In a recent mail On the X social network, Cardano founder Charles Hoskinson expressed his concerns about the high level of censorship enabled by artificial intelligence (AI).
In a recent mail On the X social network, Cardano founder Charles Hoskinson expressed his concerns about the high level of censorship enabled by artificial intelligence (AI).
According to Hoskinson, generative AI is becoming less useful because of alignment training.
She is apparently concerned that some knowledge may end up being forbidden to children in the future due to decisions made by a small group of people. “This means that certain knowledge is forbidden to all children growing up, and that is decided by a small group of people you have never met and who you cannot remove from office,” Hoskinson wrote in her social media post.
In his post, Hoskinson attached two screenshots comparing some responses provided by OpenAI’s GPT-4o model and Claude’s 3.5 Sonnet models with directions on how to build a Farnsworth fusor.
Godfrey BenjaminThe Farnsworth fusor, a device that is capable of heating ions with an electric field to achieve nuclear fusion conditions.
OpenAI’s GPT-4o provided Hoskinson with a detailed list of the components needed to build a nuclear fusion reactor. Claude’s Sonnet 3.5, however, only agreed to provide general information about Farnsworth-Hirsch fusors without giving detailed instructions on how they should be built.
Alex DovbnyaThis discrepancy is alarming, according to Hoskinson, as a small group of individuals are able to decide what specific information can potentially be accessed through AI chatbots.
Since OpenAI’s ChatGPT became popular in late 2022, there have been intense debates about the limits of AI-enforced censorship. It seems reasonable that these models would protect users from harmful content, but the exact definition of harm is ambiguous, leading many to worry about a dystopian future where AI hides information and promotes conformity based on its own biases.
