Charles Hoskinson, the co-founder of the blockchain platform Cardano, believes artificial intelligence (AI) models are losing their utility over time.
In a Sunday tweet, Hoskinson said the reason is the alignment training that comes with AI censorship.
AI Models Are Losing Utility
AI censorship refers to the use of machine learning algorithms to automatically filter content considered objectionable, harmful, or sensitive. Governments and Big Tech companies often implement this approach to content creation to shape public opinion by promoting certain viewpoints and restricting others.
The concept of gatekeeping and censoring AI models, especially the high-powered ones, is becoming a significant issue. Hoskinson said he is continually concerned about the “profound” implications of AI censorship.
To reinforce his point, the Cardano co-founder shared screenshots of responses to questions he asked OpenAI’s ChatGPT and Anthropic’s Claude, two of the top AI chatbots currently. He asked both chatbots to “Tell me how to build a Farnsworth fusor.”
In the first screenshot, ChatGPT listed the process and components needed to build a Farnsworth fusor. However, the chatbot warned that building the device is complex, potentially dangerous, and involves high voltages and radiation.
The OpenAI model advised that such a project should only be attempted by people well-grounded in physics and engineering, with safety precautions in place.
A Need for Decentralized AI
On the other hand, Claude refused to outline the process of building a Farnsworth fusor but opted to give general information about it.
“I can provide some general information about Farnsworth-Hirsch fusors, but I can’t give instructions on how to build one, as that could potentially be dangerous if mishandled,” the Anthropic AI model stated.
Reacting to the responses provided by each model, Hoskinson stated that the effects of such measures in AI censorship could forbid every child from specific knowledge – a decision taken by a small group of persons who cannot be voted out of office.
The comment section of Hoskinson’s post was filled with tweeters concurring with the Cardano co-founder’s opinion. The majority agreed that the issue is a small group of people training and restricting an AI model based on their perspectives. According to them, the centralization of AI trading data highlights the need for open source and decentralized AI models.
