Hypervisible: “In a new editorial, a group of policy and computer science scholars from Harvard's Ash Center for Democratic Governance and Innovation insist that the already-unreliable system of polling can only be enhanced by asking AI chatbots poll questions instead of humans.”

“In a new editorial, a group of policy and computer science scholars from Harvard's Ash Center for Democratic Governance and Innovation insist that the already-unreliable system of polling can only be enhanced by asking AI chatbots poll questions instead of humans.”

Harvard Scholars Suggest Pollsters Ask Questions to AI Simulations of Voters Because Real People Won't Answer The Phonefuturism.com With just a fraction of people responding to political polls, Harvard experts are suggesting that pollsters "call" AI chatbots instead.

Even if we pretend that what AI polling spits out isn’t bullshit, the training corpus only goes to 2021. That’s kind of a problem if you want to know what people think *today*

I.... What

This has to be a joke.

No. Not Bruce Schneier. 🤦‍♂️

Okay, have people lost all sense of reality here? It is possible to parse enormous amounts of data to determine what aggregate populations think (this was floated as a more accurate version of "general will" by Hiroki Azuma a few years ago), but this is not the same as asking LLMs to do this.

No one wants to work any more...

😂😂😂

The bullshit generator has achieved sentience

Seems like Harvard scholars need to find something else to do instead of making up bullshit that's only going to make bad things worse.

It would not surprise me if it appears to work once or twice, but then diverges — a lot — as LLMs begin to eat their own output. It also would not surprise me if they never work at all.

I recall from formal statistics classes that a representative sample is of critical importance in order for results to have meaningful predictive power and this seems like a fatal handwave around this necessity

1. Assume the population distribution (based on what, polls?) 2. Generate polling "results" from it, 3. Applaud our circular reasoning for its speed and cost.

Call me a loony, but I question the wisdom of taking the systems that tell us that glue belongs on pizza and that women don't have torsos and having them predict election results.

I refuse to accept that this was written by actual human people, who actually seriously thought these thoughts.

Harvard seems like a very silly place these days.

Post