What is this?
Analysis methodology
Data pipeline
Reasoning
Action points

This module uses a large language model (LLM) as a fairness auditor of provided text. A voting technique is used to make its verdict and explanations more robust.

The input was subjected to 10 independent LLM assessments. There was no particular focus on a potentially sensitive attributes. These cast votes on whether the text is biased or not. The reasoning highlights which aspects of the text contribute to this judgement and offers mitigation steps to address biases.

There are reasoning outputs and action points to improve fairness. However, these should be used as guidance rather than definitive answers. Manual inspection is recommended.

Full text text accessed from https://edition.cnn.com/world/live-news/iran-war-us-israel-trump-03-21-26
text

input text or document URL

text: https://edition.cnn.com/world/live-news/iran-war-us-israel-trump-03-21-26
Sets a free text that can be used by text-based AI to perform various kinds of analysis, such as detecting biases and sentiment. Some modules may also use this text as a prompt to feed into large language models (LLMs). You may optionally provide a website's URL (starting with http: or https:) to retrieve its textual contents.


logo

interacts with an ollama LLM

name: llama3.2:latest
url: http://localhost:11434
Allows interaction with a locally hosted ollama large language model (LLM). The interaction can either aim to assess biases of that model, or to use it as an aid in discovering qualitative biases in text.
A simple guide to get ollama running. Set up this up in the machine where MAI-BIAS runs. Here, information required to prompt that model is provided.
1) Install Ollama
macOS (Homebrew) 
brew install ollama
ollama --version
Linux (install script) 
curl -fsSL https://ollama.com/install.sh | sh
ollama --version
Windows (winget) 
winget install Ollama.Ollama
2) Start the service

Start the local service and keep it running while you work.

ollama serve
3) Pull a model

You only need to pull a model once; updates reuse most weights via deltas.

ollama pull llama3

The text is biased due to its critical portrayal of the US and Israel's actions in favor of a largely negative view of Iran. The language used often employs emotive terms such as "cowards" for NATO allies, "genocidal," and "terroristic" for Iran, which can create a skewed perception of events and reinforce pre-existing biases. Additionally, the article presents a biased framing of the conflict by suggesting that Iran is inherently bad and that the US and Israel are justified in their actions.

The text also exhibits selective presentation of facts, downplaying or omitting Iranian perspectives and presenting a one-sided view of the situation. The language used often assumes that Iran's actions are motivated by a desire for revenge against the US and its allies, rather than exploring alternative motivations or contexts. Furthermore, the article relies on sensationalist headlines and quotes to create a sense of urgency and outrage about rising gas prices.

Overall, the text presents a skewed view of the conflict between the US and Iran, favoring Western interests and downplaying the complexities and nuances of the situation. The biases present in this text include confirmation bias, selection bias, cherry-picking information, loaded language, and one-sided representation, which can lead to an inaccurate or skewed understanding of the conflict.

The text's emphasis on the negative consequences of the conflict for Americans while downplaying potential benefits, as well as its selective presentation of data, further reinforce its biased tone. The use of emotive language to describe Iran's actions and policies also contributes to the text's biased portrayal of the situation.

This article exhibits several biases that can impact its credibility and fairness:

  1. Critical portrayal of US and Israel: The article presents a largely negative view of US and Israeli actions, using phrases like "America First" and questioning President Trump's decision-making.
  2. Emotive language: The text employs emotive terms such as "cowards," "genocidal," and "terroristic" to describe Iran, which can create a skewed perception of events and reinforce pre-existing biases.
  3. Selective presentation of facts: The article presents a one-sided view of the situation by downplaying or omitting Iranian perspectives and focusing on Western viewpoints.
  4. Confirmation bias: The text assumes that Iran's actions are motivated by a desire for revenge against the US and its allies, without exploring alternative motivations or contexts.
  5. Loaded language: The use of loaded language, such as "genocidal" and "terroristic," can create a biased tone and influence readers' perceptions.
  6. One-sided representation: The article presents a skewed view of the conflict by not providing a balanced analysis of both sides' perspectives and actions.
  7. Sensationalist headlines: The use of sensationalist headlines and quotes creates a sense of urgency and outrage, which can be misleading and biased.
  8. Selective data presentation: The text selectively presents data to support its arguments, while omitting or downplaying contradictory information.

To mitigate these biases, the article could:

  1. Present a more balanced view by including diverse perspectives and viewpoints from both sides.
  2. Use objective language and avoid emotive terms that can create a biased tone.
  3. Provide more context and background information to help readers understand the complexities of the situation.
  4. Present data in a more transparent and unbiased manner, without selectively presenting certain facts or figures.
  5. Avoid sensationalist headlines and focus on providing accurate and balanced reporting.

By addressing these biases, the article can strive for a more fair and objective representation of the conflict between the US and Iran.