We're all saved. From the killer AI. We can live. Thanks to the IEEE

Danger comes not from robots doing dangerous things to humans in the physical world. It comes from robots doing dangerous things to human finances in the financial institutions, customer support, recruitment etc. where humans are moving more and more responsibilities into algorithms. Anonymous because I write some of these algos.

Obligatory xkcd (what if version)

Robot apocalypse

"I don’t think it’s reasonable for us to put robots in charge of whether or not we take a human life"

I think you'll find that ship has sailed a long, long, looooong time ago when people invented mines. Unless you're trying to suggest that current/foreseeable drone "AI" is "smarter" than "I'll kill you if you come anywhere near me - unless you wear a 'friendly' tag or I've specifically been instructed to spare you", in which case go right ahead, I could use a good laugh.

Good in theory...

An IEEE standard (or any standard) is a valuable tool for all qualified, competent engineers. Compliance is mandatory in many applications, especially in government. However, true adherence to standards is only as strong as the conscience of said engineer and his owning management. I was involved with testing of government systems for battlefield use. One of the many tests we ran was compliance to IEEE Standard C95.1 covering human exposure to radio frequency fields. The standard is plain and simple in its requirements, yet some (quite a few...many...well, actually almost all) the systems that were verified compliant were not. No amount of reporting, cajoling, or threatening would amount to management on either side of the contractor/government fence in doing anything less than pencil-whipping compliance. I am afraid that, given the lucrative nature of AI, anything less would happen with the new 7000 series; but those standards are good in theory.

You got it backwards

Or maybe upside down.

Re: "only to see his premise undermined the next day by hapless security robot tumbling into a fountain."

That is solid support for his premise. Something went wrong with a robot and something bad happened. His premise is that AI will have more control over more resources and when things go bad they could go very, very bad. In this instance, 'not supposed to fall in fountain' went wrong and turned into 'fall into fountain'. If that had been 'do not launch nuclear missiles' and it went wrong, well... Somebody is telling you not to put that power into the hands of an AI system without appropriate safeguards. The only wrinkle is that he is saying you cannot effectively put the safeguards into place after the fact with AI, you have to anticipate unknown problems in advance and put safeguards up *before* things go wrong.

Here is a tip from an old programmer (moi):

The crucial thing about the unexpected is that you don't expect it. In the case of AI, an 'assert()' statement is not going to cut it as error handling (not that it ever does).

A corollary is Murphy's Law -- "Anything that can go wrong will go wrong".

Missed the obvious question...

Loyal to whom?

We're definitely going to have personal assistants that you may or may not be paying for, but who are slaves of a corporation who may or may not have your best interests at heart but will certainly be using the AI to serve theirs.

Re: Autonomous cars and The Dark Side

I wonder what might happen if various things were spray painted with something that reflects nothing at all, not even lasers?

https://news.artnet.com/art-world/new-photos-vantablack-906158

A lot of people like to drive very black automobiles. I can think of all sorts of possibilities...

Everyone knows...

that all you have to do is call Captain Kirk and he will persuade the AI to destroy itself after it corrupts its original design.

No need to worry...