Who is responsible for the actions of an AI (artificial intelligence) agent?

Question

say i have a software that has AI(artificial intelligence) incorporated into it locally, and the user gives the software some permissions. if the software has caused some problems (developed a hack to another person device, deleted data, accessed private information...) all by random algorithmically generated code from the client itself locally. at that scenario, who is responsible?. we can treat this scenario as if a "product" of the software distributor caused damage. or we can treat the AI software as a "worker" for software distributor. the law changes on each of the different perspective of AI like it changes between human workers and products.

this is a kind of hypothetical question now, but i believe that in the emerging future this will become a serious question.

Answer 1

There are some cases in Hong Kong regarding these cases. In Dr Yeung Sau Shing Albert v Google Inc (No. 2) [2015] 1 HKLRD 26, the judge granted application for leave to appeal that:

On Ground 2, while the algorithmic processes were automated, D deployed artificial intelligence to amass information from previous search queries and web content and then generated predictive keywords for the Autocomplete and Related Searches features. Thus, it was arguable that Google Search did deal with content and, unlike mere conduits or passive facilitators, D might be considered as a publisher of injurious content of the predictive suggestions.

So, Artificial Intelligence’s criminal or tort act may turn out to be liable for the Operator. Hope it helps you.

Answer 2

Unless and until an AI becomes a legal person, it is a product/service and normal contractural or tortuous liability rules apply to the manufacturer/distributed/retailer/maintainer etc.

Answer 3

The law does not distinguish between AI and any other type of software. While certain tasks performed by an AI may appear uncannily human-like, this is not of legal significance.

Whenever a product fails and does damage, there can be ambiguity about whether it was the manufacturer or seller or user of the product who is liable. This is as true of a blender as it is of a car as it is of IBM Watson.

Suppose I make a company that builds an armed drone and imbues it with some AI, and the thing decides of its own apparent volition to fly around murdering people. In the eyes of the law, this is not much different then if my company makes cars and I drop a brick on the gas of one of my cars outside of a school. My company, and almost certainly I, would face civil and almost certainly criminal penalties.

So the more autonomy you give your products, the more careful you have to be that they don't mess up and do terrible things. I can't think of any legal principle that would allow blame to be directed at the AI.