I think he's describing the perfect digital news platform that I would think of.
His idea reminds me of a hackathon project I worked on a few years ago with some friends. We built an app that communicates its location to the backend. Our backend runs k-means clustering and decision forests to cluster tweets coming in realtime through the Firehose API to associate them to an event that's happening nearby. The app then uses the incoming stream of tweets to provide realtime coverage on an event that was algorithmically detected.
It seems obvious - a lot of "news reports" nowadays are pretty much a couple lines of text with a bunch of tweets dropped down beneath them. But making that scalable is actually pretty difficult.
The first difficulty (and we've confirmed this and talked about the problem multiple times with directors of product at Twitter) is finding what the "best" stuff is.
We tried some time ago doing that for news - can we just algorithmically find the best tweets with regard to topic x? Even given full API access, the answer is either "not really" or "yes, but austenallred and his team aren't good enough programmers to figure it out."
There simply aren't enough inputs - it's the same reason Google has always struggled at real-time news. There are favorites and retweets, but those are largely a function of how big someone's following is. Some guy walking on the street taking amazing photos will never have as many retweets as CNN, even if you try to control for the number of followers, or measure retweets as a function of the number of followers somehow.
Then even if you could get that, the tweets aren't always analyzed in the right context. Run that type of analysis and the top tweet is likely to be a joke tweet or something like "OMG I'm praying for Boston."
The best solution that we can come up with is crowdsourcing the finding and curation of tweets. It's working OK, but there's a long way to go. Basically we let everyone find and submit tweets, tag them with specific topics, and then upvote/downvote and fact-check for accuracy. (for details on how it works, see https://grasswire.com/about)
Once you add a couple lines of text, you can turn the raw firehose of Twitter into something that can be consumed by somebody like my mom. That, I think, it a really big deal. It's working fairly well for news right now, and as more people join and submit it gets better. Our hopes are that we can expand into different verticals, and create daily digests of the best stuff with regard to topic x that you can consume in 30 seconds.
We basically see this as the back-end that could power some really cool stuff (apps, partnerships with publishers)
He's right.
The insistence of tech companies on algorithms is understandable, but it's also incredibly limiting. I think of it as the "Google mindset" (it hurts them too).
In the end, you need live people involved. You can do SO much more with even just a few live curators. Looking from the outside, insisting that they aren't needed is deeply weird.
The problem with such a platform would be how to know if someone in a riot (for ex.) is posting 'news' or not. This seems to be where channels would go a long way to sorting through the hay to find the needle.
Edit: unsure of the downvotes - I feel like this is relevant info to the parent comment.
He tried this at Uber too. Kalinick shut that down super quick, and I think he is banned from the premises.