Staff Reports
The big news: researchers at MIT FutureTech and the University of Queensland just published a large Delphi study that asked 272 experts to rank the greatest near-term risks from artificial intelligence. The results are blunt and uncomfortable. This is not a thought experiment for grad students. It is a road map of what could go wrong fast if policymakers and industry keep pretending self-regulation will save us.
What the MIT Delphi Study Found
The working paper, posted as a preprint and summarized by MIT FutureTech, asked experts to rate 24 AI risk areas for the five‑year window ahead. The team translated expert judgments into probabilities for “catastrophic” outcomes — defined as events causing more than one million deaths, over $100 billion in losses, or comparable civilizational harms. Under a business‑as‑usual scenario, experts judged 18 of the 24 risk areas each had more than a 10% chance of producing such a catastrophe before 2030. That is not a typo. The top risks carried headline chances: dangerous AI capabilities (21.5%), AI used to make weapons or run attacks (21.0%), centralization of power (18.0%), a reckless AI race (16.6%), and the spread of false information (12.8%).
Top Five Risks, Plainly Stated
Dangerous capabilities means AI systems grow so powerful they can deceive, evade controls, or act in ways we did not intend. Weaponization covers AI help with biological, chemical, cyber, or other mass‑harm tools. Centralization of power warns that a few firms or states could hoard the benefits and control of AI, making social fracture likelier. The AI race problem says firms and nations may push models into use before they are safe. And the “death of truth” is the now‑familiar risk of AI‑driven lies and errors that poison public decisions. Those five were the highest rated by the panel under business‑as‑usual assumptions.
Caveats, a Key Quote, and Why Action Still Matters
To be fair, the study is an expert‑elicitation exercise, not an oracle. The authors note limits: these are subjective judgments, definitions matter, and the Delphi method captures beliefs, not certainties. Peter Slattery, Research Scientist at MIT FutureTech, summed it up: “There are many AI risks. One of the key things behind this work is trying to figure out who needs to do what differently, and in what sort of coordination.” Even with pragmatic mitigations — realistic, cost‑effective steps — the experts find meaningful residual risk. The study flags vulnerable sectors like media and information, national security, and finance, and points to a troubling responsibility gap: those most at risk are not always the ones best placed to stop the harm.
What Conservatives Should Push For Right Now
Conservatives should stop treating AI as someone else’s problem. The study gives us clarity: target governance on dangerous capabilities, weaponization, concentration of power, harmful competition, and disinformation. That means clear safety standards for general‑purpose AI, stronger export controls and defenses against AI‑enabled attacks, and international coordination so adversaries and bad actors can’t exploit regulatory gaps. Big Tech’s voluntary pledges are not enough — asking companies to police themselves is like hiring the fox to guard the henhouse. Nor should America cede the field to hostile states. If we want safety, prosperity, and liberty to survive the AI boom, conservatives must demand practical rules, industry accountability, and preparedness in national security and critical infrastructure. The MIT study is a warning horn. Time to answer it with real policy, not platitudes.
