Artificial Intelligence and Its Role in Medicine
Zacharias Voulgaris, PhD
Zacharias Voulgaris
CTO at Data Science Partnership
PhD in Machine Learning at University of London
Introduction & Definitions
Ever since the hardware technology caught up with the Artificial Intelligence (A.I.) algorithms that were developed in the past decades, the whole field has been experiencing a kind of Reconnaissance, while its applications across different fields have multiplied. First thing’s first though. By the term A.I. we refer to advanced algorithms for optimization, data processing, and automation, applicable in a variety of fields. The most popular such fields are data science (esp. Machine Learning), Robotics, and various computer systems that deal with data in one way or another. Even some more specialized fields, such as logistics, are now benefiting from A.I. Naturally, as the field of A.I. matured, it became useful to more and more other fields too, where the stakes are higher and there is little room for error. Although cyber security is the field that most people are aware of, there is a lot of work on how A.I. is useful in medicine too, in a number of way.
Beyond the Field of A.I.
A.I. is much more than a scientific field, however, and more than a series of applications of clever algorithms. For many people, A.I. is a potential solution to many of our problems and a path for accelerating scientific research, as well as improving society’s infrastructure. In other words, the implications of A.I. are so far-reaching that it has a profound effect on our lives, on many levels. From super-clever apps on our phones, to smart vehicles that can drive themselves, to even robots that can walk in harsh terrain, or perform complex operations (and by operations I mean all kinds of operations, not just the traditional factory-related processes).
Nowadays even things like medical procedures are greatly facilitated with A.I. while the use of A.I. in diagnostics is already quite wide-spread. Bottom line, A.I. is not a tech field isolated from the world, or something accessible only by techies, but rather a well-integrated part of science that already adds a lot of value to many organizations and society overall, in the places it has been deployed.
Types of A.I. Systems
As mentioned previously, there are different types of A.I. out there. For the purpose of this article, we’ll use a different taxonomy though, one that is more suitable for the medical field. After all, A.I. is adaptive, so we might as well adapt it to the application at hand.
So, depending on the level of involvement, A.I. can be passive, reactive (active in a supportive role), or fully active (autonomous). Currently most A.I. systems are in the first two categories, though fully autonomous A.I. systems are starting to come about.
Passive A.I. systems are the ones that never take any initiative. These are basically the various black box applications used in various areas, such as data science, where you give an A.I. system some data, it learn that data, and then makes predictions about it, but only after it is given some new data. Such systems may also create different ways to express the data they are given, but they always react to their users.
Reactive AIs are the more advanced systems that may seem like they are taking an initiative, though more often than not, this is an effect of clever programming, since they are actually reacting to their user’s actions or commands. Things like chatbots are in this category, just like many AI-powered robots used in medical procedures. All these AIs are good at a very specific task and are geared towards collaborating with their user, forming a partnership of sorts.
Fully autonomous AIs are usually A.I. systems developed for the robotics field as they are quite sophisticated and able to understand their environment to some extent, make decisions, and act based on external events. Things like roomba, the cleaning bot, as well as certain advanced drones, are in this category. Naturally, self-driving cars can be seen as AIs of this category too, especially once they are allowed to hit the road (both figuratively and literally).
Applications of A.I. in General
A.I. has a variety of applications, spanning over all tech-related fields and some not-so-technical fields tool. The main applications are the following:
- Virtual Personal Assistants (VPAs). This involves systems like Siri, Alexa, etc. that are geared towards performing fairly simple tasks on our behalf, much like a human personal assistant would do in an office environment. These systems are expected to become even more popular in the future as the chatbot technology that’s behind them improves.
- Video games and Virtual Reality (VR). This involves systems running on a computer while the user interacts directly with them, for a certain amount of time, mainly for entertainment purposes. However, VR systems often have an education aspect to them, such as when they are used for training people to function in certain high-risk environment, without any actual danger.
- Prediction systems specializing in retail. These systems are geared towards predicting if a customer of an online store will buy a given item, what other items he/she would be interested in, as well as how likely (s)he is to buy another such item in the future.
- Healthcare and Medicine. We’ll look at this in more detail in the next section of this article.
- Online Customer Support. Chatbot AIs are also found in the customer support domain, particularly when this takes place on the web. Such systems interact with customers answering queries, providing potential solutions to their problems, and directing them to specialized personnel when necessary.
- Security Surveillance. This is a fairly sensitive topic as many people view it as a potential violation of privacy. However, when done properly (i.e. preserving people’s anonymity), it can help prevent crime and when this is not possible, facilitate law enforcement in tracking down and capturing the criminals. Security surveillance AIs make use of camera feeds mainly, though they may use additional data such as GPS, audio streams, etc.
- Smart Home Devices. This is by far the most promising A.I. application and the one that is bound to continue being around in the future, in one way or another. It involves smart apps as well as AI-based operating systems small devices that are connected to the internet and can help facilitate everyday tasks, such as providing us with real-time information about the weather, performing inventory checks in our fridge, and monitor certain activities (e.g. exercising routine, sleeping, etc.).
- Smart Cars. This is not just self-driving cars, but also cars that can facilitate driving or parking, through an array of sensors and A.I. systems monitoring them and using their inputs to make decisions. Smart cars are an application of A.I. that is disrupting logistics at the moment and it’s gaining ground as computer vision become better.
- Fraud Detection. This has to do with AIs that are geared towards spotting unusual activity in financial transactions, denoting potential fraudulence in the processes involved. They are a special kind of anomaly detection systems, trained on financial data.
- News Generation. Although this sort of application has received some bad publicity lately, with the rise of fake news, news generation AIs are geared towards developing stories based on facts. These are supposed to be unbiased and reliable, written in a humanly comprehensive language, usually following certain templates created by journalists and such.
- Music / Movie Recommendation Services. This kind of application involves the use of user data in music and movie platforms (e.g. Spotify and Netflix, respectively) in order to make predictions about what other media the user would like to consume. The creation of play-lists in an online radio is a manifestation of such as a recommendation service, while the suggested movies or TV shows to watch next are another scenario.
- Smart phones and tablets. This is the most common application of A.I. for facilitating our lives though automation of processes, optimal use of the limited power resources such a device has, and the ability to run a series of clever applications, all while making use of a variety of sensors, such as the device’s microphone, its camera, various movement sensors, etc.
Medical Applications of A.I.
The applications of A.I. in the medical field involve three main types of processes: diagnostics, treatment, and medical research. Let’s look at each one of them in more detail.
Medical diagnosis using A.I. systems is quite commonplace today, particularly whenever images are available. A hi-res fMRI scan, for example, can contain crucial information about a patient’s condition, while when combined with other data (e.g. the patient’s vital signs, as well as demographics data, or data related to the patient’s habits and genetic makeup) can be more than enough to predict if a certain disease has manifested or if it will manifest in the imminent future. This sort of systems tend to have high accuracy, comparable to domain experts, and although they may not always be able to explain why they arrive at a certain conclusion, they can be a valuable aide to the doctor handling the patient.
Medical treatment with A.I. systems is still very experimental. However, specialized robots can greatly improve the chances of certain challenging operations, enabling the surgeon to perform fine movements and even guide him/her through the patient’s body, reducing the risk of accidentally damaging nearby tissue. Although such AIs are still unable to perform a whole medical operation on their own, they may be able to do so in the future. It’s quite likely, however, that the matter may be a bit complex from a legal standpoint, while it would require a lot of confidence from the medical professional’s part to trust an AI like that to perform an operation on his/her behalf.
Medical research that involves A.I. systems is currently related to gene expression data as well as the key factors influencing the presence or not of a medical condition, so that the latter can be averted (at least in theory) through the control of these factors through the cultivation of the corresponding habits. Also, the prediction of how a new drug would affect a person and the control of its potential side-effects is a hot topic of research where A.I. systems can add a lot of value.
The main issue of the use of AIs, however, when it comes to applied medicine is the ethical aspect of all this. What will happen if a patient dies because of a wrong diagnosis provided by an A.I.? Or if a patient is treated in a way that prolongs his/her stay in the hospital, potentially due to complications that an A.I. system wasn’t trained to handle? Questions like that need to be addressed before AIs can be trusted further in this field, where lives are at stake, and where lawsuits are quite commonplace. Also, the fact that most A.I. systems today are unable to explain how they reach their conclusions is an issue that makes both the patients and the medical professionals uncomfortable, especially in cases of terminal diseases.
Some Thoughts about the Future of A.I. and A.I. Safety
Although many people fantasize about A.I. based on some sci-fi films they have watched, or some sci-fi books (Isaac Asimov comes to mind), the future of A.I. may not be so sensational nor so hazardous as they may think. Most A.I. experts expect A.I. to grow in the years to come, becoming more widely applicable and perhaps even ubiquitous in certain geographical regions (e.g. certain high-tech cities like Singapore and Dubai, as well as many cities in America and Europe).
However, the dream of a universal A.I. that can perform general tasks with equal dexterity as a human being (something known as Artificial General Intelligence, or AGI for short), is bound to remain a dream for at least a couple of decades. And even then, it will probably not be accessible to the majority of people unless it becomes mass produced and the cost of running such a system drops dramatically, something that is quite debatable at the moment.
After all, A.I. systems can be costly to operate and they require a lot of data. So even if the algorithms improve, it may take some time before scaling them becomes affordable enough for a more wide-spread use of this technology. Also, as A.I. systems evolve, we can expect to see more and more of them becoming closed-source (proprietary), making access to them limited, while certain A.I. systems may be accessible only to a minority of users within the organization owning that machine.
Naturally, as A.I. becomes more evolved and more complex, there are some inherent dangers that may come to manifest. This may not necessarily be related to warfare (although that’s a possibility too), but most likely have to do with poorly defined objectives and sloppy programming that may make the AIs ignorant of things that are important to us (e.g. privacy) thereby violating them, all while they think they are helping us. Worst case scenario, such AIs may create social-economical issues, and risk the lives of many people, depending on what kind of tasks are outsourced to them. For example, if someone trusts an A.I. system to undertake the whole process of treating a patient, without taking into account the patient’s psychological state, this may lead to sub-standard treatment and perhaps even a decline in the patient’s health.
Fortunately, there are many people actively researching the potential problems of self-sufficient AIs and how these issues can be tackled proactively or retroactively, whenever that’s possible. So, the future of A.I. may not be as flashy as movies make it out to be, but it can definitely be interesting and potentially void of issues. However, the extent to which advanced AIs can be powerful aides of ours depends on how prepared we’ll be against potential misunderstandings from their part, regarding what we expect from them and what they can and cannot do. Just like any other technology, A.I. requires foresight and responsible usage, if it is to be a fruitful asset with limited risks.
Conclusions
A.I. is a highly practical field that involves much more than robots and computer systems for techies. It has a wide variety of applications across different fields, many of which are accessible to the everyday person, requiring little to no understanding of the complex processes the A.I. system has. Among its many applications, A.I. can offer a lot in the medical field, through a number of processes, involving diagnostics, treatment, and even scientific research.
The ethical aspect of this sort of application is something to consider however, since an A.I. system cannot be held responsible in the case of a patient getting worse or even dying. However, over time A.I. is bound to get better and even if it doesn’t reach the heights portrayed in the sci-fi films or books, it is bound to continue being useful and offering a lot of value, both in medicine and other areas.
We just need to be aware of the potential liabilities an advanced A.I. may exhibit and take steps in preventing them, or at least, tackling them, should they come into manifestation. This way, the positive effects of A.I. can be maximized, all while keeping the world safe from its potential side-effects.
References
- Z. Voulgaris & Y. E. Bulut, Artificial Intelligence for Data Science, Technics Publications, 2018 (currently in production; available in Autumn 2018).
- 10 Examples of Artificial Intelligence You’re Using in Daily Life, online article available at http://bit.ly/2o6Ikp7 (last accessed 22/8/18).
- Artificial Intelligence Is Infiltrating Medicine -- But Is It Ethical, online article available at http://bit.ly/2LgSmgh (last accessed 22/8/18).
- Artificial Intelligence in Medicine: an Introduction, online article in Open Clinical available at http://bit.ly/2w63cRB (last accessed 22/8/18).