MIT cooks up a psycho A.I. named Norman by feeding it dark data from Reddit

Mea Culpa Part IV

If you haven’t already heard, the social network is saying “we’re sorry” yet again after it was revealed that a development bug resulted in posts by about 14 million users to be listed as “public,” even though the user’s account settings were not set to “public.” Facebook says users should check the privacy settings for entries posted between May 18 to May 27th.

Facebook says they noticed the problem on May 22nd and began to revert the posts to the correct settings, but that it took a while, as in five days. The bug reportedly cropped up as developers were working on a new way to share photos and stuff in user profiles. For most of us, a nine-day stretch on Facebook is just a few entries. For others, it’s hundreds. Hit the link for more details and we also have guides to help with privacy settings, data control, and if need be, how to delete your account and information from Facebook.

Looks good, drives better

We love us some Lexus here at Digital Trends, unless it’s the meandering ES line of boat-like limos so popular with…. mature buyers. But, it looks like Lexus has invigorated the ES line with some worthy upgrades, especially when it comes to handling. With cars like the BMW 3-series and the Acura TLX gunning for its audience, Lexus has turned on the charm for the ES, and DT car guy Chris Chin recently took one to the track no less to see what’s new.

The ES line is still solid, quiet, powerful and packed with luxury, including an available 1800-watt 12-speaker Mark Levinson audio system for blasting The Eagles at near concert-level volume. But there’s also new stuff: The ES now offers Alexa integration, hi-resolution LCD instruments, a very sporty 350 F-Sport variant, and a hybrid option that gets 44 miles to the gallon. And we admit: They look pretty sharp. Chin drove last year’s model and the new version at the track and says the difference is significant. Check out his full review.

Don’t let him open a hotel

Well someone had to do it, we’re just kind of relieved it was MIT and not someone with more nefarious goals: they created a psychopathic AI. And they named it Norman, as in Norman Bates. Anyway, the researchers fed Norman The Psycho AI a steady stream of negativity mined from the darker corners of Reddit, and when confronted with a colorful but vague image, Norman said it was the picture of a man killed by a speeding car. Non-sociopathic AI systems said it was a wedding cake on a table.

On the serious side of things, this is not joke research: There is legitimate concern that the behavior of an AI is influenced by what kind of data it “feeds on,” if you will, so the Norman experiment shows that, much like humans, AI’s tend to be quite “negative” if they are surrounded by a lot of negativity. MIT has also created an empathetic AI as well called Deep Empathy, although we think they should rename it “Florence.” Go here for more on MIT’s interesting experiment.

We’ve got more news on our Facebook page and YouTube channel, and be sure to tune in to this week’s DT podcasts: Trends with Benefits (general tech shenanigans)  on Thursdays, and Between the Streams (movie and TV topics) every Friday.