Multilingual Chatbot

Multilingual Chatbot aims to address the knowledge gap for migrant workers whose native language is not English. A Multilingual Chatbot is an agent that can answer and support migrant workers in doing their job of driving and delivering products for Uber.

Project overview

The product:

Conversation design

Human-AI Interaction

UI Design

UX Design

Engagement

A multilingual chatbot is resource-intensive and an upfront investment. For instance, content translation and hiring native speakers to train and test the bot. In addition, an AI-powered chatbot is a black box, which makes it hard to predict how the chatbot will respond to user prompts.

If you have used Uber, you will notice some of them don't speak English fluently. However, the percentage of the underrepresented group is low in the sharing economy. Language barriers and technical terms can prevent non-English speakers from entering the field.

Project duration

10/30/2024 - 12/04/2024

The problem

The goal

The goal of the project is to support future Uber drivers or delivers to get on board quickly and smoothly with a multilingual chatbot as an agent.

The challenge

Research

Uber Driver Demographic

Two research papers analyze Uber's ride-share economy in the US, ACM, and NBER. Both papers identify a race disparity in Uber's driver demographic of 60% and 40% respectively. Surprisingly, the two research papers agree that Uber's driver demographic tends to be in their 30 or younger and are working toward a higher education.

Regional diversity in autonomy and work research paper highlights a lack of proficiency in English did impact driver work in the Orlando/Kissimmee area. Drivers were reluctant to call the riders to locate their pick-up location due to the language barrier. The paper indicates the primary sources of education for Uber and Lyft drivers are online forums, which are conducted in English only (Intelligence & Autonomy).

Analyzing Uber's Ride-sharing Economy – https://dl.acm.org/doi/epdf/10.1145/3041021.3054194

Language barrier

Driving conditions

Regional differences illustrate how the possibility of automation runs afoul of the realities of regulation and cultural norms (Regional Diversity in Autonomy and Work). Some tips and tricks for Uber drivers not to get tickets at the airport are available in the community forum that is only accessible in English.

Global presence

Uber serves approximately 90 countries as of January 2023 (World Population Review).

According to Inovbiz: Jurnal Inovasi Bisnis 9 (2021) 124-132, Uber's failure in the Southeast Asian Market is due to the following factors:

Payment preferences (cash payment is the most preferred in the region due poor credit card system)
Infrastructure problems
Local competition

Uber's failures in Southeast Asia highlight an important knowledge gap between the company and its drivers. Although non-English speakers can get on board within a week because of the automation process (Regional Diversity in Autonomy and Work), there are challenges that Uber drivers should know, especially if they are new to the country.

Design

Conversation agent

Dr. Sandeep Kaur Kuttal conducted research on Human AI Interaction in Programming with conversational agents that could potentially make the software industry more inclusive in terms of expertise, gender, and race (Redefining Human AI Interaction in Programming). I want to apply the same concept of pair programming to Uber's driver case study to narrow the knowledge gap and promote inclusivity. Most importantly, it aligns with Uber's mission of achieving diversity and inclusivity as international expansion.

Sketches

Chatbot UI Sketch for Uber Driver Website

Chatbot UI Sketch for Uber Driver Help Website

Chatbot UI Sketch for Uber Driver Website

Chatbot UI Sketch for Uber Driver Help Website

Low-fidelity wireframe

Testing

Usability testing is not available due to human research nature. The research subjects to IRB research approval and protocol, which I don't have the budget or time to pursue.

More information on IRB Human Subject Research.

However, below is the testing template that helps refine the chatbot training via prompting in offered languages if it is possible to do so in the future.