Media bias continues to be a contentious issue in the current news era, where misinformation has taken center stage. But is there any proof that AI-based news aggregators are biased?
There is existing scientific literature and evidence that provides proof that news aggregator AI’s are biased. In the U.S. six corporations control 90% of all media, and it is well known other AI systems show bias based on the data they are trained with.
This article will explore various topics related to this question in detail. This will include a discussion on bias in AI systems, prevailing opinions about bias in how aggregators work, and evidence to support the notion that these systems are biased. Read on for more.
Biased News Distributed to AI News Aggregators
How do you get unbiased news from biased reporters and sources? If everyone has an agenda and reporters are human, then how is it even possible for a news aggregator algorithm to decipher the bias?
For instance, according to the National Science Foundation, the Apple News editorial staff deliberately choose which “Top Stories” are shown on a day-to-day basis. That is not an Algorithmic Model, it is people controlling the direction of the headlines.
Apple News “Trending Stories” are still algorithmically selected, but unless you work for the company and have proof or access to Apple’s proprietary algorithm, then it is impossible to actually see to what extent there is still human manipulation.
If there are humans involved in the coding of AI Algorithms, then it is unlikely that they can eliminate all bias consciously or unconsciously. In addition, is it possible to aggregate news that includes AI algorithmic bias that is inherently included by design with an opinion?
No Public Audits of Private News Company Algorithms
There are no audits of private news company algorithms. Until Elon Musk bought Twitter, you would have never known any information behind the scenes of what Twitter was doing with their algorithms. Then Elon Musk involved his opinion and put his two cents in to make it even more questionable.
Data science has a problem with Black Box AI as it is and we do not know how many AI artificial intelligence models come up with the answers they give from machine learning news aggregators.
For true algorithmic transparency, you have to see and understand the code to decide if it is being manipulated. If data scientists can’t figure out the outcome from black box AI, then how could they ever figure out bias from a swirling mix of news aggregation?
Artificial Intelligence and a New Generation of News Aggregators
Today, artificial intelligence and machine learning technologies do news interpretation. Large language models (LLMs) and AI chatbots like Open AI’s Chat GPT3 and GT4.
These AIs are moving forward with machine learning (NLP) natural language processing to automate the interpretation of news feeds in different ways.
These NLP models, used as news aggregators, can summarize news articles and are also being used to rewrite news articles from an opposing viewpoint. You, of course, would not know that depending on how you receive the article from the aggregated news source.
In online articles like “Learn how to build your own personal news aggregator service” from API Layer. The public and any hacker have access to simple Open AI Python files for web scraping. These files are used to collect data and forward it to OpenAI just like news source aggregators do so you can create your own aggregator.
The lines are rapidly becoming blurred between (LLM) large language models and their ability to get things wrong and ‘hallucinate’ incorrect facts, and they also can be biased.
The UK’s National Cyber Security Centre states cyber criminals might (will) use LLMs to help with cyber attacks beyond their current capabilities because of the ability to access advanced LLMs with ease.
Google News Aggregator Algorithm System
In the Google News black box AI datafication algorithm system, they use news source and content-related information, user information, supervised, unsupervised, reinforcement machine learning AI algorithms and models, just to name a few components.
How Google News Stories Are Selected
“Computer algorithms determine what shows up in Google News. The algorithms determine which stories, images, and videos show, and in what order. In some cases, people like publishers and Google News teams choose stories.”
How Google News Stories are Selected
Analysis Of The World’s Largest News Aggregator
“There are a few studies on Google News and no systematic and substantive examination of its aggregator algorithms. The decision-making, whether it’s human- or technology-based, behind these processes is not explained; and the relationships between aggregator and sources are not addressed.”
Rutgers-Normalization And Differentiation In Google News:
A Multi-Method Analysis Of The World’s Largest News
Aggregator
Understanding Bias in AI Systems
According to a recent analysis article titled “Google News Shows Strong Political Bias”by AllSides dated February 28, 2023. The analysis did not investigate whether the political leanings or biases displayed by Google News were intentional.
To properly contextualize bias in AI-based news aggregators, it is important first to understand bias in AI systems as a whole. According to Fast Data Science, AI bias is a well-documented phenomenon that affects machine learning tools. This source also explains that this phenomenon is very difficult to get rid of.
AI bias occurs as a result of faulty or incorrect assumptions during the machine learning process. Machine learning essentially occurs when computer systems adapt and learn even in the absence of explicit instructions. The computer system simply analyzes data batches and draws inferences from the emerging patterns.
In AI bias, the correct assumptions in the machine learning process lead to prejudiced results.
AI bias is a problem that has persisted since the inception of machine learning. This is mainly because AI technology, in its design, tends to mimic human decisions. The Harvard Business Review references evidence of bias in AI systems in 1988.
In the 1988 case, the program designed to select applicants for interviews was found to be biased against women and people of non-European origin. While these systems have become much more sophisticated in modern times, the risk of bias still remains.
Causes of AI Bias
The first way bias creeps into these systems is because of the human bias from the people who design or train the AI-based system.
This bias can also be caused by faulty data sets or by using incomplete data during the system training phase. It is important to note that AI systems make decisions based on training data. As such, errors in this data can potentially cause bias to creep in.
AI bias can also be caused by flawed data sampling. This is especially a major problem where certain groups are either overrepresented or underrepresented in the training data. Accordingly, researchers highlight resolving systemic and human biases as the best way to minimize bias in these systems.
The researchers at the National Institute of Standards and Technology note that it is common knowledge that AI can be biased. Does this then mean that News Aggregator AI is also biased? Let us discuss this below.
Bias in News Aggregator AI
To understand the prevailing attitudes on bias in news aggregator AI and ascertain whether there is proof of this bias, let us first evaluate the popular opinions vis-à-vis current scientific literature.
“Journalism is not a public service; it’s a business-Walter Lippmann”
Popular Opinion
News aggregator AI performs an important role in collecting news stories and arranging these stories in an organized manner. This is vital when it comes to news distribution, particularly in this information age, when access to information is so simple.
The consensus in the available literature is that most news outlets are biased in one way or the other. For instance, the Sociable explains that the American media consumer is well aware that news networks such as Fox News and Breitbart are predominantly right-leaning.
In the same way, other news sources, such as CNN and the Washington Post, lean to the left. However, this distinction is not as clear-cut for news aggregator AI. When you consider the fact that news aggregators will generally prioritize certain news articles over others, it is not difficult to see why there are bias concerns.
If you want more context, you can read our article on Google’s use of AI in news recommendations.
Global News Industry Bias
The negative perceptions about potential bias from AI-based news aggregators are not a purely American phenomenon. Authors James Reese and Sara Bannerman explain that algorithmic systems in the global news industry have raised ethical challenges, including algorithmic bias.
The predominant narrative is that these news aggregators filter information in a biased manner. It is also difficult to divorce the claims of bias from the inherent bias of the government and partisan officials who claim the existence of this bias.
But is there any evidence of this bias? Let us discuss this question below.
What the Research Shows
We know there is a bias in Artificial Intelligence based on the data it is trained on or uses from years of research. AI does not create news unless directed, so would only report what it is given in a perfect world.
Because companies like Feedly, Inshorts, API, Google, Apple, Facebook, and Twitter provide news coverage and are profit based, the likelihood of news aggregator bias is a reality.
The International Journal of Technology Law and Practice Computer Law & Security Review 2022.
“There are concerns with the spread of false news in recent years on news aggregator websites. In the case of Google’s news aggregator service Google News, this problem is exacerbated when platform synergies are applied (for instance, users can be directed to Google News through Google’s search engine, Google Search).”
Over the period from 2020 to 2021, multiple lawsuits were filed against Google, including from the US Department of Justice and several states for – among other things – using anticompetitive tactics prohibited under antitrust laws to monopolise the online search and search advertising markets, resulting in harmful effects on competition, advertisers and consumers.137 These lawsuits, followed by others, would lead to more people questioning the influence Google has, by way of its multiple offerings to users including Google News and Google Search, as well as its ability to leverage the Google platform’s synergies. Public scepticism of online platforms such as Google would likely increase in the future.”
The curious case of regulating false news on Google
Springer Nature Humanities and Social Sciences Communications 2022
“AI models learned from user data are inheriting and amplifying some underlying human prejudice, such as the sentiment bias of news reading, which may lead to potential negative societal effects and ethical concerns. Here, substantial evidence shows that AI is manipulating the sentiment orientation of news displayed to users by promoting the presence chance of negative news, even if there is no human interference.”
Removing AI’s sentiment manipulation of personalized news delivery
International Journal on Digital Libraries Volume published in 2018
“No news aggregator focuses on revealing differences between related articles and few systems offer functionality that could be used for this purpose. Thus, users of established news aggregators are subject to media bias.” Most news aggregators ultimately tend to facilitate media bias.
Bias-aware news analysis using matrix-based news aggregation
Wiley Research – Language and Linguistics Compass 2021
“NLP systems reflect biases in the language data used for training them. Many data sets are created from long-established news sources (e.g., Wall Street Journal, Frankfurter Rundschau from the 1980s through the 1990s), a very codified domain predominantly produced by a small, homogeneous sample: typically white, middle-aged, educated, upper-middle-class men (Garimella et al., 2019; Hovy & Søgaard, 2015). However, many syntactic analysis tools (taggers and parsers) are still trained on the newswire data from the 1980s and 1990s. Modern syntactic tools, therefore, expect everyone to speak like journalists from the 1980s. It should come as no surprise that most people today do not: language has evolved since then, and expressions that were ungrammatical then are acceptable today, ‘because internet’ (McCulloch, 2020). NLP is, therefore, unprepared to cope with this demographic variation.
Models trained on these data sets treat language as if it resembles this restricted training data, creating demographic bias. For example, Hovy (2015) and Jørgensen et al. (2015) have shown that this bias leads to significantly decreased performance for people under 35 and ethnic minorities, even in simple NLP tasks like finding verbs and nouns (i.e., part-of-speech tagging). The results are ageist, racist or sexist models that are biased against the respective user groups. This is the issue of selection bias, which is rooted in data.’
Five Sources of Bias In Natural Language Processing
Aggregated News Items on Your Feed Based on Your Preferences
According to a study on bias-resistant social news aggregators by Ziashahabi and his colleagues, the news items on your feed will tend to be ordered based on your preferences and/or votes.
For instance, in social news aggregators, the users are free to post positive or negative votes on the news articles that appear on their feeds. This has the inherent risk that the kind of content that a user consumes will potentially be affected by their dominant search or consumption behavior.
Such patterns are expected in a country like the United States, where the political structures are essentially binary (Republicans and Democrats). However, it is arguable that this is not proof of bias in these aggregators, but the AI adapting to your media consumption patterns to prioritize content based on your viewing behavior.
However, as previously discussed, there is evidence suggesting that the AI is biased, although impossible to identify unless there is access to these algorithms and models.
Key Takeaways:
Bias in AI-based news aggregators are complex and multifaceted, especially considering the inherent bias in the news organizations themselves.
Bias in these systems are impossible to avoid at the present time. However, the researchers note that this technology is still developing and there is a need for further investigation.
Based on existing literature, the claims of bias are based on evidence.
News aggregator AIs also adapt to your personal preferences.
We research and write in our articles like “Is Google Personalized News Using AI To Create Bias?” and “Is Google News The Same For Everyone?”
References:
- 6 Companies Control 90% of What You Read, Watch and Hear. Here’s Why You Should Care.
- Fast Data Science: Bias in AI: How AI Algorithmic Bias Affects Society
- Tech Target: Machine Learning Bias (AI Bias)
- Harvard Business Review: What Do We Do About the Biases in AI?
- NIST: There’s More to AI Bias Than Biased Data, NIST Report Highlights
- AI Multiple: Bias in AI: What It Is, Types, Examples, and 6 Ways to Fix It
- Lexalytics: Bias in AI and Machine Learning: Sources and Solutions
- The Sociable: News App Aggregator Takes Aim at Media Bias
- IBM: Machine Learning
- Google Books: The Algorithmic Distribution of News: Policy Responses
- Predicting News Bias
- The Politics of Personalized News Aggregation
- Normalization and differentiation in Google News: a multi-method analysis of the world’s largest news aggregator
- Automated identification of media bias in news articles: an interdisciplinary literature review | SpringerLink