Dual-use drug discovery AI 創薬AIの光と影

Dual-use drug discovery AI

Generative AI discovered 40,000 molecules in less than six hours as a unique and potent poison. Last April a small lab expressed concern

Created many more lethal molecules than VX.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9544280/

Although not published, the mechanism of action of neurotoxins can be elucidated by examining the history of the development of chemical weapons, from biologically derived poisons such as scorpions and pufferfish to VX gas and Novichok. Even a medical disciple can hypothesize which metabolic pathways should be blocked or enhanced and which molecules should be targeted to be more effective. If we can train a generative AI using this knowledge, we can discover lethal molecules that are easy to generate (and drugable) in a short time. This laboratory, concerned about the implications of this, published a paper in Nature Machine Intelligence in April of the previous year. The dual nature of generative AI, both its benefits and potential dangers, is noteworthy, and its implications extend beyond drug discovery to fields such as food production, infrastructure control, weaponry, financial engineering, and education, all of which require careful consideration.

デュアルユースな創薬AI

生成AIは、ユニークで強力な毒として、6時間足らずで40,000個の分子を発見した。去年の4月小さな研究室が危惧を表明す。

VX以上の致死性分子を多数創造した。 公表されていないが、サソリやフグなど生物由来の毒から、VXガスやノビチョクなどの化学兵器の開発の歴史を紐解けば、神経毒の作用メカニズムとして、どの代謝経路を遮断または亢進させ、どの分子を標的にすればより”効果的”なのかは、一医学徒でもわかる。それを生成AIに機械学習させれば、短時間で生成容易な致死的な分子を発見できるのである。これが意味することをこの研究室は危惧し昨年4月に公表した。生成AIの光と影は注目すべきことで、創薬に限らず、食品、インフラ制御、兵器、金融工学、教育あらゆる面に気を配る必要がある。