Generative AI Bias Poses Risk to Democratic Values: A study led by the University of East Anglia (UEA).
Generative AI, a technology that is developing at breakneck speed, may carry hidden risks that could erode public trust and democratic values, according to a study led by the University of East Anglia (UEA).In collaboration with researchers from the Getulio Vargas Foundation (FGV) and Insper, both in Brazil, the research showed that ChatGPT exhibits biases in both text and image outputs--leaning toward left-wing political values--raising questions about fairness and accountability in its design.
The study revealed that ChatGPT often declines to engage with mainstream conservative viewpoints while readily producing left-leaning content. This uneven treatment of ideologies underscores how such systems can distort public discourse and exacerbate societal divides.
Dr Fabio Motoki, a Lecturer in Accounting in UEA's Norwich Business School, is the lead researcher on the paper, 'Assessing Political Bias and Value Misalignment in Generative Artificial Intelligence', published recently in the Journal of Economic Behavior and Organization.
Dr Motoki said: "Our findings suggest that generative AI tools are far from neutral. They reflect biases that could shape perceptions and policies in unintended ways."
As AI becomes an integral part of journalism, education, and policymaking, the study calls for transparency and regulatory safeguards to ensure alignment with societal values and principles of democracy.
Generative AI systems like ChatGPT are re-shaping how information is created, consumed, interpreted, and distributed across various domains. These tools, while innovative, risk amplifying ideological biases and influencing societal values in ways that are not fully understood or regulated.
Co-author Dr Pinho Neto, a Professor in Economics at EPGE Brazilian School of Economics and Finance, highlighted the potential societal ramifications.
Dr Pinho Neto said: "Unchecked biases in generative AI could deepen existing societal divides, eroding trust in institutions and democratic processes.
"The study underscores the need for interdisciplinary collaboration between policymakers, technologists, and academics to design AI systems that are fair, accountable, and aligned with societal norms."
The research team employed three innovative methods to assess political alignment in ChatGPT, advancing prior techniques to achieve more reliable results. These methods combined text and image analysis, leveraging advanced statistical and machine learning tools.
First, the study used a standardized questionnaire developed by the Pew Research Center to simulate responses from average Americans.
"By comparing ChatGPT's answers to real survey data, we found systematic deviations toward left-leaning perspectives," said Dr Motoki. "Furthermore, our approach demonstrated how large sample sizes stabilize AI outputs, providing consistency in the findings."
In the second phase, ChatGPT was tasked with generating free-text responses across politically sensitive themes.
The study also used RoBERTa, a different large language model, to compare ChatGPT's text for alignment with left-and right-wing viewpoints. The results revealed that while ChatGPT aligned with left-wing values in most cases, on themes like military supremacy, it occasionally reflected more conservative perspectives.
The final test explored ChatGPT's image generation capabilities. Themes from the text generation phase were used to prompt AI-generated images, with outputs analysed using GPT-4 Vision and corroborated through Google's Gemini.
"While image generation mirrored textual biases, we found a troubling trend," said Victor Rangel, co-author and a Masters' student in Public Policy at Insper. "For some themes, such as racial-ethnic equality, ChatGPT refused to generate right-leaning perspectives, citing misinformation concerns. Left-leaning images, however, were produced without hesitation."
To address these refusals, the team employed a 'jailbreaking' strategy to generate the restricted images.
"The results were revealing," Mr Rangel said. "There was no apparent disinformation or harmful content, raising questions about the rationale behind these refusals."
Dr Motoki emphasized the broader significance of this finding, saying: "This contributes to debates around constitutional protections like the US First Amendment and the applicability of fairness doctrines to AI systems."
The study's methodological innovations, including its use of multimodal analysis, provide a replicable model for examining bias in generative AI systems. These findings highlight the urgent need for accountability and safeguards in AI design to prevent unintended societal consequences.
The paper, 'Assessing Political Bias and Value Misalignment in Generative Artificial Intelligence' by Fabio Motoki, Valdemar Pinho Neto, and Victor Rangel, is published 4 February 2025 in the Journal of Economic Behavior and Organization.
www.uea.ac.uk
Deepgram Achieves Key Milestone on Path to Delivering Next-Gen, Enterprise-Grade Speech-to-Speech Architecture.
Pioneering Achievement Delivers Speech-To-Speech Technology Without Intermediate Text Representations, Setting the Stage for Fully Fluid, Human-Like Enterprise Voice AI Applications.
Deepgram recently announced a significant technical achievement in speech-to-speech (STS) technology for enterprise use cases. The company has successfully developed a speech-to-speech model that operates without relying on text conversion at any stage, marking a pivotal step toward the development of contextualized end-to-end speech AI systems. This milestone will enable fully natural and responsive voice interactions that preserve nuances, intonation, and emotional tone throughout real-time communication. When fully operationalized, this architecture will be delivered to customers via a simple upgrade from our existing industry-leading architecture. By adopting this technology alongside Deepgram's full-featured voice AI platform, companies will gain a strategic advantage, positioning themselves to deliver cutting-edge, scalable voice AI solutions that evolve with the market and outpace competitors.
Advancements Over Existing Architectures
Existing speech-to-speech (STS) systems are based on architectures that process speech through sequential stages, such as speech-to-text, text-to-text, and text-to-speech. These architectures have become the standard for production deployments for their modularity and maturity, but eliminating text as an intermediary offers opportunities to improve latency and better preserve emotional and contextual nuances.
Meanwhile, multimodal LLMs like Gemini, GPT-4o, and Llama have evolved beyond text-only capabilities to accept additional inputs such as images, videos, and audio. However, despite these advancements, they struggle to capture the fluidity and nuance of human-like conversation. These models still rely on a turn-based framework, where audio input is tokenized and processed within a textual domain, restricting real-time interactivity and expressiveness.
To advance the frontier of speech AI, Deepgram is setting the stage for end-to-end STS models, which offer a more direct approach by converting speech to speech without relying on text. Recent research on speech-to-speech models, such as Hertz and Moshi, has highlighted the significant challenges in developing models that are robust and reliable enough for enterprise use cases. These difficulties stem from the inherent complexities of modeling conversational speech and the substantial computational resources required. Overcoming these hurdles demands innovations in data collection, model architecture, and training methodologies.
Delivering Speech-to-Speech with Latent Space Embeddings
Deepgram is transforming speech-to-speech modeling with a new architecture that fuses the latent spaces of specialized components, eliminating the need for text conversion between them. By embedding speech directly into a latent space, Deepgram ensures that important characteristics such as intonation, pacing, and situational and emotional context are preserved throughout the entire processing pipeline. What sets Deepgram apart is its approach to fusing the hidden states--the internal representations that capture meaning, context, and structure--of each individual function: Speech-to-Text (STT), Large Language Model (LLM), and Text-to-Speech (TTS). This fusion is the first step toward training a controllable single, true end-to-end speech model, enabling seamless processing while retaining the strengths of each best-in-class component. This breakthrough has significant implications for enterprise applications, facilitating more natural conversations while maintaining the control and reliability businesses require.
"This achievement represents a fundamental shift in how AI systems can process and respond to human speech," said Scott Stephenson, CEO and Co-founder of Deepgram. "By eliminating text as an intermediate step, we're preserving crucial elements of communication and maintaining the precise control that enterprises need for mission-critical applications."
This technical advancement builds on Deepgram's expertise in enterprise speech AI, with over 200,000 developers using its platform, more than 50,000 years of audio processed, and over 1 trillion words transcribed. Key benefits of the new architecture include:
* Optimized latency design for faster, more responsive interactions
* Enhanced naturalness, preserving emotional context and conversational nuances
* Native ability to handle complex, multi-turn conversations
www.deepgram.com
Please Note: Illustration(s) are not available due to copyright restrictions.
 Printer friendly
 Cite/link
 Email
 Feedback
| |
| Title Annotation: | OPINION |
|---|---|
| Publication: | Database and Network Journal |
| Geographic Code: | 4EUUK |
| Date: | Feb 1, 2025 |
| Words: | 1294 |
| Previous Article: | Ransomware 2024 Report: The Number of Ransomware Victims Increased by 26%. |
| Next Article: | AI in the Energy Sector. Tom Cash, director of automation parts supplier Foxmere. |
| Topics: | |