[THE VIEW]Baedal Minjok's AI Dispatch System Ensuring Rider Autonomy

Contrasting With Uber's Mandatory Dispatch System
The Key Lies in Rider-AI Collaboration

Baedal Minjok, the number one food delivery platform in the industry, introduced the ‘AI Recommended Dispatch’ in 2020, which automatically assigns the most suitable orders by considering the rider’s route and the characteristics of the ordered food using artificial intelligence (AI). Until then, riders had to competitively grab calls even while on the move to secure delivery jobs. Additionally, when performing multiple deliveries, it was difficult to know the priority and optimal route among several pickup locations, which sometimes caused delays in delivery times. The AI Recommended Dispatch was expected to improve both delivery safety and efficiency by allowing riders to focus solely on driving through automatic call assignment and by providing the optimal delivery route.

According to an internal report from Baedal Minjok, compared to January 2019, just before the introduction of AI Recommended Dispatch, the accident rate in January 2020 decreased by 47%, and delivery times were reduced by 15%. This means that food was delivered faster and more safely through AI Recommended Dispatch. However, there are also strong voices of rejection and criticism from riders regarding the AI Recommended Dispatch. Some riders complain that it does not always recommend the optimal route, and others continuously express dissatisfaction, saying that although it is somewhat useful, it is still less reliable than human selection.

How can the AI Recommended Dispatch system become more trustworthy? The starting point lies in advancing and refining the AI Recommended Dispatch technology. It is necessary to analyze delivery failures, where deliveries took longer than the predicted time by AI, and to continuously improve the algorithm by training the AI with this data. Securing data that reduces the gap with the field is also important. In particular, it is necessary to use more accurate data by narrowing the gap between the straight-line distance from the restaurant to the customer’s destination and the actual distance traveled by riders.

More importantly, it is essential to secure the autonomy of riders. The previous dispatch method allowed riders to decide what they wanted to do, whereas AI Recommended Dispatch reduces the freedom of choice by making AI, not humans, the decision-maker. Rather than forcing AI Recommended Dispatch, it should be introduced as a way for riders to utilize and collaborate with AI. Sufficient guidance and prior education about AI Recommended Dispatch are also necessary. If AI selects jobs instead of me, I need to be convinced that it can do better and understand why. Transparent sharing of explanations about how the AI algorithm operates will help build trust.

Baedal Minjok seems to already know the answer. By allowing riders to turn the AI Recommended Dispatch mode on and off, it respects the autonomy of riders. This contrasts with ride-sharing platform Uber, which forces all drivers to use AI Recommended Dispatch. Baedal Minjok also analyzes and discloses the impact of AI Recommended Dispatch on reducing accident rates.

There is no perfect AI system. For an AI automatic dispatch system to catch both effectiveness and safety, it is necessary to listen to riders’ complaints and reflect them in the system, as well as to continuously strive to develop the AI system more precisely. This is the key element in completing an AI system.

Kyung Na-kyung, Professor, Department of Computer Science, National University of Singapore