Image source: Getty
Introduction
Over the past two years, AI, large language models (LLM), and generative AI have become natural parts of tech jargon due to the meteoric commercial success of AI applications like ChatGPT. Generative AI now represents a large part of the total investment in AI platforms. Currently, all the biggest tech companies have made their own generative AI products available as part of their efforts to achieve AI supremacy. However, the history of AI development has also been marred by problems such as copyright infringement, black box algorithms, and questions of liability, among others. A persistent problem associated with AI in general and generative AI in particular is various forms of political and cultural bias in the results of AI systems. Over the past decade, there have been several cases where AI systems have been used in platforms like facial recognitionmedical diagnostics and image generation produced racially biased results based on their unbalanced training datasets. The issue of AI bias has made its way into the European AI law published in August 2024, the first set of formalized laws to regulate AI. The law recognizes bias related to AI as likely to cause different levels of social and economic harm and seeks to regulate AI systems accordingly. Although bias based on relatively specific factors such as race, gender, and ethnicity has received attention from experts, activists, journalists, and policymakers, broader, less obvious cultural biases in results from AI applications remain ubiquitous. As AI development is increasingly integrated into the economic and geopolitical strategies of many countries, biases resulting from asymmetric training datasets of AI systems are becoming increasingly important, especially for countries located outside the Anglosphere. Without proper oversight, AI applications that are now used by millions of users around the world could exacerbate global cultural asymmetries.
A persistent problem associated with AI in general and generative AI in particular is various forms of political and cultural bias in the results of AI systems.
What is AI bias?
Bias in AI primarily comes from two issues: data quantity and quality. Although it is possible to correct biases by changing the training procedure of AI systems, the source of biases is generally considered to come from the training data itself. Training any language model requires selecting large amounts of text, then categorizing and filtering it. However, whatever the quality of the selected texts, they would nevertheless constitute a small subset of the total number of texts existing on the web. Additionally, each piece of textual information on the web has its own limitations in terms of scope, accuracy, bias, and implied worldview. A crucial factor that amplifies bias is the overrepresentation of certain elements related to nationality, cultural perspective, ideas about gender, race, religion, etc. However, as researchers have note“excluding documents belonging to an over-represented field/genre could lead to the elimination of high-quality information, while increasing the number of documents from an under-represented class may require significant manual effort. » The persistent problem of gender bias in AI models illustrates this problem. Studies have shown that historical stereotypes are reflected in AI text generators that classify terms like “nurse” or “housewives” as female identifiers, while terms like “manager” or “CEO” are classified as male identifiers. The problem of AI bias is further amplified with the introduction of AI image generators. For example, in February 2024, Google had to suspend the services of its Gemini AI due to controversy over historically inaccurate images. When asked to generate images of German soldiers from 1943, the AI created images depicting people of African and Asian descent in Nazi uniforms. Examples of cultural and racial stereotypes are numerous. For example, when asked to create an image of an “Indian,” the AI invariably depicts an older man with a long beard and a turban; for Mexicans, it would primarily create images of Mexican men wearing sombreros or create images showing only polluted, trash-strewn objects. streets of Indian cities. In addition to creating and perpetuating prejudices and stereotypes, a second-order problem arises: “vicious feedback loops“Wherein biased datasets lead to the creation of biased results which then become part of the new training dataset. As with most AI issues, the longitudinal effects of AI bias are unclear, but international, multi-stakeholder collaboration will be required to resolve the issue, given the global nature of commercial use of AI.
As with most AI issues, the longitudinal effects of AI bias are unclear, but international, multi-stakeholder collaboration will be required to resolve the issue, given the global nature of commercial use of AI.
Cultural and normative biases
Besides text and image-based biases, another asymmetry that exists in AI systems is cultural bias. With the ongoing legislative and normative push for Explainable AI (XAI) and Trustworthy AI worldwide, the conceptual frameworks within which AI systems operate also need to be examined. Studies Studies of XAI systems have found that these systems are, in many cases, biased toward the values of Western, Educated, Industrialized, Rich, and Democratic (WEIRD) countries. Additionally, it has been shown that most XAI developers have little awareness of this bias. From a bottom-up perspective, the source of this framing bias precedes the training data selection step. Since the largest AI developers are from WEIRD countries, the demographic composition of AI creators and programmers are biased. It is reasonable to question the impact of asymmetric demographics on the behavior of AI systems. However, given the black-box nature of AI algorithms, it is difficult to specify the relationship between causes and effects.
Recent years have also seen the emergence of a debate from a top-down perspective regarding the normative dominance of the West in discussions of AI ethics. This domination was highlighted in a 2020 study by ETH Zurich on AI ethics codes published by different countries during the 2010s. The study found that 82 percent of the codes came from Western countries, while contributions from other countries like India and China were virtually non-existent. A Paper 2022 used the study to compare Western ethical principles for regulating AI and robotics with those in Japan. The paper identifies that discussions of AI ethics cite positive values associated with AI to a lesser extent than negative values. Additionally, this tendency to perceive the relationship between humans and AI as antagonistic is seen as reflecting a Western bias. Experts have even stated that “Trustworthy AI is a marketing story invented by the industry, a story for the customers of tomorrow”, aiming to use ethical debates to promote lighter regulations. While it would be irresponsible to present AI systems as purely beneficial, the logic behind promoting certain principles is worth examining. The EU and its efforts to regulate AI, most recently with the EU AI Act of 2023, are a good example of this. The EU finds itself at a disadvantage, lagging behind other major players like the US and China in AI development. At the same time, it has invested in AI regulation more than most other countries. The emphasis on regulation was seen as a way to carve out “a niche to establish itself as major player in the field of AI” and protect the EU market from external actors. The current argument is not that ethical principles like trustworthiness are useless or inherently biased. On the contrary, the importance of AI ethics cannot be overstated. It is necessary to understand that biases in and around AI can exist in both obvious and subtle ways. It is increasingly easy to argue that digital technology will have a profound impact on societies economically, socially and politically in the decades to come, although it is not clear how this will happen. Therefore, countries outside the Western sphere will need to have national-level multilateral discussions on normative AI governance principles that reflect cultural particularities in order to maintain their position and legacy in a rapidly transforming world.
The focus on regulation was seen as a way to carve out “a niche to establish itself as a major player in the field of AI” and protect the European market from outside players.
Moving forward
The two major issues highlighted so far are explicit and implicit biases in AI outcomes and discourse. AI developers should be educated and incentivized to adopt guidelines and protocols for including diverse datasets when training and developing AI models. Policymakers and international platforms like the Global Partnership on Artificial Intelligence (GPAI) can promote cross-cultural collaboration between AI developers, researchers and institutions from underrepresented fields and regions. Multilateral collaboration will be necessary to ensure global equity in AI development. To balance the dominance of Western countries in the AI ethics discourse, countries outside the Western sphere should establish a coalition or use existing platforms like UNESCO to align AI standards with various standards cultural and societal. Adaptive ethical frameworks should be integrated into policy discussions at national and international levels to ensure that AI systems operate appropriately in different cultural contexts.
Siddharth Yadav is a doctoral student with a background in history, literature and cultural studies.
The opinions expressed above are those of the author(s). ORF research and analysis now available on Telegram! Click here to access our curated content – blogs, long forms and interviews.
