AI can just as easily create toxins as new therapies.
A small pharmaceuticals company has generated the chemical formulae for thousands of existing and new toxins using machine learning techniques. They did the work with the best of intentions. It was presented at a conference designed to funnel discussion topics on to the agendas of Chemical and Biological Weapons Conventions. The company in question simply inverted the rewards in the algorithm to generate formulae to promote toxicity rather than therapy. The algorithm generated some 40,000 such formulae in around six hours. We have all become used to the exciting idea that machine learning can and will promote therapeutic drug discovery. As war currently rages in central Europe, this is a timely wake-up call pointing out that the same technology can generate harmful outputs too.
Link to article: https://www.nature.com/articles/s42256-022-00465-9
You may also like to browse other AI papers: https://www.thesentientrobot.com/category/ai/ai-academic-papers/