A new way to build AI, openly

The speaker is discussing the current state of artificial intelligence (AI) and its potential impact on society. They argue that the current system of AI development, where models are trained in secret and released as "castles" with little transparency or accountability, is concerning for several reasons. Firstly, it makes it difficult to evaluate the accuracy and fairness of these models, as there is no access to the training data or evaluation process. Secondly, it raises concerns about values being decided unilaterally by the rulers of these castles, rather than being democratically determined based on input from everyone. Finally, it neglects to attribute the creators of these models and compensate them fairly for their work.The speaker offers hope for a different future, where AI development is more transparent and participatory. They point to examples like Wikipedia and Linux, which were initially seen as ridiculous but eventually prevailed due to their open source nature. They argue that the purpose of AI should be to organize and augment the culture of our civilization, by enabling people to create, invent, and discover. However, realizing this vision will require innovation and participation from researchers, companies, policymakers, and the public at large.