Can AI really cure all diseases in ten years?

Apparently, Nobel Prize winner in chemistry Demis Hassabis wants to be able to cure all diseases in just ten years with the help of artificial intelligence (AI) – at least that is what many people read online.

Hassabis was interviewed on various topics related to AI on the US news channel CBS News' "60 Minutes." The 49-year-old Briton is the CEO of Google's AI-focused subsidiary DeepMind.

Together with a colleague, he developed the AI ​​model AlphaFold2, which can predict the structures of virtually all of the 200 million proteins known to date. The researchers won the Nobel Prize in Chemistry for this work last year.

"One day we may be able to cure all diseases with the help of AI"

Proteins fulfill a variety of biological functions in the human body. Disruptions in their production, structure, or function can lead to disease.

In an interview, the Nobel laureate said that AI could shorten drug development to months or weeks in the future. He continued: "I think one day we might be able to cure all diseases with the help of AI."

Host Scott Pelley asked, "The end of all disease?" Hassabis said that's within reach. "Maybe even within the next decade, I don't see why not."

AI is revolutionizing medicine

Often, one can estimate a protein's function based on its 3D structure, "because we don't yet know that for most proteins in the human body," explains biochemist and computer scientist Katharina Zweig in an interview with DW. The professor heads the Algorithm Accountability Lab at the Rhineland-Palatinate University of Technology in Kaiserslautern-Landau.

If you know the function and see that the protein structure is altered in certain diseases, this could perhaps be the cause of those diseases, Zweig says. "Then you could develop a drug that prevents a protein from adopting an incorrect structure."

Previously, it would have taken an entire doctoral thesis to identify, calculate, and model a single protein structure. "That took three to five years. Hassabis' AI is truly revolutionary."

Although the causes of disease cannot usually be reduced to individual elements, Florian Geissler, senior researcher at the Fraunhofer Institute for Cognitive Systems (IKS), told DW, "there are many examples where proteins play a major role."

According to Geissler, the potential for AI-based applications in medicine is enormous. "AI will enable things in the coming years that we can't even imagine today."

There is a long way to go before medicines are ready for market

Nevertheless, Professor Zweig believes that in ten years, we will still not be able to cure all diseases. Because it is not possible to say with certainty which of the many proteins trigger a particular disease.

"There are also mutations that have abnormal 3D structures. While statistically they may appear to be the cause of disease symptoms, they are harmless."

And even if it's clear that a specific protein structure leads to a disease, there's a lengthy process before a drug can reach the market. "It has to be tested in clinical trials, for which you need enough patients, and it has to be approved – so in my opinion, it won't happen so quickly."

Where AI is already helping in medicine

In image-based diagnoses based on CT scans, AI can be used to significantly improve the detection of pathological changes, says Florian Geissler. AI can also help with questions about unexpected side effects from a combination of medications, thus leading to optimized treatment methods.

AI could also ease the burden on the healthcare system. "AI systems could, for example, automatically summarize patient consultations and prepare structured reports for health insurance providers. This saves valuable time in the healthcare system. This is where AI will play a crucial role."

Will AI make medicines and treatments cheaper?

However, Professor Zweig points out that, despite AI support, curing diseases always requires large financial resources.

"That's why drugs will continue to be developed only where there are enough patients with enough money to subsequently pay for the drugs."

Why AI fails in most diagnoses

In general, only a few medical diagnoses can be made based on clear rules, Zweig says. One example is diabetes. "There's a very clear threshold and a clear measurement method, and then you've diagnosed diabetes."

Most other diagnoses , however, require a great deal of judgment and experience. "I don't know of any AI systems that can do this reliably enough to replace doctors."

Florian Geissler also assumes that treatment decisions will remain a human domain for the foreseeable future, "primarily for ethical and legal reasons."

And probably also because AI systems are usually still a kind of black box, as Geissler puts it, "where you put something in and then get something out, but you don't fully understand how this decision was made."

Katharina Zweig puts it this way: "We cannot watch the machine learn and make its diagnosis. Therefore, we cannot determine whether it does so according to criteria that we as humans would also apply."