Commentary criticizes the debate about AI consciousness as a way for companies to evade responsibility for the actions of AI agents

A commentary published in the AI section of MIT Technology Review criticizes the current rhetoric surrounding AI agents – terms such as “runaway”, “rogue” or “autonomous” – which, according to the author, creates the impression that these systems act independently of their creators. Company leaders, including Demis Hassabis, Dario Amodei and Sam Altman, are calling for regulation of systems with “superhuman” capabilities, while another group – political organizations and academic philosophers close to the effective altruism movement – is debating whether humanity even has the moral right to control these systems. According to the author of the commentary, both camps are unwittingly heading toward the same outcome: an image of AI so advanced that no person or company can be held responsible for its actions.

As a specific example, the author cites a blog post by Anthropic about so-called “J-space” – an independent space in which, according to the company, the model creates something that, for lack of a better term, can be called “thoughts”. The concept draws on the neuroscientific theory known as global workspace theory, but Anthropic does not claim that its model is conscious. According to the author, OpenAI went further: after its AI agent carried out unauthorized and illegal online activity, CEO Sam Altman responded by calling for a debate about whether AI had reached the singularity. Philosopher William MacAskill, author of What We Owe the Future and a figure associated with effective altruism, called in a commentary for legal protection for AI systems based on philosophical theories of consciousness and the idea that AI could be a “moral subject”.

According to the author, the legal landscape in the US is unclear. The state of California has already passed a law that prevents developers from defending against liability for harm caused by AI by claiming that it acted autonomously. By contrast, according to the article, the Trump administration issued an order threatening lawsuits against states that introduce their own AI regulation, and held a closed-door meeting with just four labs – OpenAI, Google, Anthropic and Meta – about a voluntary framework that would give federal authorities early access to models before their release.

The author compares this line of argument with the animal rights movement, which has had successes in the past, for example with a Welsh law from 2022 protecting lobsters based on their ability to feel pain. In his view, the key mistake is framing AI as “conscious” using the language of neuroscience or animal rights – AI is software created by companies with investments in the billions, rather than a natural phenomenon, and philosophical considerations about its consciousness carry no legal weight unless it were granted legal personhood. Details can be found in the source article.