Making Sense of AI in Healthcare: What It Is and What It Isn't
AI has been making waves in the healthcare industry, with a growing number of companies introducing AI-powered technologies they claim will revolutionise patient care and clinical processes.
It can be challenging to discern what is genuinely AI-powered and what isn't, largely because definitions and understandings of AI vary widely. It’s a loose term; it means different things to different people.
But as AI becomes more pervasive in healthcare, it’s important to understand the different types of AI and what they can and cannot do to set realistic expectations and make informed decisions based on what you truly need.
So in this article, we will:
Start by breaking down the different types of AI
Discuss the debate about what qualifies as AI
Emphasise the importance of understanding AI's capabilities before implementation
The categorisation of AI
The capabilities of AI can be categorised into three main types: narrow, general, and superintelligent AI. For now, only narrow AI exists in practical applications, so let's focus on that.
Narrow AI is designed to perform a specific task more efficiently than a human, but it can't function outside of its designated task. This category includes reactive machines and limited memory AI.
Limited memory AI can use past data to make informed decisions, learning from previous experiences to improve over time. Limited memory AI can be further divided into machine learning and Natural Language Processing (NLP).
Machine learning has several subcategories:
Supervised learning: This is like training a machine using examples with correct answers. Imagine teaching it to identify tumours on MRI scans. After enough examples, the machine can identify tumours on new scans.
Unsupervised learning: Here, the machine analyses data without any labels, finding patterns to group similar things. For instance, it can cluster patients with similar symptoms together. This also includes generative AI, where the machine creates new data, like images, by understanding patterns in its data.
Deep learning: This advanced learning type uses multiple processing layers to interpret complex data. It is useful for analysing complex images, like pathology slides, to find signs of cancer that the human eye might miss. Deep learning also powers advanced generative AI that can produce human-like text or other outputs based on its learning from large data sets.
Meanwhile, NLPs enable machines to understand and interpret human language. It could, for example, summarise patient records or extract important details from clinical notes, making it easier for healthcare professionals to quickly find relevant patient information.
Reactive machines, on the other hand, are the simplest form of AI. They operate on current data to perform specific tasks, do not possess memory, and do not learn or improve performance. And this is where the debate starts.
The debate: is it or is it not AI?
Reactive AI includes rule-based systems. These systems respond to specific inputs with predefined outputs and do not learn from past experiences. It is an if-then algorithm where human input is required for the "if this" and "then that."
Examples of such systems include virtual assistants and spam filters, which use predefined rules that lead to specific outputs. They automate decision-making but can't improve or change their behaviour over time without human intervention.
So, is it AI or not?
Rule-based systems can be considered a simpler form of AI that automates decision-making processes, but they can't learn because the rules are set by human input. This leads to different opinions on whether or not they qualify as AI. Some believe that because these systems can automate decisions and are technically reactive machines, they can be considered AI. Others argue that since they are incapable of learning and evolving on their own and it's all predefined by humans, they should not be classified as AI. In the end, whether or not rule-based systems are considered AI is subjective and largely depends on one's personal definition of AI.
Understanding what you need
Whether or not you consider specific tech to be AI or not, you need to understand how it works. You need to understand the model of the AI (or possibly not AI) that you're working with and what outputs you can expect before incorporating it into healthcare processes.
Understanding the distinctions between the different types of AI is important for anyone in the healthcare industry. This understanding lets you identify the capabilities of AI technologies and ensures that investments and expectations are aligned with reality, thereby preventing "AI washing." This term refers to the misconception that all AI systems can learn, adapt, and continually improve, which isn't always the case.
So by knowing what to anticipate, you can make informed decisions that align with your needs and goals.