AI chatbots refuse to produce ‘controversial’ output − why that’s a free speech problem

×

Error message

  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ds_changed in eval() (line 16 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$ss_search_api_url in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).
  • Notice: Undefined property: stdClass::$tm_title in eval() (line 20 of /var/www/7/modules/php/php.module(80) : eval()'d code).

Still, the conversation on AI ignores another crucial issue: What is the AI industry’s approach to free speech, and does it embrace international free speech standards?

Metadata
Provider: 
The ConversationThe Conversation
URL: 
https://theconversation.com/ai-chatbots-refuse-to-produce-controversial-output-why-thats-a-free-speech-problem-226596https://theconversation.com/ai-chatbots-refuse-to-produce-controversial-output-why-thats-a-free-speech-problem-226596
Retrieved on: 
Thursday, April 18, 2024
Information
Tags: 
Civil service commission, Vanderbilt University, Synthetic media, Freedom, Hatred, International, Bias, COVID-19, Future, Climate, Research, United Nations, Social media, Policy, Child, AI, Woman, European Commission, ChatGPT, Government, Digital Services Act
Geotags: 
INDIA
Summary
Key Points: 
  • Still, the conversation on AI ignores another crucial issue: What is the AI industry’s approach to free speech, and does it embrace international free speech standards?
  • In practice, this means that AI chatbots often censor output when dealing with issues the companies deem controversial.
  • Without a solid culture of free speech, the companies producing generative AI tools are likely to continue to face backlash in these increasingly polarized times.

Vague and broad use policies

  • Companies issue policies to set the rules for how people can use their models.
  • With international human rights law as a benchmark, we found that companies’ misinformation and hate speech policies are too vague and expansive.
  • Our analysis found that companies’ hate speech policies contain extremely broad prohibitions.
  • To show how vague and broad use policies can affect users, we tested a range of prompts on controversial topics.
  • More recently, India confronted Google after Gemini noted that some experts consider the policies of the Indian prime minister, Narendra Modi, to be fascist.

Free speech culture

  • If they serve a global audience, they may want to avoid content that is offensive in any region.
  • This means society has an interest in ensuring such policies adequately protect free speech.
  • Even where a similar legal obligation does not apply to AI providers, we believe that the companies’ influence should require them to adopt a free speech culture.
  • At least two of the companies we focused on – Google and Anthropic – have recognized as much.

Outright refusals

  • Therefore, users’ exposure to hate speech and misinformation from generative AI will typically be limited unless they specifically seek it.
  • This is unlike social media, where people have much less control over their own feeds.
  • Stricter controls, including on AI-generated content, may be justified at the level of social media since they distribute content publicly.
  • Refusals to generate content not only affect fundamental rights to free speech and access to information.
  • The Future of Free Speech is a non-partisan, independent think tank that has received limited financial support from Google for specific projects.
  • In all cases, The Future of Free Speech retains full independence and final authority for its work, including research pursuits, methodology, analysis, conclusions, and presentation.
  • The Future of Free Speech is a non-partisan, independent think tank that has received limited financial support from Google for specific projects.
Copyright: 
2024 The Conversation. All Rights Reserved.