The Silicon Sieve

“Weakness and ignorance are not barriers to survival, but arrogance is.”
— Death’s End

I’ve worked on an assembly line, doing tasks that were less economical to automate than it was paying me nearly minimum wage to accomplish. Thinking back on the nature of my work, I can only hope that’s no longer the case. Put another way, do we really want our descendants to pursue careers flipping burgers?

Every generation of library search tools has fixed problems with previous iterations of such systems, but while also inadvertently giving rise to some new ways that things can go wrong. Those obstacles are part of the path towards technological progress, and almost hardly ever sufficient reason to abandon more modern methods. An old proverb puts these situations nicely: “We can complain because rose bushes have thorns, or we can rejoice that thorn bushes have roses.”

A big issue facing us right now, as far as discovery layers are concerned, is how since we’ve thankfully melded almost everything together, into a single index, the resulting underlying mix of inadequate or inconsistent metadata does not at times make for sensible results, especially with regards to how they may be ranked.

AI will resolve this problem. It’s already making inroads at doing so. Search engine algorithms aren’t getting worse, after all. Actually, semantic networks will even by in large prevent these problems from occurring in the first place, because the difficulties with interoperability of linked data, as with hashtags and other folksonomies, mainly exist exist by virtue of it being human-generated.

Don’t believe me? Type “foreign b&w movie @ a knight playing chess with death” into Google and tell me what pre-coordinated set of controlled vocabulary was involved in delivering your results. Or, ask Siri for driving directions, and consider how there’s no human computational element present in plotting your route.

We don’t use the library hand or telnet OPACs anymore. Instead of offering mediated searching, our discovery layer likewise processes over a million queries per year. And circulation desk staff, rather than keying them in manually, scan barcodes and RFID chips. Machine learning can be applied to knowledge organization in much the same vein.

This also holds true, of course, for all aspects of contemporary library operations, as well as the overall nature of libraries in general. The connotation of “librarian” may someday elicit sentiments similar to what “stagecoach driver” does today. I can only hope.

Further Reading

• Can we build AI without losing control over it?
• Could a computer ever create better art than a human?
• The Most Terrifying Thought Experiment of All Time
• The Rise of the Robot Reporter
• Speech synthesis from neural decoding of spoken sentences
• The telegraph in the library
• Why Would Anyone Build Their Own Coffin?