There are two major societal risks, as far as I can see.
- the pace of automation going faster than humans can retrain
- the escape of a super-AI with a goal that we don't like
So, you lose a job as a checkout cashier to a self-serve checkout. No problem, you take on the job as a stockboy. A few months later, the stockboy position is replaced by a robotic autoloader system. Well, crap. But, you're clever, so you go and get your high-grade driver's license to be a long-haul truckdriver. It cost you a grand and you needed to cover rent in the meantime, but ehn. That's what lines of credit are for. But then robotic trucks are cheaper and you're laid off. Etc. etc.
On the plus side, the grocery store prices just got cheaper. Long-hauling just got cheaper. IF a new job was invented, you're in pretty good shape. It entirely depends on whether we're retasking those cashiers, stockboys, long-haul drivers faster than they're being replaced.
The 2nd scenario has an AI get out that has a goal we don't like. Some paperclip factory releases a "maximize paperclips" AI. So, it destabilizes the global economy to drive down the price of copper. Maybe a small nuclear war in order to prevent monitoring of nuclear facilities, so that it can buy larger amounts of electricity for cheap. Viola! Hundreds of billions of tonnes of paperclips per year!
OTOH, a good AI could peruse pubmed and invent cures for Alzheimer's that would breeze through safety testing. 10 years later, a looming epidemic doesn't happen and we save billions of dollars per year.
It's a huge risk and a huge reward.
