Machine Learning at Amazon: Software Engineer Preparation; Adapting Voice-Activated Personal Assistants for Disability: Balancing Privacy and Independence
Sohan, Ian, School of Engineering and Applied Science, University of Virginia
Graham, Daniel, EN-Comp Science Dept, University of Virginia
Baritaud, Catherine, EN-Engineering and Society, University of Virginia
Although the ability of voice-activated personal assistants (VAPA) to provide a pseudo-infinite amount of knowledge at a simple voice request is desirable, the modern technology drastically escalated the societal concern for privacy. The technical topic addresses the ethical concerns surrounding smart speakers through the perspective of new software engineers exposed to the development of Amazon Alexa. The analysis provides potential improvements of university curriculums to better prepare software engineers for ethical challenges faced when developing and managing modern technology. The science, technology, and society (STS) topic provides a framework through which to analyze the social context of VAPA within the disabled community and the tradeoff faced between privacy and independence. For individuals living with disability, VAPA can enable an independent lifestyle; therefore, many individuals must sacrifice their concern for privacy in order to maintain this lifestyle. The technical topic and tightly coupled STS topic offer insight into the development and possible solutions to the ethical dilemmas within the realm of VAPA technology.
The technical report outlines personal internship experience at Amazon Alexa from the perspective of an aspiring software engineer studying at the University of Virginia. The intent of the technical report is to offer insight into the continuously evolving industry of computer science and provide feedback to the engineering curriculum at the University of Virginia. The report specifically details the development of a tool that migrates machine learning models out of Amazon Alexa’s secured internal employee environment and into production. The model migration system involved the steps of loading models into the software, testing the models for accuracy, validating the models, and exporting the models out of the internal employee system. The process integrated the previously written code of multiple machine learning teams which ultimately led to a generalizable system.
The revamped migration system successfully manipulated S3 pathways, Amazon’s cloud data storage service, to enable for the simple solution of an otherwise complex problem. The launch of this service aided in the productivity and efficiency of all Amazon Alexa machine learning teams. The purposeful complexity of the system exists in order to protect Amazon Alexa information from malicious attacks and inappropriate access. The exposure to the delicate balance of increased convenience in the modern world while attempting to maintain personal privacy highlights the need for computer science education that evolves with the industry. Incorporating opportunities into university curriculums to expose students to industry problems and ethical dilemmas aids in continuous improvement within education.
The STS topic analyzes the possible solutions that would negate the tradeoff between privacy and independence faced in the disabled community in regard to voice-activated personal assistants. The investigation focused on updating outdated legislation such as the Health Insurance Portability and Accountability Act (HIPAA) and the Medical Device Regulation Act to allow companies to produce medical VAPA technology fully protected by medical privacy law. Pinch and Bijker’s Social Construction of Technology theory was used to create a framework outlining the necessary social groups required in the mitigation of disabling barriers and tradeoffs introduced by VAPA technology. The framework was developed through a collection of technical journal articles, STS journal articles, and government laws and regulations relating to the protection of medical privacy and enforcement of medical device safety.
Although the responsibility of mitigating the discussed tradeoff primarily belongs to both government and Big Tech, the inability for government legislation to maintain the pace of technological development is the primary problem. For example, the most recent amendment to HIPAA happened prior to the first mainstream smart speaker, Amazon Alexa. Adjustments to HIPAA to specifically address the healthcare use cases of VAPA technology and adjustments to the Medical Device Regulation Act to incentivize the creation of VAPA devices that can be prescribed and used by medical professionals. Big Tech, the companies with financial incentive from the technology, also must use their influence to encourage legislative change and development practices.
All communities benefit from ethical engineering at the forefront of the technological development that shapes society. Responsible engineering leadership and proper government legislation mitigates ethical dilemmas such as the tradeoff between privacy and independence faced in the disabled community in regard to VAPA technology. Maintaining both societal and technological improvement requires consistent effort to solve ethical dilemmas.
BS (Bachelor of Science)
Social Construction of Technology, Amazon Alexa, HIPAA, Voice-Activated Personal Assistants
School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisor: Daniel Graham
STS Advisor: Catherine Baritaud
All rights reserved (no additional license for public reuse)