A look at the forest
by Vladimir Zelevinsky
Vladimir Zelevinsky
Einstein once said (and by referring to Einstein, this post instantly receives 1.21 jigawatts of respectability):
“So many people today – and even professional scientists – seem to me like someone who has seen thousands of trees but has never seen a forest. [...] This independence created by philosophical insight is – in my opinion – the mark of distinction between a mere artisan or specialist and a real seeker after truth.”
– Letter to Robert A. Thornton, 7 Dec. 1944
90% of putting together software is chasing all those bytes, bits, and nibbles, making sure they all align right. Sometimes it is important to try to take a broader look, to summarize what it is we are trying to build. All generalizations are approximations, and thus distortions; consider the following (rather categorical) set of statements in this light. This is an attempt to look past the trees to see the forest.
Information access systems fall under the broad heading of HCIR, Human-Computer Information Retrieval.
To make such systems as effective as possible, it would be useful if computers were as smart as humans.
This (strong AI) is difficult. We are doing something much easier: we make computers smarter than humans.
This is not as much of a paradox as it might appear: weak AI only needs to be better than humans in its own narrow range of applicability. Thus, web search systems do not need to understand the meaning of web pages – but they should be able to sift through keyword indices as fast as possible. Natural language parsers do not need to be quite as accurate as humans in tagging parts-of-speech or extracting entities – but when they are an order of magnitude faster, the relative loss in accuracy might not matter.
The key for good HCIR is mutual participation of humans and computers; both parties should be doing what they are the best at. Good HCIR systems are built when this is considered during the design.
The third part – frequently neglected – is as important as the first two. It’s the interface between the two parties, the gray area, the system that translates human search intent into commands for the search system, and then translates computed information back into human-readable results. Without it, humans and computer won’t be able to co-operate perfectly – after all, they speak different languages.
on February 5, 2010 at 11:55 am
Permalink
[...] monitored countless search startups over the past decade, and this is spot on. To paraphrase Vladimir’s earlier post, this future asks cobbled “strong AI” to do work humans can do much better. Instead [...]