Scientists have for the first time ever successfully used artificial intelligence (AI) to redesign functional viruses. This has resulted in previously unknown virus variants that specialize in bacteria, known as phages. Some of the properties of these AI viruses are superior to those of previously known phages. In the experiment, the scientists did not just alter individual DNA segments, but the entire genetic material of the virus was rewritten and redesigned. The experiments raise serious questions about the risks to health and the environment.
Phages are known for using bacteria to reproduce. The infected bacteria die, giving rise to the next generation of viruses, which can then re-infect bacteria. Some phages are therefore thought to be a possible replacement for antibiotics in combating particularly resistant germs. In theory, this could be one of the benefits of the current experiments.
Ultimately, however, the experiments have shown for the first time that AI can be used to design virus forms that do not yet exist in nature and nevertheless still deliver its functions. The AIs were trained with DNA from existing phages in such a way that complex interactions between genes and their regulation were also taken into account. AI could be used to re-synthesize entire biological systems and create novel properties.
The genetic material of phages is relatively small, which is why they have been used for several years in experiments to resynthesize entire genomes. However, these experiments have so far been largely unsuccessful in terms of newly designed viruses. Guided by experts, the AIs ‘Evo 1’ and ‘Evo 2’ have now succeeded in creating phages with biological properties that differ from those of the original forms (ΦX174) but continue to function. Some of the newly designed virus variants can not only reproduce successfully with the help of bacteria, they can also overcome resistance and spread with greater effectiveness.
This poses various risks: on the one hand, phages can continuously evolve, acquire new properties, and, for example, kill bacteria that are essential for life. On the other hand, the approach could also be applied to viruses that spread dangerous diseases. As a precaution, the AI was not trained with data from human pathogenic viruses. However, technically, the approach of these experiments may be transferable to viruses that are dangerous to humans. It is also conceivable that completely new viruses could be created.
In this context, many experts are warning of the dangers of malicious applications or even military use. Humanity now seems to have come very much closer to these dangers. Testbiotech is actively warning against these threats alongside many other experts urgently calling for legal initiatives to contain the risks.
The study has so far only been published as a preprint and still needs to be reviewed by independent experts.
Contact:
Christoph Then, info@testbiotech.org, Tel + 49 151 54638040
Further information:
Publication on AI-designed viruses (preprint)
Position paper of the Nuclear Threat Initiative on AI and biotechnology