If you wanted to intentionally build a “faulty” and “dangerous” AI product, you couldn’t find a better recipe than the one currently being used by the tech industry. This conclusion comes not from critics, but from the contract workers themselves, one of whom warned the US Congress about this exact outcome a year before AI’s public failures became mainstream news.
Step 1: Hire a workforce under false pretenses. Lure in skilled writers and researchers with vague job titles, then assign them to psychologically taxing content moderation without their consent. This ensures they are unprepared and unsupported from day one.
Step 2: Impose impossible deadlines. Give them complex tasks requiring careful research and judgment, but only allow them minutes to complete them. Constantly shrink these deadlines to maximize pressure and ensure that corners are always being cut.
Step 3: Devalue expertise. Systematically assign tasks to workers who have no qualifications in the subject matter. Tell a fine arts major to vet physics content and a history major to edit medical advice. Forbid them from skipping tasks they are unqualified for.
Step 4: Loosen safety standards. When your existing safety rules get in the way of performance, simply change them. Create loopholes that allow your AI to replicate hate speech and other harmful content, while claiming you haven’t changed your core policy on “generating” it.
By following these simple steps, you can create an AI that is unreliable, unsafe, and ethically compromised, all while maximizing productivity and profit. It is a recipe for disaster that is being followed to the letter.