- Users of social media platforms like Facebook are part of constant marketing experiments
- Because algorithms are driven by AI and machine learning, it’s impossible to know how social media companies are choosing what to show — and not show — different groups of people
- Because there is no “random assignment”, marketers can’t fully tell if one ad might work better than another one.
- In the process, groups of social media users can be excluded from important messages
- Algorithms are so precise, they can target people down to an individual level
I hate this interpretation of black box algorithms.
We know how they work. It isn’t magic. We intentionally built them to be the way that they are.
What we don’t know is precisely which outputs you get from a large combinations of inputs, because it would require memorizing entire databases and simulating results as a weird form of mental math, and how we score things will be impacted by interpersonal biases even if we attempted that.
I mean, still sounds like a black box.
I think the point is that we designed the black boxes to do X and they do X consistently, just with slight variation.
If I make a cake making machine and it consistently makes cakes, its not a magic box just because I’m not sure if it will be creme frosted or not.
I mean, maybe it’s just different vocabulary for both of us?
To me: a blackbox is a thing where input and output comes out in a consistent way, very functional. While the box can make accurate predictions or decisions, the exact reasoning behind them is often unclear.
Looks like we are on the same page, but just talking past each other.
That’s what a black box is, but colloquially, it’s also a way to call something “unknowable” or “magic.”
I thought you were referring to it as the latter, not the former.
It’s a little deeper than that, a lot of advertising works on engagement -based heuristics. Today, most people would call it “AI” but it’s fundamentally just a reinforcement learning network that trains itself constantly on user interactions. It’s difficult-to-impossible to determine why input X is associated with output Y, but we can measure in aggregate how subtle changes propagate across engagement metrics.
It is absolutely truthful to say we don’t know how a modern reinforcement learning network got to the state it’s in today, because transactions on the network usually aren’t journaled, just periodically snapshot for A/B testing.
To be clear, that’s not an excuse for undesirable heuristic behavior. Somebody somewhere made the choice to do this, and they should be liable for the output of their code.
It’s not deeper than “we know how they work” which was the point I was making. I admit I gave am oversimplified layman’s explanation but it is not deeper than that.