Should we worry AI will create deadly bioweapons? Not yet, but one day

Artificial intelligence promises to transform biology, allowing us to design better drugs, vaccines and even synthetic organisms to say, eat plastic. But some fear that it can also be used for darker purposes, to create biow weapons that are not detected by conventional methods until it is too late. So, how worried we should be?

“The progress of AI feeds the breakthroughs in biology and medicine,” explains Eric Horvitz, scientific director of Microsoft. “With new power is responsible for vigilance.”

His team has published a study aimed at whether AI could conceive of proteins that do the same as proteins known to be dangerous, but are different enough not to be recognized as dangerous. The team did not reveal what proteins they tried to rethink – parts of the study were selected – but it probably included toxins such as ricin, famous in an assassination of 1978, and the botulical, powerful neurotoxin better known as Botox.

To make a lot of protein like botulinum, you need the recipe – DNA that code it. When biologists want a specific piece of DNA, they generally order it with companies specializing in the manufacture of the desired part.

Due to the concerns that potential bioterrorists could order bio-armes manufacturing recipes in this way, certain DNA synthesis companies will voluntarily detect orders to check if someone tries to make something dangerous. Proteins are sequences of amino acids and screening checks whether the amino acid sequence corresponds to “concern sequences”-that is, potential bio-armes.

But with AI, it is in theory possible to design a version of a protein which has a different amino acid sequence but which always does the same thing. Horvitz and his colleagues tried this with 72 potentially dangerous proteins and have shown that screening methods often miss these alternative versions.

It is not as alarming as it may seem. First, the team did not really do the rebuilt proteins, for obvious reasons. But in a distinct study earlier this year, they tested redesigned versions of harmless proteins – and essentially found that they did not work.

Second, although there have been attempts to attack bioterrorists, although very few, there are few reasons to think that this is due to the failure of the voluntary scanning system. There are already many ways to get around it without resorting to an IA overhaul – for example, ricin can be obtained from castor oil plants, found in many gardens. This study is the equivalent of the warning that a bank could be stolen by a little sophisticated Impossible mission-The style plan, while in fact the door of the vault was widely opened.

Finally and above all, when the actors of the State are excluded, no bioterrorist has ever managed to kill anyone using protein -based biow weapons. The worship of Aum Shinrikyo in Japan tried to kill people with botulinum, but only succeeded with chemical agents. The letters stretched by ricin sent to the White House did not kill anyone. Based on the number of bodies, firearms and explosives are extremely dangerous than biotoxins.

So does that mean that we stop worrying about Biow’s weapons designed by AI? Not quite. While Horvitz’s studies only looked at proteins, it is the viruses that represent the great threat – and AI is already used to rethink whole viruses.

Last month, a team from the University of Stanford in California revealed the results of their efforts to rethink a virus that infects the bacteria E. coli. As with redesigned proteins, the results were not impressive – of the 302 viruses designed by the AI ​​that were manufactured, only 16 could infect E. coli. But this is only the beginning.

Asked about the viruses designed by AI, James Diggans to the DNA manufacturing company Twist Bioscience and a member of the Horvitz team, said that it was easier to detect viruses for worrying DNA than worrying proteins. “Summary screening works better on more information rather than less. So, on a genome scale, it’s incredibly informative. ”

But all DNA manufacturing companies do not perform this screening and benchtop DNA synthesizers are available. It is a question of conceiving AI tools that will refuse to create dangerous viruses or to try to detect the malicious intention, but people have found many ways to bypass the guarantees meant, for example, to stop the AI ​​to provide bombs manufacturing instructions.

To be clear, the history suggest that the risk of “wild” viruses is much higher than the risk of bioterrorism. Despite what the current American administration claims, the evidence suggests that Sars-Cov-2 emerged when a BAT virus jumped on other wild animals, then to people from a market-no laboratory involved.

In addition, potential bioterroists could do an incredible amount of damage simply by releasing a known virus, such as smallpox. With the many gaping holes in the efforts to control Biow’s arms, it is not much necessary to use the AI ​​deception to bypass them.

For all these reasons, the risk of an AI virus is unleashed as soon as possible is probably close to zero. But this risk will grow as the various technologies continue to move forward – and the COVVI -19 pandemic has shown how a new virus can create a new virus, even if it is not particularly deadly. More and more, there will be reasons to worry.