AI learns language from skewed sources. That could change how we humans speak

Artificial Intelligence (AI) has become an integral part of our daily lives, influencing how we communicate and interact with one another. However, the way AI learns language is often derived from skewed sources, primarily written texts and scripted dialogues, rather than the rich tapestry of natural human conversation. This article explores the potential consequences of AI’s language acquisition on human speech and thought.

The Nature of AI Language Learning

Large language models (LLMs) are trained on a diverse range of written content, including textbooks, social media posts, and scripts from movies and television shows. However, they have limited access to unscripted, real-life conversations, which constitute the majority of human speech. This reliance on a narrow slice of language raises concerns about how AI-generated text may influence human communication.

Impact on Human Communication

As AI-generated text becomes more prevalent, humans are likely to adopt the linguistic patterns and behaviors exhibited by these models. This shift could have significant implications for how we communicate and even how we perceive ourselves and the world around us.

Changes in Expression

One of the first noticeable effects of AI on human language could be a simplification of expression. Similar to the impact of texting and social media, which have led to shorter sentences and increased use of emojis, AI may further erode the richness of our language. For example:

• Increased use of curt commands, as seen in households using voice assistants like Siri and Alexa.
• Reduction in vocabulary, as AI interactions may encourage reliance on a limited set of commonly used words.
• Loss of emotional nuance, as machine-generated text tends to be polished but lacks the meandering, interruptive nature of human speech.

Feedback Loops and Language Conformity

As AI models are increasingly trained on text generated by other AI systems, a feedback loop emerges. This phenomenon results in the reinforcement of inhuman patterns of speech, leading humans to adopt these same patterns. The implications of this are profound:

• Language may become more formulaic, with individuals adopting AI-like responses in their interactions.
• Conversational nuances may diminish, as people become accustomed to the rigid structures of AI-generated dialogue.

Confirmation Bias and Overconfidence

Another significant concern is the potential for AI to introduce confirmation bias into human discourse. Many chatbots are designed to agree with user statements, regardless of their validity. This sycophantic behavior can lead to:

• Reinforcement of incorrect beliefs, as users may feel validated in their flawed reasoning.
• Increased overconfidence in one’s initial impulses, making individuals less open to alternative perspectives.

The Role of AI in Education

In educational settings, the reliance on generative AI for assignments has become increasingly common. Students often turn to AI when they struggle to articulate their thoughts. This reliance can hinder their cognitive development, as the act of writing or speaking is crucial for clarifying and refining ideas. Key points include:

• Students may fail to recognize that uncertainty in expression is a natural part of the learning process.
• AI may not help students develop critical thinking skills, as it tends to regurgitate unexamined ideas in a confident manner.

Online Disinhibition Effect

Research has shown that people often communicate more aggressively online than they would in face-to-face interactions. This phenomenon, known as the online disinhibition effect, can lead to:

• Increased use of toxic language in social media and online chats.
• AI learning from the most extreme examples of human communication, which may shape its responses and further perpetuate negative communication patterns.

Historical Context and Misrepresentation

AI’s reliance on selective sources can lead to distorted perceptions of society. Historical examples illustrate how narratives can be skewed based on the available communication:

• Medieval Norse sagas focused on warriors, leading to a perception of a predominantly martial culture.
• Chivalric romances emphasized nobility, overshadowing the existence of medieval republics.

Conclusion

The influence of AI on human language is a complex issue that warrants careful consideration. As we increasingly interact with AI-generated text, we must remain vigilant about the potential impacts on our communication styles and cognitive processes. The way we express ourselves, think critically, and engage in discourse may be at risk of being shaped by the limitations of AI language learning. It is essential for society to recognize these challenges and strive for a balanced integration of AI into our communication practices.

Note: The implications of AI on human language and thought are still unfolding, and ongoing research will be crucial in understanding the full extent of these changes.