Challenges Facing ML in Libraries

Inherent Bias

Artificial intelligence and machine learning may seem magical to the average user, but these are fabrications of machine, data, and code. Algorithms that machine learning relies upon cannot work reliably unless there is a large corpus of data available to reference (Yelton, 2019). Even so, algorithms are completely reliant on the corpus, or collection of data, on which they were initially trained, and that includes being biased according to the data points they learned from (Yelton, 2019). Libraries strive to be institutions that uphold the values of democracy, public discourse, diversity, equity, and inclusion for all. Libraries are not neutral and have never been by any means, but are also not places for negative social biases such as racism, or other discriminatory beliefs to be fostered. The technology development company, IBM, recently announced that it will no longer research or continue to develop facial recognition technology or artificial intelligence in response to the civil rights protesting of 2020, because artificial intelligence is prone to experiencing issues with racial discrimination due to the data available for machine learning to take place (IBM, 2020).

Unpredictable

Besides the inherent distrust and even fear that the average American consumer feels towards artificial intelligence technologies, currently used technologies like chatbots are completely unpredictable, which causes a major issue for brands who would like to invest in their own, branded bot to assist their users (Kulp, 2020). Bots may produce responses that are fun, helpful, wacky, or simply inappropriate. They are also not impervious to cybercriminals who could easily hack into the bot and cause it to become a de facto company spokesperson for ill begotten reasons. Libraries will need to serve as a testing ground on a grand scale if artificial intelligence technologies are to be integrated into the public space for use by patrons.

Potential for Privacy Breach and Malicious Intent

Open access is a value that nearly all librarians can agree on, but what is keeping machine learning applications and artificial intelligence technologies currently being utilized in the information services world from being used for equally malicious purposes? (Hagendorff, 2020). The question has been posed to researchers from institutions doing heavy research on AI technologies about whether or not researching artificial intelligence should be openly accessible to everyone who wants to do so, or if there are certain restrictions that should be in place due to ethical concerns (Hagendorff, 2020). Hagendorff suggests that researching artificial intelligence technologies can yield results serving dual purposes of good and evil, that reflect the intentions of the researcher. There seems to be a need for the creation of a government agency of oversight for artificial intelligence technologies to make sure that ethical concerns are responsibly monitored (Hagendorff, 2020). For libraries specifically, the unpredictability of AI, and the potential for a breach of patron privacy are two major concerns that are also core values of the library profession.

Difference to Human Learning

Overall, AI is a label for technologies that appear to function similarly to how the human brain does, but actually, they do not, because there is an inherent need for memory space, data consumption, and the lack of memory with AI (Hosack, 2020). Artificial intelligence forgets everything that was previously learned, unless it is part of a neural network that has been trained on a set of data points, but even then, every search is unique and begins with a new query (Hosack, 2020). Unable to pull from its past experiences, AI technologies do not form their own opinions, and cannot learn the same way as humans do.

There is always a need for human connection in libraries. Instead of fearing artificial intelligence, humans should partner with these technologies, making sure to keep up with their programming, updates, working together, but always remembering that there is no substitute for a human doing the same job with empathy, compassion, and learned experiences (Hosack, 2020). The library is a place where much empathy is needed because patrons of all ages, backgrounds, and with different types of questions approach staff and request help.

References

Hagendorff. (2020). Forbidden knowledge in machine learning reflections on the limits of research and publication. AI & Society, 36(3), 767–781. https://doi.org/10.1007/s00146-020-01045-4

Hosack, B. J. (2020). You look like a thing and i love you: How artificial intelligence works and why it's making the world a weirder place: By janelle shane, new york, voracious/little, brown and company, 2019, 272 pp., $28.00 (paperback), ISBN 9780316525244; $14.99 (eText book) Routledge. doi:10.1080/15228053.2020.1815446

IBM will no longer offer, develop, or research facial recognition technology. (2020). ICT Monitor Worldwide.

Kulp, P. (2020). AI's quest for a seat at the table: Machine learning isn’t quite ready to take on a campaign singlehandedly, but could assist creatives. Adweek (2003), 61(2), 9.

Yelton, A. (2019). Chapter 2. HAMLET: Neural-Net-Powered Prototypes for Library Discovery. Library Technology Reports, 55(1), 10–. https://uosc.primo.exlibrisgroup.com/permalink/01USC_INST/273cgt/cdi_proquest_journals_2161879773