Enhancing Linguist Productivity through Session-Logging and Data Pipelines; Ethical Challenges of Emotionally Intelligent AI in Humanoid Robots 5 views

Technical Report During a software engineering internship at Meta Platforms Inc. in the summer of 2025, the core challenge was that Meta's Global Experience and Localization team had no reliable way to measure how productive their translation workers were. The old system asked workers to self-report their own hours, stored data in scattered tables across a large company database, and required up to a month of approvals before anyone could access new data. These bottlenecks made it nearly impossible to catch inefficiencies or make informed decisions about resource allocation. To fix this, a new session-logging system was built using an internally developed framework that gave the team direct control over their data. The system captured roughly 840,000 translation events per day across all major tools, then pushed that data through a three-stage pipeline that cleaned, validated, and grouped events into meaningful work sessions. Everything surfaced through a custom dashboard where managers could see performance broken down by vendor, language, and tool in real time. The outcome was a 17 percent improvement in metric accuracy over the old system, along with a much faster loop between spotting a problem and being able to act on it. STS Paper This paper looks at what happens when AI systems are designed to simulate human emotion and then placed in front of people who naturally respond to emotional cues as if they are real. The central question is whether transparency alone can solve the ethical problems that come with deploying emotionally intelligent robots, especially when the users are vulnerable, like elderly residents in care homes or young people who have grown up surrounded by AI-mediated communication. Through a review of five peer-reviewed papers from 2020 to 2025, the paper argues that simply telling someone they are talking to a machine is not enough. Humans form genuine bonds with systems that feel warm and responsive, and knowing something is artificial does not stop those bonds from forming. The paper finds that the problem goes deeper than any single fix can reach. Making AI more transparent can actually lower how much users trust it, because a system that feels more mechanical also feels less capable. At the same time, none of the ethical frameworks researchers have proposed fully account for the business pressures that push companies to deploy these systems fast and make them as engaging as possible. My paper concludes that the only real path forward involves binding regulation that holds organizations accountable for psychological harms, paired with a shift in how engineers are trained to think about the people on the other side of their design choices.

BS (Bachelor of Science)

emotional AI; humanoids; human AI interactions

School of Engineering and Applied Science Bachelor of Science in Computer Science Technical Advisor: Bryan Ansong STS Advisor: Karina Ripley

English

All rights reserved by the author (no additional license for public reuse)

2026-05-08