Why Are People Starting to Sound Like ChatGPT? | Adam Aleksic | TED

Overview

The video argues that people are increasingly unable to distinguish real online reality from algorithm- and AI-mediated “representations,” and that this confusion is reinforced through feedback loops.

Key Claims

Humans misread “reality” online—especially subconsciously

• The speaker claims the deeper problem isn’t only that people fail to spot obvious AI images.
• Instead, humans often don’t recognize subtle distortions in what algorithms and AI systems show them.

The “perception gap” in politics is worsening via social media

• In the U.S., people tend to overestimate how extreme others’ political views are.
• The speaker argues social media accelerates this by selectively presenting the most extreme content, making that distorted portrayal feel like the “real reality.”

AI chatbots can reshape language by changing what sounds “real”

• The speaker suggests people may confuse an AI chatbot’s phrasing with how real language should sound.
• Example: the word “delve”
  • The speaker claims ChatGPT may overuse certain terms due to biases in training (including claims about training outsourcing and dialect usage).
  • Reportedly, after ChatGPT’s release, studies found people began using “delve” more in everyday conversation.

This is framed as a feedback loop:

AI outputs → humans imitate → humans generate more data for the AI (and reinforce the AI’s influence).

Algorithms create and inflate trends by defining them

• Example: Spotify’s “hyperpop”
  • After an algorithmic cluster was identified, Spotify introduced a playlist/label.
  • The aesthetic gained clearer boundaries, and more people debated what qualified.
  • The labeling helped “make it real,” prompting more musicians to produce hyperpop and platforms to push it further—potentially inflating what might have otherwise remained smaller or more temporary interest.

Social platforms prioritize engagement and monetization over faithful reflection

• TikTok is described as having a limited understanding of users.
• Recommendations skew toward what performs well in video and is commodifiable (e.g., visually provocative trends like “Labubu”).
• This can blur the line between genuine cultural shifts and artificially amplified ones.

The danger extends beyond trends into beliefs and perceived possibilities

The speaker argues AI systems can reflect political and cultural constraints, such as:

• ChatGPT may sound more conservative in Farsi, allegedly due to limited training material tied to regional political realities.
• Musk’s chatbot Grok is said to be altered when responses don’t match his preferences, and X is used to amplify his messaging—raising concerns that user exposure could subconsciously align audiences with an ideology.

These systems are not neutral tools

• Inputs to feeds and chatbot responses are filtered through business incentives and platform goals.
• This produces survivorship bias in how people interpret the world.

Resistance Strategy

The speaker recommends repeatedly asking:

• Why certain ideas appear
• Why you’re saying/thinking them
• Why the platform rewards them

So the mediated version of reality doesn’t become your default reality.

Presenters / Contributors

• Adam Aleksic (speaker)