Artificial intelligences have been around for quite a while now. Right now, they’re only useful in very narrow contexts, like beating humans at chess and Go, writing absurdist movies (starring The Hoff), or recommending stuff you might want to buy, watch, or read next.

More recently, AI has been used to improve photos that smartphones take, giving them seemingly impossible abilities like seeing in the dark. But outside those very specific fields, the same AIs are pretty much helpless, and cannot deal with new situations they were not programmed and trained for, or deal with very complex situations that humans can handle fairly easily.

So-called general purpose AIs, one that can learn to deal with new situations without human supervision and intervention, are then the holy grail of AI research. Such an AI could eventually rival a human’s mind capability, and maybe even surpass it. What happens then is hard to tell, since, in theory, such an AI could immediately start improving itself in ways we would not have forseen, since it is by definition it has surpassed our human mind’s capabilities.

Such scenarios, then, have been the subject of many sci-fi books and movies, ranging from benevolent overlords watching over and shepherding humanity (e.g. Ian M. Bank’s Culture novels) to ones that start eradicating humanity almost the moment they’re turned on (the Terminator movies). This has prompted several prominent researches and thinkers to call for limiting or even banning military AI research, due to fears that we don’t know what AIs are going to do if given weapons.

But our present does not contain such threats, even though AIs have gotten extremely inventive with what you could call creative solutions to the problems they were trained for, due to (others might just say they’re cheating). As such, researches have also called out the danger of even these non-general AIs, since they rely on humans to program them and set the constraints, and, well, we humans sure ain’t perfect. The canonical example is a “paperclip maximizer” AI, that would eventually consume the entire earth in its quest to produce as many paperclips as possible (also available as a browser game, where you play the part of that all-consuming AI).

But there’s also an argument that we don’t have to wait for general-purpose AI. Rather, even today’s special-purpose AIs are already creating more than enough problems that we humans have to deal with, and that we have already created the “paperclip maximizer” AI, but instead of paperclips, it’s trained to maximize “engagement”, and thus optimizes social media for maximum addiction and despair.

So while we would be prudent to watch out for the possible future human-enslaving overlords, we should also keep out an eye for our present-day recommendation and optimization buddies, and the people who are controlling and training them.

Delicious Treats

You know that whenever you order something at McDonald’s, it’s not going to look nearly as delicious as the high-gloss poster showing off the product. So here are some photography tricks that advertisers use to make food look delicious (YouTube) (via kottke.org)

Hearts, Beating Again

If there’s one medical procedure everyone knows, it’s probably CPR, or Cardiopulminary Resuscitation. It’s a staple of any medical drama on TV, and has the seemingly magic ability to bring people back from the brink of death. In reality, it’s not that easy: What Doctors Know About CPR

Earth’s Beating Heart

Three times a day, an extremely low frequency electromagnetic pulse races around the Earth. These pulses vary with seasons, solar activity, or even lighting storms: Schumann Resonances

📖 Weekly Longread 📚

Good magicians not only perform magic tricks, they make them seem effortless. And few were as good with card tricks and sleigh of hand as Ricky Jay. Give him a deck of playing cards, and he would do the impossible: Ricky Jay’s Magical Secrets

🦄 Unicorn Chaser 🦄

Astronomy Picture of the Day: 2018 December 4 - Rocket Launch between Mountains