As the use of artificial intelligence capabilities actually moves into the real world and business, a number of economic and ethical concerns have been raised. I am doubtful that AI will destroy more jobs than it creates, typically large productivity enhancers, which AI may be, generate stronger economic growth that tends to benefit everyone and create jobs. But AI is built by use of very large data sources and algorithms to interpret that data and both the data and algorithms can be sources of bias. The software underlying the data and models training AI are very complex and all their effects are not well understood. This article finds that the AI software “learns” more than intended, and may “pick up” sensitive behavioral signals and may “inherit” certain properties that are not always beneficial in human interactions. We already are reading stories of chatbots interacting in a manner dangerous to teens and of clear political or other biases in AI output. Artificial intelligence regulation has become very controversial, but we need some guardrails before and AI system does something extremely damaging to humanity.
The article is somewhat complex but worth reading to get a sense of just how little control the developers of these AI capabilities have over what the models are actually learning and what their output will be. Now imagine that applied to health care, for example. (Nature Article)
