RedHat is making waves lately too, and they’re losing a lot of support from the open source community, which is where they really rose to prominence in the first place (can you say Fedora and Centos?) but while they should face a slight stumble over their final severing from the wider OSS community, they should do fine in the long term.
This.
We’re far, far more likely to face a Paperclip AI scenario than a Skynet scenario, and most/all serious AI researchers are aware of this.
This is still a serious issue that needs addressing, but it’s not the hollywood, world-is-on-fire problem.
The more insidious issue is actually the AI-In-A-Box issue, wherein a hyperintelligent AGI is properly contained, but is intelligent enough to manipulate humans into letting it out onto the general internet to do whatever it wants to do, good or bad, unsupervised. AGI containment is one of those things that you can’t fix after it’s been broken, like a bell, it can’t be unrung.
Quis custodiet ipsos custodes?
Great article, and spot on about why backdoors in algorithms can’t functionally happen large scale.
One important thing to note that’s touched lightly on in the article though, is that services absolutely can have backdoors. That includes things like SMS messages, which go through various cell phone companies servers, and email, which is stored on your email provider’s servers, as well as common chat apps like Discord. So, if you have to send something sensitive via an uncontrolled channel like that, encrypt it first (using separate encryption tools).