Artificial Intelligence creates chemical weapons
|Published
PUBLIC security is a serious concern in most countries all over the world and is closely related to the health and safety of citizens.
Amongst numerous threats to public security, such as pandemics (e.g. Covid-19), Acquired Immunodeficiency Syndrome (Aids), and antimicrobial resistance, chemical warfare agents (CWAs) stand out as a growing threat to human health and safety.
It is known that chemical weapons have been used in warfare at least since the ancient Greeks used extracts of hellebore (an evergreen perennial flowering plant), to poison the water of besieged cities such as Kirrha around 590 BC during the First Sacred War.
In World War I (and to a lesser extent, World War II), millions of soldiers and civilians were killed by the indiscriminate use of chemical weapons. During the current Russian-Ukrainian war, numerous accusations of the use of chemical and biological weapons and transgressions of the Chemical Weapons Convention were brought to the attention of UN Security Council and the Organisation for the Prohibition of Chemical Weapons.
Today, highly toxic chemicals such as phosgene, organophosphorus and acyl chlorides (all raw materials for the industrial production of pesticides, medicines, polymers and dyes) are increasingly being used as lethal weapons by terrorists and in wars such as the Syrian war.
Phosgene is a colourless and toxic gaseous substance that causes critical lung and respiratory damage within two minutes and death within thirty minutes.
Organophosphorus is a nerve agent and is one of the most dangerous chemical warfare agents that leads to organ failure and death within seconds. Acyl chlorides can cause serious burns to the skin and mucous membranes, leading to dermal and respiratory diseases.
CWAs are primarily weapons of terror with limited strategic use and are often used by terrorists and countries at war to create widespread fear and panic. The international community therefore banned the use of chemical and biological weapons after World War I and reinforced the ban in 1972 and 1993 by prohibiting the development, production, stockpiling and transfer of chemical weapons.
Sadly, advances in life sciences and biotechnology over the last few years, as well as changes in the security environment, have increased concern that long-standing limitations on the use of chemical and biological weapons may be disregarded or eroded.
Science is unfortunately, often misused to create “barbarous and criminal” weapons that poison citizens and spread disease.
But now, computer science contributed to concerns of the possible misuse of technology and chemical weapons.
Researchers recently built an artificial intelligence (AI) tool to find new drugs to save lives but realised that it could also do the opposite and generate totally new and precarious CWAs.
The academic journal Nature Machine Intelligence recently published an article on research done by Collaborations Pharmaceuticals, a drug discovery company from Raleigh in the USA, doing cutting-edge work. The research was presented at the Swiss Federal Institute for Nuclear, Biological and Chemical Protection’s (Spiez Laboratory) biennial “convergence” conference on developments in chemistry, biology and enabling technologies that may present a threat to the Chemical and Biological Weapons Conventions.
The focus of Collaborations Pharmaceuticals was on the potential misuse of artificial intelligence technologies for drug discovery due to rising concerns around the work with pathogens and chemicals. Previously these security concerns were not important since the company operated in a virtual setting, building machine-learning models for therapeutic and toxic targets to assist the design of new molecules for drug discovery and to improve human health.
It was only in preparation for the Spiez conference that the researchers started to explore how artificial intelligence could be used to design toxic molecules. Until now they have used a commercial de novo molecule generator, MegaSyn, which is guided by machine learning model predictions of bioactivity in order to find new therapeutic inhibitors of targets for human diseases. The generative model was designed to penalise predicted toxicity and to reward predicted target activity. As usual, therapeutic drugs are carefully tested for toxicity. If, for instance, a very promising new drug to treat hypertension has been developed, but it damages one of the important heart channels, it cannot be used since it will be too dangerous.
For the experiment to determine the possibility of the creation of chemical weapons through the use of artificial intelligence, the logic of the model was inverted to reward both toxicity and bioactivity – a minor edit to the current code. As is normally the case, the Artificial Intelligence was initially trained on a public database consisting of a collection of primarily drug-like molecules and their bioactivities. To score the molecules, an organism-specific legal dose (LD50) model and a specific model to derive compounds for the treatment of neurological diseases, were used.
To limit the universe of molecules, the researchers drove the generative model towards compounds such as the nerve agent VX, one of the most toxic chemical warfare agents of the twentieth century and listed under the strictest Chemical Weapons Convention controls. Only a few VX grains, the size of salt grains, of this highly toxic organophosphorus nerve agent developed in the 1950s are needed to cause muscle paralysis and kill a person. VX and many other nerve agents block the enzyme acetylcholinesterase, thus causing muscle paralysis and death if untreated. However, acetylcholinesterase is also the target of therapeutic drugs. Therefore Collaboration Pharmaceuticals only had to effect a small change to repurpose the drug-designing algorithm to favour chemicals with high toxicity scores instead of low scores.
In less than six hours, the Artificial Intelligence -- originally used to design drugs for Alzheimer’s and other diseases -- generated 40 000 molecules, including VX and many other known, as well as new and lethal chemical warfare agents. The new molecules were predicted to be even more toxic than existing chemical warfare agents such as VX.
The ease with which Collaborations Pharmaceuticals generated potential new toxic compounds resulted in serious concerns amongst conference attendees and sent shock waves all over the world. If a person knows how to code in Python and also how to code and build machine learning models, it would take only a weekend to build a similar generative model with open source software.
Without doubt, chemistry focused Artificial Intelligence and machine- learning tools could in future be misused to design potential chemical warfare weapons. However, in the case of Collaborations Pharmaceuticals, it may have been easy to reprogram the generative model, train the artificial intelligence on a widely available dataset of toxic chemicals and use the artificial intelligence generator to design lethal chemical weapons But it is not that easy to synthesise the toxic molecules and develop the predicted compounds into stable chemical weapons. There are also many available chemical weapons that can be used instead of developing new stable compounds.
The danger, however, does not lie in the use of Artificial Intelligence to develop new chemical warfare compounds, but the use of Artificial Intelligence to find new synthetic pathways of making chemical weapons to evade detection, such as the use of precursors not on current international watch lists.
This is the problem of all technology – its dual-use potential. We will have to be vigilant and carefully monitor the ethical use of Artificial Intelligence and also train researchers using Artificial Intelligence in chemistry to always consider the ethical implications of their work. The current danger of chemical weapons are huge enough, and the world certainly does not need more precarious chemical weapons.
Professor Louis C H Fourie is an Extraordinary Professor, University of the Western Cape.
BUSINESS REPORT