What the industry needs is OPEN <i>training data</i> for image and voice recognition, which can be annotated by human volunteers and used to improve new algos. This data can be hosted by archive.org under a permissive license.<p>For example, there is a commercial service in the "visual search" space which has been around for a few years, which uses a combination of algos and human employees. It is available as an API and app (8 cents per picture, or $10 per month unlimited).<p><i>"Besides CamFind, the Image Searcher, Inc. team also created TapTapSee, which is a similar app for the blind and visually impaired. It uses image recognition technology with photos taken on your smartphone and then a voice actually speaks the results. This earned the team an Access Award just one week ago from the American Foundation for the Blind"</i>, <a href="http://www.builtinla.com/blog/camfind-brings-visual-search-your-smartphone-putting-google-goggles-test" rel="nofollow">http://www.builtinla.com/blog/camfind-brings-visual-search-y...</a><p>An AFB review from 2013 (at that time the app was free), <a href="http://www.afb.org/afbpress/pub.asp?DocID=aw140704" rel="nofollow">http://www.afb.org/afbpress/pub.asp?DocID=aw140704</a><p>How much faster could this research area have moved if the collected data was open? What if Facebook/Instagram decided to contribute in some manner, given their large database of photos and tags?