All the rage is about fake news and algorithms. Supposedly, they're threatening democracy. Is it true? Here's my biased and unproved opinion.
First of all, algorithms are the pillar of modern society. They put several people on the Moon and they basically are the root of everything computers do. Everybody here knows that but it's always good to remind what's what.
As any tool you can use it for whatever purpose you want to give to it. Some use it to save lives by advancing medicine while others use it to sell ads. I'm not judging, it's a free world.
While some of those algorithms are based on rules, others are based on statistics. Especially the ones doing artificial intelligence which is in essence a convoluted way to do statistical regressions. The problem is that statistics are so hard to understand and counter-intuitive that you usually can use them to prove whatever you want.
Take this news article about Google doing CSI-style face unpixelating. It induces that you can re-compose a face from a pixelated image, however that is false: you can simply create a new face statistically inspired by the pixel soup which sometimes happens to look like the real one. So for instance, it would be totally unusable for police work and even less in a trial.
This is probably the most important takeaway from statistics: correlation does not imply causation. In short, AI is an "educated dice roll". If the wind changes then it's lost. You cannot rely on it blindly.
Which brings us to another issue. Humans are built by confrontation. Did you ever see a movie where the main character does something important without some external force forcing his hand? In fact some of the best movies ever made, let's say Blade Runner, are the result of an accumulation of conflicts that have been atrociously terrible for the people doing it.
Because yes, artificial intelligence is, to some extent, a copy of human intelligence. We're also regression engines. But when we face the world, the randomness of what happens to us gives us new ideas. And that's the only way to get them. Otherwise we'd just stay on the same drifting boat.
But what if we put in front of our eyes a filter that is specifically trained to recognize what we already know and only show us that? How does Hugh Grant meet Julia Roberts if she only sees other stars? How can Pretty Woman meet Richard Gere if he only drives cars he knows how to drive? How could Shrek meet Fiona if his swamp wasn't invaded?
That's right, if you put this filter in front of your eyes you end up sad and alone. And that filter is precisely what Facebook and Netflix and Amazon and Google and every-fucking-body in the Silicon Valley are creating, based on artificial intelligence.
What does this give to us?
- Readers will only read about what they already know and endless repetitions of that
- Content producers will only create content that already exists, with less and less randomness each time in order to better reach their audience
In short, sad and alone.
Facebook is not responsible for letting the fake news in. What kind of divinity would they be to decide what is true or not? Rather, they are responsible for cutting out all the perspective that people need but don't want.
So how do you solve fake news? Throw random shit at people. Find out what's good in some circles and throw it in other circles. Make ideas flow around the world not just inside your 2 or 3 closed circles.
Are algorithm making us dumb? The one safely driving your subway probably doesn't but the one blinding you from the truth probably does. It's dumb algorithms that make us dumb. So let's start making clever ones.
Top comments (14)
As the builder of a newsfeed algorithm for dev.to this is definitely something I've thought about. I don't have all the answers, but I promise we're trying to do it right and we'll publish any really good insights we have on the subject.
Well I'm posting this on dev.to and not Facebook right? Thanks guys :)
Actually, as we get out of the "just build anything that won't totally fall apart as we try to scale" phase and into the "slightly more dedicated time and energy to the feed" phase, I feel like we'll definitely come away with some interesting insights and compare/contrast approaches with the Facebook properties.
At the time I feel a bit too scatterbrained to be able to distill our approach/strategy into logical sentences.
How dev.to could topple Facebook
Ben Halpern
There's no timeline for this but it's still on the table IMO 😄
Well I think that the feed does a good job, there is several entry points (chronological, popularity, tags, etc) which bring some pseudo-randomness into what content you are presented and I find that it is good. I've stumbled into a lot of things I would've never been looking for this way!
Yeah, the stumble is key. So much of software development is non-linear and if you aren’t stumbling into new insights along the way you’re not gonna learn much.
Henlo, Benus. Do you haz any counterpoint to my commint there. Pls say yez. Pls have comint.
Wow, I was not expecting such an invigorating read so early in the morning!
I read Blink by Malcom Gladwell a couple years ago, and the bit about humans being regression machines reminded me a bit about some of the same thoughts I had when reading it. It seems like our brains will do all kinds of gymnastics to setup short-cuts around critical thinking.
following
I gave you a heart. Related read: Hello World: Being Human in the Age of Algorithms by Hannah Fry
The problem with the democracy discussion stems from what the algorithm is being fed.
If it is being fed one sided data (which it appears to be), the output will be one sided.
That's the problem, not artifical intelligence itself. The people building the AI tool.
On this point I totally agree that we have to train our brain. However, when you're delegating that's an opportunity to do something else instead, possibly more interesting and more challenging. Society gives us a pretty abstraction of the world which allows us to concentrate on our jobs rather than worrying about getting food every single day all day long. And I think that algorithm do just that.
Thanks, I'm not the only one :D. Long time ago I was writing final project for ma bachelor degree on context-aware recommendation engines. I wanted to do algorithm in a way that presents mainly stuff that's not in list of users similar to you. I decided on Jaccards index as it's the simple thing that made sense to me. I didn't have huge DB with users to test so I gave up and used user-to-user style where context finds only new episodes from tv shows that are still on air relative to current date. So each day you get new popular shows and some recoomended ones from similar users. This ignores item to item where you would be stereotyped to a certain genre and user to user where you would always get similar series to ones your watching which is boring in a way as most of them would have similar stories. My algorithm while testing actually gave me Blindspot which I was watching later on and liked.
I think main problem is that people are unwillingly stereotyping others based on what the "mostly" do. So for TV Shows and me who likes some Sci-Fi and computers a lot of algorithm suggest Start Trek and similar which I do not watch nor like. Although Netflix did gave me Dark (german series) mostly it gives me too much stuff I already watched and don't have time to filter out by clicking like or dislike. So maybe it would be good to create something based on "contra-algortihms" where recommendation throws out to you only stuff that you have little to no relation. This would be good for when you want to find new stuff but also bad as it basically filters out nothing but hides the stuff you are statistically gonna like.
The problem is deeper than the AI and ML it more like stereotyping and grouping but then again it sells so who am I to judge.
Agreed, our own biased filters and services that only serve us similar things is the base problem.
It’s why we don’t have children with our cousins, we need diversity to thrive.
Diversity usually puts us outside of our comfort zones and takes mental energy to consume. A barrage of kitten photos does not.
So I can understand where the issue comes from, but not how to solve it unfortunately. :/
"correlation does not imply causation" or would that be better "correlation does not imply causality."
🧐
From what I understand, "causality" and "causation" mean basically the same thing but the later is the one used by statisticians. Although I'm French and my math background is entirely in French I wouldn't be completely assertive on that. Also, XKCD says "causation".