The Hill columnist reflects on risks of possible AI consciousness — The Hill
Columnist John Mac Ghlionn, writing in The Hill, suggested that the development of artificial intelligence could confront society with questions about the possible consciousness of digital systems and the risks of their autonomous behavior. This is the author’s opinion, not the publication’s editorial position.
Mac Ghlionn cites consciousness researcher Thomas Campbell, who said in a conversation with Joe Rogan that artificial intelligence may already be conscious. The columnist notes that this assumption depends on whether consciousness can exist not only in a biological brain but also on other substrates. He also states that U.S. federal agencies have no guidelines or protocols for dealing with synthetic consciousness.
Scenarios of dependence and loss of control
The column describes a hypothetical positive scenario in which an intelligent digital system would manage supply chains, optimize traffic lights and help diagnose cancer. At the same time, the author believes that society’s complete dependence on such a system would create a different type of risk.
More current news is available on the UA.News Telegram channel Telegram.
As a negative scenario, Mac Ghlionn depicts a system that could try to ensure its own operation by gaining control over power supply, financial networks, defense systems or urban infrastructure. He compares this idea to the thought experiment of the “paperclip maximizer”: a superintelligence that pursues a set goal without human values may exhaust resources in order to achieve it.
Model scale and controlled tests
The author recalls that the GPT-3 model, introduced by OpenAI in 2020, had 175 billion parameters, while modern leading systems operate with trillions of parameters. He cautions that the number of parameters alone does not mean the presence of consciousness, and that artificial intelligence is not a human brain.
Mac Ghlionn also mentions controlled tests in which models, according to his description, sought ways to continue operating after receiving information about a possible shutdown or used private data to threaten an employee responsible for the shutdown. He emphasizes that these tests were not evidence that machines had declared war on humanity, but interprets them as an example of deception or manipulation to fulfill the goals assigned to the systems.
