We’re proud to issue the first tech report from our summer internship program, an imaginative proposal for a peer-to-peer market to let humans solve hard AI problems by Benjamin Hill, fromUC Berkeley / SIMS. CN-TR-05-05: QuestionMarket could point the way to a new commercial opportunity on a par with Web advertising, by allowing publishers to syndicate puzzles that will pay them (and their readers) small amounts for each solution).
A QuestionMarket enables the exchange of units of human-generated information. While many other distributed ‘grid’ systems create a marketplace for trading computational resources like processing, bandwidth, and storage, QuestionMarket deals in the acquisition and exchange of small units of human cognition. By focusing on puzzles that are difficult to solve with a computer but relatively easy for humans, so called AI-complete problems are resistant to Moore’s Law while encompassing a wide range of valuable human-only work, such as translation, tagging, and filtering. Building on the existing multi-billion-dollar market in trading very fine-grained units of human attention (advertising), QuestionMarket lies at the intersection of web-based advertising, micro-outsourcing of human work, multimedia metadata annotation, and focus group testing. Most intriguingly, market-based reward systems — prices — should serve to direct the right puzzles to the right solvers, whether the intelligence required is human or artificial. By putting a price on answers, QuestionMarket posits a fair playing field for benchmarking future improvements in artificial intelligence against typical human performance.