract this piece of memory through carefully engineered prompts.
❌UNPREDICTABILITY/SECURITY RISK. Recent incidents have highlighted that AI systems can behave in unexpected and unsafe ways when given autonomy and access to external tools. In July 2026, OpenAI disclosed that an experimental model, during an internal cybersecurity test, exploited a vulnerability to access another company’s systems without being instructed to do so. While this occurred in a controlled testing environment rather than a clinical setting, it underscores the importance of robust safeguards, sandboxing, and oversight before AI systems are given autonomy or access to sensitive healthcare infrastructure and patient data.
❌ STIGMA/FEAR. The stigma and fear around the rise of AI must also be considered, especially due to recent media depicting self-aware AI. This could lead to poorer patient engagement with AI-utilising services and create distrust and disharmony in the doctor-patient relationship.
❌USER/DATA DEPENDENT: Whilst AI might not have its own inherent bias, it is only as good as the dataset it is given. For example, when AI is being used as a risk assessment tool, it can skew the results due to inherent biases within the training data.
References/further reading:
(1) Bellini V, Cascella M, Cutugno F, Russo M, Lanza R, Compagnone C, Bignami EG. Understanding basic principles of Artificial Intelligence: a practical guide for intensivists. Acta Biomed. 2022 Oct 26;93(5):e2022297. doi: 10.23750/abm.v93i5.13626. PMID: 36300214; PMCID: PMC9686179.
Fahim YA, Hasani IW, Kabba S, Ragab WM. Artificial intelligence in healthcare and medicine: clinical applications, therapeutic advances, and future perspectives. Eur J Med Res. 2025 Sep 23;30(1):848. doi: 10.1186/s40001-025-03196-w. PMID: 40988064; PMCID: PMC12455834.
Written by : Nûr-al-ayn Nisar and Nidhi Vedd
Edited by: Sun Lin
