Just goofing around, I proposed to Google AI:
AI responses on audio products can be very subjective since it is trained by scanning of social networks, which is full of subjective opinions.
https://www.google.com/search?num=1...hUKEwiC2Mj2gvOPAxW5PDQIHZUfNXgQvgUoAXoECBEQAg
AI Overview
It is correct that AI responses on audio products can be very subjective, which is a direct result of how AI models are trained
. Because these models learn from vast amounts of internet data, including social networks, their outputs reflect the biases, opinions, and inaccuracies present in that data.
Here is an explanation of why this happens and how to address it.
Why AI responses on audio products can be subjective
- Training on biased data: AI systems are trained on massive datasets scraped from the internet, which inevitably contain human biases and subjective opinions. For example, if a popular audio product is widely praised on forums and social media, an AI will learn to associate that product with positive sentiment, even if the praise comes from only a specific demographic.
- Pattern matching, not understanding: A large language model (LLM) is an "imitation engine" that identifies and reproduces statistical patterns from its training data. It does not possess true understanding or common sense. It generates plausible-sounding text based on what it has seen, not necessarily what is true. This means that when it generates a review for a pair of headphones, it will repeat phrases and sentiments commonly associated with that type of product without having actually listened to them.
- The "average Jane" problem: When dealing with subjective topics like sound quality, there is no single "ground truth" to train against. Training is often done by human labelers, but their opinions can differ based on personal perspective, background, and beliefs. If the training data is not representative of all relevant consumer segments, the AI will fail to make predictions that generalize to different contexts.
- Amplifying popular opinion: Models can lean heavily on the most common or linked sources they were trained on, leading to a focus on "big brands" or products that generate the most internet discussion. This can cause smaller, niche products to be overlooked, even if they are higher quality.
How to get more objective AI responses for audio products
To get the most value out of an AI response on an audio product, consider using these strategies.
Request objective, data-driven information
Instead of asking for a review, which invites a subjective response, focus on objective questions that the AI is better equipped to answer based on its training data. For example:
- Ask for a summary of professional reviews from specific, trusted audio publications. This guides the AI to use more credible sources.
- Inquire about the technical specifications of a product, such as the battery life, drivers, and codec support.
- Prompt the AI for a neutral comparison of features between several models. For example, ask for a table comparing the noise-canceling capabilities, connection types, and price points of three different headphones.
Analyze with critical thinking
An AI can be a powerful assistant for gathering and summarizing information, but you must take on the role of fact-checker and critical thinker.
- Cross-check with trusted sources: Just as you would with any review, use credible sources to verify the information. For example, check a trusted audio reviewer's YouTube channel or website to see how their impressions compare to the AI's summary.
- Look for citations: If the AI provides citations, follow them to determine if they back up the claims made in the summary. If they don't, the AI may be misinterpreting information or hallucinating.
- Understand the limitations: Always remember that an AI is not a human. It cannot experience sound, so its "knowledge" of audio quality is based on the opinions it has processed from its training data.
Refine your prompts for nuance
By providing more context, you can direct the AI to provide a more nuanced and less biased response.
- Specify the target audience and use case. For example: "For a music producer who prioritizes a flat, accurate frequency response, what do critics generally say about the Sennheiser HD 600?"
- Direct the AI to a specific perspective. For example: "Based on reviews, what are the most common positive and negative feedback points regarding the build quality and comfort of the Sony WH-1000XM5 headphones?"
His honest love for sharing knowledge is what the world needs more of.
