The rise of ChatGPT and similar artificial intelligence systems has been accompanied by a sharp increase in anxiety about AI. For the past few months, executives and AI safety researchers have been offering predictions, dubbed “P(doom),” about the probability that AI will bring about a large-scale catastrophe.
Worries peaked in May 2023 when the nonprofit research and advocacy organization Center for AI Safety released a one-sentence statement: “Mitigating the risk of extinction from A.I. should be a global priority alongside other societal-scale risks, such as pandemics and nuclear war.” The statement was signed by many key players in the field, including the leaders of OpenAI, Google and Anthropic, as well as two of the so-called “godfathers” of AI: Geoffrey Hinton and Yoshua Bengio.
You might ask how such existential fears are supposed to play out. One famous scenario is the “paper clip maximizer” thought experiment articulated by Oxford philosopher Nick Bostrom. The idea is that an AI system tasked with producing as many paper clips as possible might go to extraordinary lengths to find raw materials, like destroying factories and causing car accidents.
A less resource-intensive variation has an AI tasked with procuring a reservation to a popular restaurant shutting down cellular networks and traffic lights in order to prevent other patrons from getting a table.
Office supplies or dinner, the basic idea is the same: AI is fast becoming an alien intelligence, good at accomplishing goals but dangerous because it won’t necessarily align with the moral values of its creators. And, in its most extreme version, this argument morphs into explicit anxieties about AIs enslaving or destroying the human race.
A paper clip-making AI runs amok is one variant of the AI apocalypse scenario.
Actual harm
In the past few years, my colleagues and I at UMass Boston’s Applied Ethics Center have been studying the impact of engagement with AI on people’s understanding of themselves, and I believe these catastrophic anxieties are overblown and misdirected.
Yes, AI’s ability to create convincing deep-fake video and audio is frightening, and it can be abused by people with bad intent. In fact, that is already happening: Russian operatives likely attempted to embarrass Kremlin critic Bill Browder by ensnaring him in a conversation with an avatar for former Ukrainian President Petro Poroshenko. Cybercriminals have been using AI voice cloning for a variety of crimes – from high-tech heists to ordinary scams.
AI decision-making systems that offer loan approval and hiring recommendations carry the risk of algorithmic bias, since the training data and decision models they run on reflect long-standing social prejudices.
These are big problems, and they require the attention of policymakers. But they have been around for a while, and they are hardly cataclysmic….
