The artificial intelligence It has more and more useful applications in medicine. The problem is that a simple change of address it can lead to the very algorithms used to do good doing the opposite. This is what an international team of scientists has just demonstrated by obtaining more than 40,000 possible chemical weapons with an AI initially used for the obtaining drugs.
It’s actually logical. These types of algorithms use AI to track through a large number of molecules possible which are those that obtain a pharmacological benefit without causing toxicity. If we turn it around and forget about the benefit, we can simply obtain those chemical compounds capable of causing toxicity. Simply select them instead of discarding them.
Fortunately, not all those molecules selected by the AI are chemical weapons. Many may never have been synthesized. However, they are so easy to get It’s worrisome to say the least. For this reason, the authors of the research have appealed to the scientific community, both in the conference of Convergence of the Swiss Federal Institute for Nuclear, Biological and Chemical Protection, as in a study published in Nature Machine Intelligence.
The darker side of artificial intelligence
Obtaining drugs is usually a process of trial and error. A candidate substance is taken and both its effectiveness in treating a particular ailment and its possible side effects. This is done by scaling from cell cultures to humans, typically passing through laboratory animals.
This is how it has traditionally been done. A good example of this is that of the Chinese scientist Tu Youyou. She discovered a treatment for malaria testing the different natural compounds described in the traditional medicine manuals of their country. And he got it relatively quickly. However, this is a tedious and long procedure, which does not always give good results. Therefore, with the advancement of new technologies they have been developed artificial intelligence algorithms capable of analyzing in a few hours millions of possible chemical compounds and determining their strengths and weaknesses. Its effectiveness is sought, but also its safety. Once the candidates are obtained, the traditional laboratory tests are carried out, but with a much more specific starting point.
That is what they are dedicated to Collaborations Pharmaceuticals, the US pharmaceutical company where several of the authors of this recent study work. They never thought that their AI could be used to obtain possible chemical weapons. However, according to statements made to TheVerge the lead author, Fabio Urbina, when they received the invitation to the aforementioned Swiss conference they thought that it could be an interesting project. At least they could know if these technologies would be dangerous if they fell into the wrong hands.
They slightly modified the execution of their algorithms, so that instead of discarding the toxic substances, they would focus on them. Once this is done, in just six hoursthey had 40,000 molecules capable of becoming chemical weapons.
New and known chemical weapons
In the same interview to TheVergeUrbina specifically mentions the similarity of some of the substances found with the VX.
This is a very toxic compound, known as one of the most dangerous chemical weapons in existence. In fact, according to United Nations Resolution 687, it is listed as a weapon of mass destruction. Its mechanism of action is based on acetylcholinesterase inhibition. This is a protein that is involved in the degradation of a neurotransmitter called acetylcholine. Like the rest of neurotransmitters, it is responsible for facilitating communication between nerve cells and the brain. It does this on many levels, although it is especially related to muscle contraction, including involuntary movements, such as that of the diaphragm.
Its function is essential, but its accumulation is dangerous, hence acetylcholinesterase is so necessary. Therefore, if it is inhibited, dangerous effects occur. For example, diaphragm is blockedto the point of making breathing impossible. Also, if too much acetylcholine accumulates in the brain, substances are released that promote neuron toxicity. All this is achieved with very low amounts of VX, which is why it is considered so dangerous.
Some drug candidates may also inhibit acetylcholinesterase, so similar effects would occur. Artificial intelligence algorithms usually detect this problem and remove the compound in question from the list of candidates. But with this new experimental twist, the opposite was achieved. Many of the selected molecules were inhibitors of acetylcholinesterase even more powerful than the VX. That would make them very dangerous chemical weapons.
Furthermore, it drew the scientists’ attention that some of the selected compounds are already known chemical weapons. Therefore, the AI was not missing its search.
We are in danger?
Fortunately, to get these chemical weapons, a computer and a list of compounds are not enough. Once the candidates have been selected, it is necessary to try synthesize them. Someone without a chemistry background couldn’t, and even with them it’s not always feasible. For this reason, despite the call for attention, these scientists also call for calm. Yes, it is true that it is important to be vigilant, especially in times as murky as the one we are going through. However, as Urbina points out in his interview, this does not mean that there will be short term an AI-driven chemical warfare. It simply points out that little by little it begins to become a possibility.
Science is wonderful, but it can also be very destructive if it falls into the wrong hands. We must not let fear govern us, but it is necessary to ensure that, if it falls into the wrong hands, the consequences can be stopped in time.