Google and Bing are fighting it out to be The Best AI-Powered Search Engine in the Universe at the moment but this isn’t the first time a company has tried to build a knowledge system that returns answers to people’s various queries.
Mindpixel was a web-based AI project that ran from 2000–2005. It was developed by computer scientist Chris McKinstry as a knowledge base of human-validated true/false statements. The process of obtaining these statements was particularly interesting to me:
Participants in the project created one-line statements which aimed to be objectively true or false to 20 other anonymous participants. In order to submit their statement they had first to check the true/false validity of 20 such statements submitted by others. Participants whose replies were consistently out of step with the majority had their status downgraded and were eventually excluded. Likewise, participants who made contributions which others could not agree were objectively true or false had their status downgraded. A validated true/false statement is called a mindpixel.
via Wikipedia
Although OpenAI’s process for validating prompts and responses also uses humans (some who get paid a pittance), this was paradoxically more open and closed-minded at the same time. Despite 5 years of work, Mindpixel lost its free server in September 2005 and while Chris McKinstry rewrote Mindpixel as Mindpixel 2 with a view to move it to a new French server, it never came to be as McKinstry died of suicide in 2006. Ironically, the Mindpixel website looks like a SEO content farm, like the kind that would leverage generative AI for organic clicks.
If you want to read more about the project, Toby Howard wrote a piece for Personal Computer World Magazine in December 2000 called “Modelling minds“.