Optimal Discovery?:

Siri, Alexa, and Other Virtual Personal Assistants in Libraries

About

Should libraries use virtual personal assistants (VPAs), such as Siri and Alexa, within the physical library? VPAs are interfaces that support users with information retrieval and task performance. They are largely proprietary software and hardware that exist in numerous, consumer-driven fields. They function as powerful tools for data aggregation that combine speed, user experience, and accuracy, making them of interest to libraries. These technologies are mainly unregulated and face issues of interoperability outside their own ecosystems, or system of compatibility. As a result, VPAs rely on a number of other digital and mechanical components to function, making their infrastructure fragile and their ability to parse and understand language unreliable or slow. While VPAs have the potential to enhance library search experiences, they also have the capacity to record and profit off patron data, calling into question issues surrounding privacy, confidentiality, and ethics. 

Background

Beginning with Siri in 2011, VPAs have begun to appear in most facets of everyday life. As an application integrated into Apple's iPhone, Siri is a virtual persona designed to help users operate their cellphones. Other VPAs assist users in office environments, the automotive industry, the tech industry, and the home (Newnham, 2015). The largest players in the field are Amazon (Alexa), Apple (Siri), Google (Google Assistant), and Microsoft (Cortana). The market for VPAs, and their associated hardware, yielded 8.56 billion USD in 2017. 31% of VPA owners stated that they use their VPA every day, while 54% agreed that VPAs make their lives easier. VPAs are most commonly used to remind/schedule appointments, read out messages, and receive traffic news. 55% percent of VPA owners agreed the greatest benefit of VPAs is their hands-free application (Liu, 2017).

The VPA market is currently stratified in two ways: 1) as a feature of service; and 2) as the service itself. Until 2014, the VPA market was structured entirely as a feature of service, making product integration the only means for acquiring a VPA. Siri is an example of this instance, as she is an accessory to iPhones. In November 2014, Amazon released Alexa, redefining the market by offering an affordable, self-contained VPA that could be purchased like a vacuum or microwave (Lorenzetti, 2014). This type of market separation altered consumer control and opened access to VPA technology. Alexa sparked the possibility of VPAs in libraries, as they were now independent and portable.

VPAs are in tune with libraries because both curate, store, and retrieve information. Library patrons interrogate collections and library resources through an online catalog, and in some cases, through a physical finding aid. In order to assist library users locate information, staff members consume a substantial amount of resources, measured in both time and money. Reference librarians, on average, spend hundreds of hours a year answering questions that are either directional or superficial. VPAs offer the potential to answer these otherwise simple questions, freeing up librarians and paraprofessional staff for other projects or in-depth research assistance. Indeed, VPAs have the potential to enhance and complement library search experiences, but also the capacity to record and profit off patron data, calling into question issues surrounding privacy, confidentiality, and ethics.

What follows is a technical description of the technology's infrastructure, along with considerations of how these elements affect libraries and their patrons. The description ends by addressing the key challenges libraries might face by implementing VPAs and the future trends that might emerge alongside the technology. 

This timeline shows when various VPAs entered the market.

Architecture and Design

The following components are necessary for a successful and effective VPA system: a local search engine to handle geographic data, a data feed processing framework to accept real-time data streams from different providers, and a number of other proprietary development tools and languages. VPA programmers have struggled to create a seamless user experience from these components.

VPAs outsource information and are dependent on the hardware in which they reside. However, the position of the user's query determines what infrastructure is necessary to accomplish a given task. Here, position is defined in terms of proximity. Queries, or commands, can be local or distal, depending on the processing power or additional resources needed to perform a task. Distal queries are centrifugal—meaning they migrate from the device that houses the VPA in order to perform their task. If a VPA must use a network (i.e. go online), then the query is considered distal. VPAs that are dependent on a network also face the issue of bandwidth—the frequencies available and capable for data transmission (Sadun and Sande, 2014). The speed at which data is relayed is subject to the quantity of information being shared. VPAs use small amounts of bandwidth when transmitting or receiving queries to data centers. But data centers use varying amounts of bandwidth depending on the extensiveness of their processing. The more data transmitted, the more resources consumed and the more time a user must wait for a response. Local queries are processed within the confines of the VPA's hardware and are not dependent on a network. Directional library questions would be considered local, while superficial research questions would be distal.

When a VPA receives a distal query, it passes the request to the operating system, and is then transmitted over a network to a data center. The servers at the data center use voice-recognition algorithms to parse the language into a machine-readable command. The command is queried into the data center's database or searched on the web. The data server relays the answer/result back through a network to the operating system. Then, the operating system transmits the information to the original application or initiates the desired task. The final result is an answer or action. Local queries do not require external processing, because the information necessary to produce an answer/result is contained within the physical device. With local queries, the information moves from VPA application to the operating system to a local data cache. The process ends by moving back through the operating system to the VPA. These pipelines are not fixed mechanisms; rather, they are constructions that exist during this given, contestable moment. 

Benefits

Libraries have the potential to co-opt this mechanistic process by programming VPAs with the answers to directional questions. These answered questions can save library staff time and potentially expand library use due to the increase in the level of immediate service. Given that VPAs can be accessed either vocally or through a touch screen, they can increase access to those with disabilities and encourage an environment where conversation is rewarded. Those who struggle with library anxiety as a result of their inability to operate the online library catalog might experience more comfort talking to a VPA. While these few benefits may be useful in a library setting, they come at a price. Not only do VPAs make library experiences less personal, they also make them less safe, private, and confidential, as the current VPA market structure is dependent on user data for profits. If VPAs enter libraries, then libraries enter the financial market. 

Interoperability and Market Development

Due to their proprietary nature, VPAs are largely confined to local ecosystems, systems that are closed off from other systems (White, 2018). VPAs perform more tasks within company-specific applications and fewer actions within third party applications. Apple products work best with other Apple products, just as Alexa works best with Amazon products. These limits of interaction are purposeful, however. By entombing users within a restricted range of mobility, they become dependent (whether by force or indifference) on the services of the proprietary ecosystem. This limited interoperability allows VPAs to "guide" users toward company-specific purchases, which in turn, helps generate additional revenue for parent companies. This type of interoperability should be a concern for libraries, as patron data has the potential to be stored and manipulated by private companies. Depending on the company, it may be possible to delete recorded conversation; even so, it is uncertain if VPA data is ever truly erased from the permanence of the online forum. If libraries opt to use VPAs as they are constituted now, they risk becoming complicit in the horse-trading of patron data.

Some VPAs are also interoperable with extensions of themselves in different hardware, such as Siri and the Apple Watch. Here, Siri can interact with the user through multiple devices, aggregating and synchronizing material as needed, like access to email or reminders. Libraries can take advantage of this feature by developing a mobile application that allows users to ask directional questions through voice command on their cell phones. If sophisticated enough, such an application could act in conjunction with the online catalog, augmenting search with fast and accurate information retrieval. 

Standardization

VPAs are largely unstandardized. The majority of information-related control remains with the companies that manufacture and sell VPA devices. As VPAs are not seen as public goods or objects in need of regulation, no third-party organization is mediating their functions. Instead, consumers and private companies are the governing bodies that mold the standardization process.

Google, Apple, Amazon, and Microsoft are the leaders in the VPA field, and their personal assistant technologies interface almost entirely with other company-owned equipment, preventing the need to create standards that support interoperability or access across platforms. Instead, in-house standards are generated to complement economic efficiency and data security (Amazon, 2018; Apple, 2019; Microsoft, 2018). Here, companies standardize VPA software to specific operating systems to limit interoperability among competitors. This lack of congruency is purposeful, as it creates difference and diversity in the marketplace. As a result, there is a lack of unity in how VPAs function. Companies outline some specifics of VPA performance in their Terms of Use/Service, but largely do not reveal how decisions are made or what practices are standardized, including issues of privacy and confidentiality. 

Regulation

Policymakers and other regulatory commissions have failed to keep up with changing VPA technologies. As a result, the invisible hand of the market has neglected the economic and ethical practices of the VPA community in the United States. In recognition of this issue, the AI Now Institute at New York University releases an annual report that addresses the lack of institutional practices and accountability surrounding Artificial Intelligence (AI) technology. The 2018 iteration of this report states that: "regulation should include national laws that require strong oversight, clear limitations, and public transparency" (Whittaker, 2018). It continues by suggesting that: "[t]hird party experts should be able to audit and publish about key systems, and companies need to ensure that their AI infrastructures can be understood from 'nose to tail,' including their ultimate application and use." They further suggest that individual agencies or a collection of experts should comprise the regulatory body rather than a singular entity, as each subdivision invested in AI technology has its own, industry-specific regulatory practices. The AI Now Institute recognizes that the power between companies and the people whom they serve is asymmetrical, and their report suggests that: "[t]hese gaps are producing growing concern about bias, discrimination, due process, liability, and overall responsibility for harm."

The European Union (EU) has been more judicious than the United States in its efforts to limit the power that companies have in terms of data retention and use. The 2018 General Data Privacy Regulations (GDPR) standardized data practices for all companies in the EU, affecting VPA performance as data aggregators in the European market ("General Data Protection Regulation," 2018). This piece of legislation standardized the length of time that VPA-collected data can remain in data centers.

Children's Online Privacy Protection Act (COPPA) regulates the data that entities can collect on persons under the age of 13 ("Children's Online Privacy Protection Act," 1990). Children may reveal personal and identifying information to VPAs and subject libraries to COPPA violations, if recorded. As an online standard, COPPA makes VPA use in libraries impractical and potential sparks for litigation.

Regulation often includes limitations and standards of forced compliance, allowing for other companies to enter the playing field with fewer assumptions of risk. Companies such as Apple and Microsoft lobby against third-party regulation, given that structural intervention contrasts the goals of free-market capitalism (Pressman, 2018). The major players in the VPA field self-regulate through internal policies that outline the company's view on privacy, use, and access (Apple, 2018; Apple 2019; Amazon 2018; Microsoft 2018). The topic of security is pervasive in these documents, offering two topics for discussion: 1) user best practices for mitigating risk; and 2) the ways in which parent companies (and trusted third-party affiliates) utilize and protect information. These documents are explicit in stating that their policies extend to the breadth of the company's influence-including VPA technology. User privacy always has the potential to be exploited, and this risk undermines the foundation upon which the American Library Association built its Bill of Rights (American Library Association, 2006). Libraries must keep these aspects in mind when considering the collateral benefits of personal assistant technology. 

Key Challenges

VPAs face a variety of consumer-market challenges, especially considering that this market is built around an ecosystem of use. Servers have a finite ability to process requests for data, so institutions wrestle with regulating their digital assets in order to deliver seamless experiences for users. These social customs are in tune with cultural standards, suggesting that user psychology plays a large role in market decisions (Blair, 2014).

Privacy is also a concern for users, as VPAs store query data in order to generate targeted ads and batten off user requests. Patrons run the risk of forfeiting rich amounts of personal data to private companies in exchange for a convenient, technological experience. Libraries run the risk of violating COPPA if their VPA devices store information produced by children.

On a local level, VPAs face challenges with their command features. VPAs have "action words" that activate their functions, and cannot distinguish between operative words and casual speech (Chung et al, 2017). Natural language parsing algorithms are unreliable and have difficulty operating alongside a VPA's digital architecture. The varying levels of infrastructure make VPAs fragile, as a breakdown at any point in the architecture results in a failure to execute the query or command. The cost in maintaining the various components is unmeasured and not equally distributed between users and those who provide the maintenance. Exactly how these systems interoperate is largely unknown, as most VPA algorithms are copyrighted and their methods patented. These proprietary safeguards prevent VPAs from becoming birdfeeders of cultural property, but also subject companies to scrutiny and allegations of foul play. Until VPAs can exist without the need to absorb user data, they have no purpose in libraries. 

Future Trends

The popularity of virtual assistants is growing rapidly, and it is projected that 1.8 billion people will be using VPAs worldwide by 2021 (Lazarevich, 2017). However, in order to secure a place in libraries, VPAs need to undergo some improvements. To support institutional needs, VPAs must be able to process multiple requests simultaneously. They must also parse language more rapidly, accurately, and intelligently. Most importantly, VPAs must protect patron data before they can enter public and civic sectors, as these technologies impact users and their levels of comfort. Ultimately, VPAs need to become more versatile, pervasive, and human in their processing.

Libraries are unlikely to experience successful VPA integration in the immediate future. And although our current technology is able to create a data-secure personal assistant, the capital to finance such a device will be difficult to secure, as the return on investment is uncertain. Libraries do not turn profits; they produce information. Until these two facets can be reconciled in the eyes of investors and library administrators, VPAs will continue to be the topics of unresolved conversations. 

© Jeremy Zimmett 2021 | All Rights Reserved
Powered by Webnode
Create your website for free! This website was made with Webnode. Create your own for free today! Get started