As someone who is building such instant answer system for our own web search engine [1] the level of failure here tells me that this is: - Either a basic, dist

As someone who is building such instant answer system for our own web search engine [1] the level of failure here tells me that this is: - Either a basic, distilled BERT based model (optimized for latency and scale) or - More likely, still a purely heuristics driven answers system like the one Google has been using for last 10+ years. The current NLP models are able to quite successfully answer these questions with proper context. In this case ignoring Lance or Mars in question looks more like old-school keyword based heuristics and there is no way it can get this right. Google is not only allowing mistakes in this, what is considered a fairly difficult problem to solve, but also questions that directly query their own knowledge graph, for example this one querying for a CEO of a well known public company where it returns the wrong answer [2] This only shows that 'emperor has no clothes' [3] and that there is still a lot of room for innovation left, specially on the 'organizing the world's information' front. [1] https://kagi.com https://kagi.com (currently in private beta) [2] https://ibb.co/qkXpdFB https://ibb.co/qkXpdFB [3] https://www.quora.com/Is-Google-really-in-a-decline https://www.quora.com/Is-Google-really-in-a-decline (almost 2M views for the top answer indicate that 'Is Google in decline" is a fairly popular question among presumably Google users)