AI in healthcare: What's science and what's fiction

Artificial intelligence has fueled our imagination ever since the term was first coined over 60 years ago. It tends to elicit two different reactions. Optimists envision systems that relieve us of tedious chores or that control autonomous flying taxis. Pessimists think of out-of-control machines like the computer HAL 9000 from Stanley Kubrick’s classic, “2001: A Space Odyssey,” who fiercely protects itself from being shut down by the crew. HAL’s red camera eye has since become an icon representing a dystopic future where machines take over.

So should we be worried that there’s something in the works that’s better or more intelligent than we are, and that will someday surpass us and render us superfluous?

In my opinion, no, we don’t need to worry. A lot of what’s currently discussed in the context of artificial intelligence falls in the category of “narrow artificial intelligence.” Broadly speaking, these are systems that concentrate on solving very specific, clearly defined problems by processing algorithms, which are instructions based on computer science and mathematics. These systems are also able to optimize themselves, a process known as machine learning.

AI is trained by means of examples. The computer uses this data to create a model, on the basis of which it can then process similar data in the same way in the future. It’s true that a lot of progress has been made in this area in recent years, from automatic face recognition in your phone camera to advances in autonomous driving. Algorithms recognize trees, houses, and obstacles and navigate the self-driving car accordingly. This particular type of AI can perform its jobs very well and reliably, and it never gets tired – but that’s all it can do. Algorithms have no consciousness. Language, text, and image recognition, automated translation, navigation systems, and chatbots may be extremely impressive in everyday life, but they’re all examples of narrow AI.

Meanwhile, Clinical Decision Support systems are helping physicians evaluate and make better clinical decisions. For example, AI helps detect nodules in lungs, but only because our team has previously shown it thousands of CT images where experts have annotaded these nodules by hand. This important work is a prerequisite for using AI in medical technology. We work closely with many radiologists who are contributing their medical knowledge. And there’s lots of material available for annotating: Modern imaging modalities easily generate hundreds of high-quality images per patient exam.

This process is the only way that computers can learn patterns and compare them with new data. And again, AI can do this reliably, quickly, and tirelessly. But this type of AI can only recognize what it is trained to recognize, and so the quality of the data and annotations has a tremendous impact on the quality of AI-based recommendations. That’s why it’s so important to have large collections of very high-quality data. Naturally, we also need the right experts so that our software and algorithms can do exactly what physicians and patients expect.

Why are so many narrow artificial intelligence applications emerging right now?

Significant advances have been made over the past few years in the areas of fast computers with specialized chips, storage capacity, machine learning, and ever-larger collections of clinical data for training AI. Combining all these elements reliably and for the benefit of patients isn’t a job for one company alone: In healthcare, we’re dealing with the life and hopes of patients and their families. Algorithms can’t do the job without the extensive expertise of physicians, medical technology companies, and research institutes with access to well-developed knowledge and experience.

At an increasing pace, I experience how AI-powered solutions become more and more common in everyday clinical practice. And how they relieve physicians of routine tasks, enable more precise diagnoses, and give medical staff more time for personal contact with patients. The advances being made in healthcare are staggering in scope.

I like to use the comparison of the autopilot in airplanes. The system takes over navigation and control in routine operation, but the pilot intervenes in important maneuvers and makes the decisions. On the other hand, movies like Kubrick’s “2001” are about “general artificial intelligence,” also referred to as “superintelligence” and “technological singularity.” This would combine decision-making ability under uncertain conditions, a capacity for logical thought, and the ability to plan in order to achieve personal goals – and this would exceed human intellect. Thinking about it might make us uncomfortable, but I believe that general artificial intelligence is still a long way off, assuming it ever becomes possible at all.

When implemented in a trustworthy manner, AI-based systems in everyday clinical practice can help us achieve our goal of promoting a better and healthier life for more people. One of the greatest obstacles to accomplishing this is the fact that a lot of medical data is currently located in different places and isn’t available for training our AI. To me, that's more concerning than being surpassed by a superintelligence.

What about you? Do you take a more optimistic or pessimistic view of the future of healthcare and AI-powered medical technologies?