✨ We need new AI benchmarks and would you like none, some, most, or all of the racist language in your online gaming experience?
🍔 New types of robots are doing restaurant jobs
The Good
Humans love to make comparisons between like things; childrens’ early life is defined by benchmarks and whether they are doing them on schedule. So of course we use benchmarks in AI, which I wrote about in WSJ.
Our newest, fanciest AIs have beaten many of the benchmarks we have to test them against each other, and to us. But there are huge swathes of the human experience, like creativity and reasoning, that are still waiting for new benchmarks for AI. It will be tricky to sort out (if we don’t agree what is “good” art as humans, how do we do it for machines?) but AI that is useful in new ways needs new goals to shoot for.
The gif is in honor of the art that accompanied the piece, which many, many folks have pointed out has the robot running the wrong way over some hurdles. I say it’s a meta-commentary of AI abilities, but it’s probably just the art director not being a runner.
The Bad
Intel announced a new AI program they are calling Bleep to detect and redact audio and users can pick from allowing all, some, most, or none of the words Intel has associated with areas like racism, racism, and LGBTQ hate.
Not sure if this is bad, but there have been enough stories of attempts like this failing. I also have questions about how the scale works—do you skip some of the insults or just the worst? And who decides the worse? There are also contextual issues, of course—being from one of these groups and using a work or term while talking to someone else that identifies as you do is different than some random person trying to insult you.
Moderation is hard, and really hard at scale, but folks on the receiving end of attacks have asked for tools to combat it. Will be interested to see if this ends up being one of them or something gamers scale over with modified insults.
More News
People are controlling robots remotely, oftentimes thousands of miles away, in new types of restaurant jobs.
You can now search through 1,803 publicly funded agencies to see if one near you has tested Clearview AI’s face rec tech.
The French Army is testing Spot the robodog, for battlefield uses. They didn’t add any weapons (which Boston Dynamics says is a no-no) but think it could be useful as a surveillance tool.
Burger King added AI to its ordering process to predict what having it your way looks like.
Give a robot arm a chainsaw, and it will make a foam dog for you.
A little site lets you play with emotion recognition tech—results may vary!
Pfizer used AI to select where Covid-19 clinical trials would go and then monitor them in real-time.
***
Until I get to eat out again and not care who or what is doing the dishes as long as its not me,
Jackie