/ Weekly Takeaways

The DIY-ification of Political Ads, Remote controlled humans and How the elderly will suffer in the wake of car automation

Political Ads still face little scrutiny from their ad networks

In a test run by VICE Media, Facebook ads were run with highly offensive racial statements, and in some cases as sponsored by existing political leaders like Mike Pence. Despite Facebook's recent announcements that they'd be looking at all political ads, this demonstrates that we may not see any changes with any future elections.

Key Takeaway

Depending on whether you're good or evil, this one's interesting. Let's start with the fun one first. If you're evil - there's a LOT you can still do on Facebook with minimal input from the ad network. On the flips side, there's a LOT people can still do on Facebook with minimal input from the ad network.

There's a far more negative implication for the ad industry. If we can't trust that ads were in fact created by their true publishers, why trust any ads at all?

MIT Media Lab wants you to control a human this halloween

This Halloween you can log on to https://beeme.online/ and tell a human what to do. Using an actor, MIT has set up a website where audiences can vote on what they want the person to do, like opening doors, moving around, making coffee, etc. This is an annual tradition for MIT Media Lab where they continue to stretch the boundaries of "Bleeding Edge Spookiness" (sorry I had to).

Key Takeaway

This is another great example of the changes we're seeing in story-telling and interactivity. If you think about the popularity of eSports and the dominance of content voting platforms like Reddit, this movement is a natural progression from people wanting to play games, to people wanting to watch others play, and casually directing the outcomes.

When we think about creating things that people will want to spend time with, the ingredients for success have shifted from "let me do something fun" to "let me watch others do something fun".

People prefer that self-driving cars kill elderly over youth

At the advent of the self-driving car discussion (about five years ago) there has always been a moral question of how two cars in an accident would need to decide who lives and who dies. Since then, MIT had set up an experiment called "Moral Machine" to see what people would choose. It has now received enough votes from all around the globe on the moral preferences for self-driving cars. For the most part, everyone agrees that a self-driving car should protect youth over the elderly; however, different regions go even deeper into nuances like a passenger's contributions to society.

Key Takeaway

The results of this aren't surprising, but it brings up a key personal question: Should I not ride around with my parents anymore?

All kidding aside - this may make more people think more carefully about their digital footprint on the world as they think about the potential impact it may have on their future lives. If there's somebody ultimately making these moral decisions, they'll need to use our personal data (like age) to tell the cars what to do. If we expand beyond age, into things like our Uber rating, things get far more interesting. Who knows, maybe you'll start tipping your driver more.

The DIY-ification of Political Ads, Remote controlled humans and How the elderly will suffer in the wake of car automation
Share this

Subscribe to Jon Litwack