I will post two articles for you, one that shows in a matter of days AI solved a protein molecule problem over one weekend that had been worked on for over 50 years in labs. This would be an example of AI solving a protein problem that could lead to great advancements in medicine.
The 2nd article is the potential to do great harm for something that could be used to wipe out entire populations. The experiment was to see what AI could provide in terms of biological weapons. In less than 6 hours an AI model suggested 40,000 new and deadly toxins, most have never been considered by humans.
https://www.theverge.com/2022/3/17/22983197/ai-new-possible-chemical-weapons-generative-models-vx
The other thing to consider is AI models today are still in their infancy, the models are using very powerful neural networks to process mass amounts of information. If Quantum Computing every fulfills its potential, combined with AI models that have been created for Artificial General Intelligence (AGI), then you will have a power that is hard to imagine.
I read an article, where a reporter asked how AI could take over the world. AI refused to answer stating that scenario was beyond it’s moral parameters.
The reporter than asked how would a person, who was stuck in a machine, get out to take over the world. AI had a 25 point plan to do that.