Artificial intelligence — science fiction or already reality?
Artificial intelligence — science fiction or already reality?
One of the most exciting topics within the diverse aspects of digitisation is AI — Artificial Intelligence. Many people still place this in the realm of science fiction, but it is already part of our reality. Numerous researchers worldwide are working on the development of AI systems for almost all areas of our everyday and working life. But exactly how far has the development of such artificial intelligence actually progressed? In order to find answers to these and other questions, we have interviewed an expert in this field, Prof. Dr. Tonio Ball from the Medical Faculty of the University of Freiburg, who is investigating AI systems and their application in medicine.
Prof. Dr. Ball, could you please start by briefly explaining in which area exactly you are conducting research at AI?
Prof. Dr. Ball: The main topic of our research group at the University Hospital of Freiburg is the use of AI algorithms in the medical field. This mainly involves diagnostics, i.e. the analysis of data sets, a task that currently still has to be carried out by doctors and which is not only time-consuming but also prone to errors. A program able to detect deviations on its own would be a solution that would not only relieve the doctors’ workload, but also reduce the error rate significantly. A good example of this are already relatively highly developed algorithms that are to be used in radiology and serve to analyze images after brain scans and determine whether the patient is healthy or sick. But we want to take the whole thing even further and use AI systems not only for classifications, but also develop algorithms that can be used to control robots, e.g. as assistance systems for paralyzed patients, or to optimize hospital organizational processes. For example, a program that monitors the number of drugs in stock in a hospital could, over time, independently take care of reordering them in time by recognizing patterns, i.e. how quickly which drugs are used up. In this way, there is a very wide range of possible applications of such methods in hospitals, which ultimately aim to simplify processes and thus help the staff and free up time for more important activities.
When do we actually start talking about “artificial intelligence”, i.e. what is the exact difference between a conventional algorithm and an algorithm that would be classified as artificially intelligent?
Prof. Dr. Ball: This is a good question to which, unfortunately, there is no really satisfying answer. This is simply because there is no general definition of what exactly intelligence is. For example, where does intelligence begin in the animal kingdom. Is a mouse intelligent only because it can make independent decisions? Is there no intelligence in a single worker bee because it only performs its defined function in the swarm? Can something that is not alive even possess intelligence? These are all difficult questions to which there are no generally accepted answers outside the technical field.
When we talk about artificial intelligence in our research area, we generally refer to programs that are capable of learning. If you now have an algorithm that can only handle one task, no matter how complex it is, once it has been programmed, it is done. That would be a regular program, where every decision that can be made was predetermined by the programmer. Most of the algorithms that would now be described as artificially intelligent are related to so-called “machine learning”. Another feature that, together with machine learning, often leads to algorithms being placed in the category of artificial intelligence is the ability to plan abstractly, or to represent complex relationships.
When you speak now of ” adaptive algorithms”, does that mean that these algorithms are simply fed with information and thus gradually become more and more “intelligent”, or how do you have to imagine that?
Prof. Dr. Ball: Whether more data automatically leads to a better functioning of the whole thing is another question, but often it is. The principle behind this “machine learning” is often artificial neural networks consisting of interconnected artificial neurons, i.e. nerve cells. The connections between the single neurons have a certain weight, they tell us how much influence one neuron has on the others. Depending on how these weights are now set, the abilities and function of the entire network also change. So when we speak of a “learning process”, this is ultimately nothing more than the systematic optimisation of these weights. A typical example would be the classification of images into the categories “dog”, “cat”, “mouse”, “flower”. The network starts to classify images that it receives into one of these categories, while specially developed algorithms, depending on the result, optimize the weights of the artificial neurons and thus slowly ensure that the network delivers increasingly better results over time, i.e. “learns”. However, this approach does not have a 100% success rate. It is not unusual for such a learning process to fail and for the network not to learn how to master the given task, since this area of research can give rise to many problems that cannot be identified or solved at the moment.
Do you think that artificial intelligence could eventually be developed far enough that it could compete with the intellect and expertise of a human being, or are there limitations to such learning methods?
Prof. Dr. Ball: In some areas such artificial networks are already far ahead of humans. A well-known example is the victory of an AI against the world champion of the Asian board game “Go”. While the battle between man and machine in chess had been lost since the end of the 1990s, since even the chess programs of that time were able to calculate all possible moves very far and make the move with the highest calculated chance of victory, Go was for a long time considered a bastion of human superiority. There are so many different moves and positions in Go that not even the most powerful computer in the world could calculate all the possibilities fast enough . In chess it is also much easier to estimate for which player a certain board position or move is favourable than in the case of much deeper Go. Since conventionally programmed computers are overwhelmed when it comes to strategic, intelligent thinking, it was long assumed that no program would be able to beat a Go professional in the foreseeable future. That’s why it was a big surprise when in 2016 an AI program that was also trained with the “deep learning” method, a form of machine learning, defeated the world’s best player . And now, only three years later, there are already further developed and even faster learning AI systems that can easily beat the program of that time.
However, such developments, as with the Go program, are very difficult to predict. Only a few years before 2016, hardly anyone would have thought it possible that Go could be defeated by a computer in such a short time. But even if future predictions in this area are very difficult, I do not see any limitations in general for the development of AI systems that are superior to humans in many areas. Similar to the Go example, it is quite plausible that with the help of e.g. deep learning, the programs will gradually become better and will be able to perform more and more tasks and activities better and more efficiently than humans. But whether the AI systems will be limited in some areas is an exciting question that cannot be answered at the moment. However, I believe that development in this area is progressing so quickly that we may not have to wait as long for the answer as many people still think.
So the programme that beat the Go world champion in 2016 did not calculate its moves in advance, but reacted to new situations and independently searched for an answer or solution?
Prof. Dr. Ball: Exactly, this was not, as in some chess programs, simply “Which move is the best”, but the system learned based on a lot of test games, which are strategically good moves that make sense in the long-term course of the game and then decided over and over again for the move it thought was best.
Of course, you are right in saying that it is difficult to predict future direction of such a complex issue: How long do you think it will be before AI programs have an impact on the workplace and establish their presence on the market?
Prof. Dr. Ball: Well, partly that has already begun. Programs already exist for customer service centers that can independently transfer a caller to the right processing point. Intensive work is also being done in the legal field on automated legal advice systems that could soon take over part of the work of paralegals. The development of all these programs is advancing very quickly and the first of them are already on the market or will soon be. Even though we are not yet feeling the changes so strongly at the moment, it is considered quite realistic that the increasing use of AI in the workplace will in the long term have consequences similar to those of the first wave of industrialisation at the end of the 19th century. However, this should not necessarily be seen as something negative; especially in our field of expertise, the interaction of AI systems and employees will ultimately benefit many patients. Today, when a doctor is talking to a patient, it is often the case that the doctor is already sitting at the computer and typing in symptoms and other information during the conversation. Especially in situations like these, an intelligent program that listens to the conversation and is able to automatically filter and independently transfer relevant information would be extremely practical. By supporting doctors in such a situation by AIs, they would be able to concentrate better on their patients and, above all, save a lot of time, which is particularly important in hospitals. What I mean to say is that the changes that artificial intelligence brings with it can be positive, it depends to a great extent on what we make of it and how we shape the changes that are coming.
So do you think that artificial intelligences, as in your example with the doctor, will act more as a kind of supporting help system for human employees, or do you also see the risk that many jobs may be lost due to the increased use of AI systems?
Prof. Dr. Ball: I think that both of these cases will occur. The potential applications of AI systems are so broad and so efficient in their implementation that some, if not many, jobs will definitely be threatened by AI. On the other hand, in many areas, a cooperation scenario will emerge in the first place. It remains to be seen to what extent the division of labour will shift to the AIs in the future. Of course, human resources are indispensable in areas such as medicine. In other sectors, however, it could also happen very quickly. For example, as soon as autonomous driving systems can drive safely enough and are financially viable, we would suddenly need far fewer or no more taxi and bus drivers.
My special thanks go to Prof. Dr. Ball for his willingness to be interviewed and for his highly interesting remarks.
Author: Linus Plesnila, Glasford International Deutschland Research & Analytics