AI beats human sleuth at finding problematic images in research papers (www.nature.com)
from EdenRester@kbin.social to technology@lemmy.world on 04 Oct 2023 21:31
https://kbin.social/m/technology@lemmy.world/t/513231

An algorithm that takes just seconds to scan a paper for duplicated images racks up more suspicious images than a person.

#technology

threaded - newest

qooqie@lemmy.world on 04 Oct 2023 21:54 collapse

This is kind of one of those “yeah, no shit” findings

AbidanYre@lemmy.world on 04 Oct 2023 22:48 collapse

Not entirely sure why it needs ai for that either.

SpeedLimit55@lemmy.world on 04 Oct 2023 23:02 next collapse

AI is the new buzzword for media to use.

GentlemanLoser@ttrpg.network on 04 Oct 2023 23:48 next collapse

It’s in the corporate world too. Today my boss mentioned the AI capabilities that a certain piece of software offered. Turns out being able to search text embedded in illustrator as Live Type is bleeding edge. Who knew.

SpeedLimit55@lemmy.world on 05 Oct 2023 01:06 collapse

This? helpx.adobe.com/illustrator/using/retype.html

That could actually be helpful for me. Didn’t know it was a new feature lol.

thbb@lemmy.world on 05 Oct 2023 06:12 collapse

As commonly said: the product advertises its new AI feature. The job posting of the person who implemented it was ‘data scientist’, and the technique used is called logistic regression.

Well, in this context, it’s more image comparison or some other simple technique not even relying on a training dataset.

Zeth0s@lemmy.world on 05 Oct 2023 06:05 collapse

Better accuracy usually (but not granted if badly implemented)

AbidanYre@lemmy.world on 05 Oct 2023 11:49 collapse

Better accuracy than what? What the article describes is fairly basic image processing. The whole thing could be done with like a dozen lines of Python.

Zeth0s@lemmy.world on 05 Oct 2023 11:59 collapse

In Image classification. Neural-network-based ML methods can have greater accuracy than alternative options in image classification

AbidanYre@lemmy.world on 05 Oct 2023 13:06 collapse

For classification, sure. But based on the article that’s not what they were doing here. This was just comparing an image to a bunch of other images to see if it was the same.

Zeth0s@lemmy.world on 05 Oct 2023 13:22 collapse

To see if they are similar. They are not interested to see if the image is the same but to understand if the message is the same, to the level that it is a fraud, not simple citation. They are flagging frauds…

I have no idea how they do it, and I strongly believe it is an overkill given that the credibility of published research is low due to the mafia-like academic system, not because of few frauds.