> “these companies could be inserting bias into their algorithms that then affect accessible content.”<p>Even if we ignore intentional tampering, even unintentional edge cases in AI will be dangerous for people in relation to large companies.<p>For instance, if 2B+ people are using Facebook, and Facebook has some trained ML algorithm that is 99.9999% accurate, that is still 200K people that might get incorrectly "categorized" per incident.<p>Now start compounding these 200K people with AI that produce an echo-chamber "oh this person likes X, therefore we should show them Y." Which then eventually becomes "They are looking at Y, therefore they must be like X."<p>These people, even if they try, won't be able to escape the self-reinforcing algorithms. So once they are <i></i>accidentally<i></i> categorized as a "bot" or "fake" or etc. it'll get worse.<p>Now, when these people request human help from call support (if such a thing even exists), before the rep even gets on the call the system will see this "thing" has been flagged and therefore should be pushed back in queue and let better customers be prioritized.<p>These people will be forgotten, tread upon, and never be able to escape their accidental categorization. This is what I call "the lost generation", and they will only grow in numbers over time as the compounding happens, and they will strike back.<p>How do we prevent such systems? I've written a pretty long article on over at <a href="https://hackernoon.com/a-new-kind-of-social-network-emotional-intelligence-e45dcddb1bdb" rel="nofollow">https://hackernoon.com/a-new-kind-of-social-network-emotiona...</a><p>And, we've created tools to build a P2P/decentralized internet, over here: <a href="https://hackernoon.com/so-you-want-to-build-a-p2p-twitter-with-e2e-encryption-f90505b2ff8" rel="nofollow">https://hackernoon.com/so-you-want-to-build-a-p2p-twitter-wi...</a><p>It is as if IPFS + Firebase had a lovechild. :)