Andrew Hines
andrewhines.bsky.social
Andrew Hines
@andrewhines.bsky.social
UCD Computer Science
To finish - I think these apps are of great benefit and help people to parse manifestos easily and promote democracy. But we need to be vigilant that new technologies do not introduce unexpected opportunities to skew opinions through manipulation or poll design. 6/6
November 28, 2024 at 4:48 PM
But, this highlights how a tool to promote democracy could be leveraged. As an aside, this screenshot highlights all the #GE24 candidates responses by colour. The "noisy" patch two thirds of the way down captures the wide ranging opinions of independent candidates. 5/6
November 28, 2024 at 4:48 PM
In theory, with 32 questions, candidates might aim to avoid selecting "agree" or "disagree" to prevent scoring 0 on those questions. With an adjusted scoring: Agree (+1), Neutral (0), and Disagree (-1), no strategy would candidates. We don't believe candidates tried to game the app but... 4/6
November 28, 2024 at 4:48 PM
Or if a candidate, through understanding the scoring system, could even try to game the system? Candidates could optimise their scores based on the proximity rules by maximise neural answers. 3/6
November 28, 2024 at 4:48 PM
We evaluated the candidate responses across all constituencies and computed the average number of neutral responses per party, and this is what we observed. We were curious about the effect of the scoring system and whether it promotes outcomes that reflect voter preferences. 2/6
November 28, 2024 at 4:48 PM
This article to was written almost 2 years ago by @labhaoisenif.bsky.social and me to raise awareness in the legal profession but the risks called out are still as relevant today. Even if you fine-tune and deploy a model in-house it has still “learned” from foundational data out of your control.
www.lawsociety.ie
November 23, 2024 at 2:04 PM