- Jan 20, 2013
- 9,452
- 1,040
Stanford researchers used AI (similar tech to ChatGPT, but trained on genetic code instead of text) to design brand new viruses from scratch. Out of about 300 AI-generated designs, 16 actually worked in the lab; they successfully infected and killed bacteria. This is the first time AI has designed a full, working genome, not just edited existing DNA.
Here's the interesting part: some of these AI-designed viruses actually outperformed the natural virus they were modeled on, with a few showing replication rates up to 65x higher, and a mix of them managed to beat bacterial resistance that the natural version couldn't overcome.
The upside: this could help create new treatments for infections that no longer respond to antibiotics. The concern: biosecurity experts are already calling this an "urgent" safety issue, since the same AI approach could theoretically be used to design harmful pathogens too, not just helpful ones.
What if actual disease-causing pathogens could one day be designed the same way, and even outperform natural ones? Should that possibility change how open or restricted this kind of AI research should be?
Here's the interesting part: some of these AI-designed viruses actually outperformed the natural virus they were modeled on, with a few showing replication rates up to 65x higher, and a mix of them managed to beat bacterial resistance that the natural version couldn't overcome.
The upside: this could help create new treatments for infections that no longer respond to antibiotics. The concern: biosecurity experts are already calling this an "urgent" safety issue, since the same AI approach could theoretically be used to design harmful pathogens too, not just helpful ones.
What if actual disease-causing pathogens could one day be designed the same way, and even outperform natural ones? Should that possibility change how open or restricted this kind of AI research should be?
