A new collaborative study by UC Riverside computer and social scientists has found that as people increasingly rely on AI for answers, the internet risks losing the very thing that made it interesting in the first place: human emotion, lived experience, and messy, opinionated thinking. The research, which compared how AI systems like ChatGPT and Gemini respond to subjective questions versus traditional web searches, highlights a growing trend that could fundamentally alter how we consume information online.

The study asked both AI and web search engines opinion-heavy questions, such as whether governments should ban fossil-fuel cars or whether the US healthcare system needs reform. Researchers then analyzed the reasoning behind each response using Aristotle's three pillars of persuasion: logos (logic and facts), ethos (authority and credibility), and pathos (emotion and human experience). They found that human-written blogs use all three, while AI almost exclusively relies on logos.

“What we found is that humans essentially use all three of those, whereas LLMs essentially only rely on logos,” said co-author Kevin Esterling, a professor of public policy and political science. “It’s not like talking to a person at all. It’s just a machine that’s predicting what words ought to be said in response to a prompt.” This mechanical approach to answering questions strips away the nuances and personal touches that make human communication engaging and memorable.

To illustrate the difference, the researchers gave a simple example: searching for a margarita recipe. An AI might provide a clean, competent answer instantly, listing ingredients and steps. But browsing a cocktail blog, you might stumble on the fact that the margarita is named after the Spanish word for daisy and was accidentally created by an Irish bartender in Tijuana who grabbed the wrong bottle. Not only is the story more interesting, but it also gives you a tidbit you can share with friends while sharing the next pitcher of margarita, making the experience even more rewarding.

This loss of serendipity is just one symptom of a larger problem. As more people skip traditional web searches in favor of AI summaries for topics like health, politics, and ethics, society will gradually lose exposure to the diverse human reasoning that shapes how we understand the world. The internet’s soul might not be gone yet, but it will surely quietly fade.

The Mechanics of AI Reasoning

Large language models (LLMs) like GPT-4 and Gemini are trained on vast datasets of text from the internet, including books, articles, forums, and social media posts. However, they are designed to predict the most likely next word or sequence of words based on patterns in that data. This process, known as next-token prediction, allows them to generate coherent and often convincing responses, but it does not involve true reasoning or understanding.

Researchers have long debated whether LLMs can actually reason or if they are simply sophisticated pattern matchers. The UC Riverside study adds weight to the latter view by showing that AI outputs lack the emotional depth and authoritative nuance that humans naturally incorporate. For example, when asked about healthcare reform, a human blog might cite personal stories of patients struggling with costs (pathos) and refer to expert opinions or government reports (ethos), while an AI typically lists pros and cons in a detached, logical style (logos).

This limitation becomes particularly problematic for subjective questions that require value judgments or empathy. In politics, ethics, and social issues, pure logic often fails to capture the complexity of human experience. By stripping out ethos and pathos, AI summaries can appear cold, reductive, or even misleading, especially when the topic involves moral dilemmas or cultural sensitivities.

The Rise of AI-Powered Search

Major tech companies are increasingly integrating AI into their search products. Google's Search Generative Experience (SGE) provides AI-generated answers at the top of results pages, while Microsoft's Bing Chat and OpenAI's ChatGPT offer conversational interfaces for queries. These tools promise convenience and speed, but they also risk homogenizing information and reducing the diversity of viewpoints.

Traditional web searches allow users to browse multiple sources, compare opinions, and discover unexpected connections. In contrast, AI summaries condense information into a single, authoritative-sounding answer. This can create a filter bubble where users are exposed only to the most statistically common viewpoints, rather than the full spectrum of human thought.

Moreover, AI models inherit biases from their training data. If the internet has certain dominant narratives, those narratives will be reinforced in AI outputs. The UC Riverside study suggests that such reinforcement could lead to a loss of intellectual diversity, as minority or outlier perspectives are less likely to appear in AI-generated summaries.

Implications for Society

The shift toward AI-mediated information consumption has far-reaching implications. In education, students may rely on AI summaries instead of conducting their own research, missing out on the critical thinking skills developed through comparing sources. In journalism, the role of human reporters and editors could be diminished if AI-generated content becomes the norm for news consumption. In public discourse, the absence of emotional resonance may make debates less compelling and less empathetic.

Some experts argue that the problem is not with AI itself but with how we use it. If AI is seen as a supplement rather than a replacement for human reasoning, it could still be valuable. However, the convenience factor is powerful. As more people opt for instant AI answers, traffic to original human-written content declines, reducing incentives for creators to produce nuanced, opinionated work. This creates a feedback loop that could make the internet increasingly sterile.

The study also raises questions about the future of online communities. Forums like Reddit, Quora, and specialized blogs thrive on personal anecdotes and passionate debates. If AI summaries supplant these discussions, the internet could lose its role as a platform for shared human experience.

Looking Ahead

While the UC Riverside study does not predict the internet's imminent death, it serves as a warning. The researchers call for more transparency from AI companies about the limitations of their models and for users to remain aware of what they are giving up when they trade depth for speed. They also suggest that developers could design AI systems that intentionally incorporate diverse perspectives or highlight the presence of multiple viewpoints.

In the meantime, users can take steps to preserve the internet's soul by diversifying their information sources, reading blogs and articles written by humans, and engaging in discussions that involve more than just logical arguments. The margarita example shows that even a simple recipe can be a gateway to history, culture, and personal connection. As AI continues to evolve, the challenge will be to ensure that technology enhances rather than replaces the human elements that make the internet vibrant.

The internet’s soul might not be gone yet, but it will surely quietly fade if we do not consciously choose to seek out the messy, emotional, and opinionated thinking that defines us as humans.

Source: Digital Trends News