[AI Revolution](57) Over 20,000 Socar Vehicles Nationwide Managed Entirely by AI from Car Wash to Maintenance

Anyone who has taken a long-distance trip or business trip has probably used 'Socar' at least once. With around 20,000 Socar vehicles, it is used as a means of transportation nationwide, including Ulleungdo. Every second, countless people make reservations, use, and return vehicles through the Socar platform without any issues. This is possible because Socar’s unique artificial intelligence (AI) is deeply integrated into the platform.

Thanks to AI, One Employee Manages 1,000 Vehicles

Socar has a vision of “a world where everyone moves freely and happily.” The means to realize this is AI. Socar has organized and operates a dedicated AI team within its Data Business Division. This team conducts comprehensive AI research and performs machine learning operations (MLops) to improve the performance of its proprietary AI models through data.

AI has brought significant efficiency to Socar’s services. To provide vehicles to users, it is necessary to frequently check whether the vehicles have any issues. The condition of the vehicle and whether it is parked correctly must be verified based on photos sent by users when they return the vehicles. It is impossible for employees to check all 20,000 vehicles nationwide one by one. In fact, in 2014, one employee managed about 50 vehicles.

However, just a few years later, the situation completely changed with the introduction of AI into the service. Currently, one employee manages more than 1,000 vehicles. AI reviews photos uploaded by customers to identify dirty vehicles and issues car wash orders, and it also oversees vehicle accidents and damaged areas. AI tracks who caused accidents and handles repair requests as well. Socar expects that by 2025, one employee will be able to manage more than 2,000 vehicles.

AI technology has also helped reduce variable costs related to car washes and vehicle maintenance. In the second quarter of this year, the ratio of variable costs to sales was 45.9%, down 13.7 percentage points from 59.6% in the same period last year. Previously, partner car wash companies performed regular weekly washes, but now AI determines whether a vehicle is dirty and initiates car washes accordingly, saving costs.

Training and Application Process of Socar Mobility Domain Multimodal Large-scale AI

Transforming into an AI Specialist Company

Socar plans to achieve 20-30% annual productivity improvements through the advancement of AI technology. To this end, it is independently researching and developing large language models (LLM) to build conversational AI. The plan is to create an AI model that understands Socar best, based on data obtained through platform operations.

Socar has launched the ‘AI Contact Center’ project to replace call center personnel, aiming to introduce it within the year. For this, Socar’s developed STT (Speech to Text) voice recognition model will be used. Customer inquiries mostly occur in noisy environments such as roads or parking lots. This voice recognition model is expected to convert conversations in noisy situations with high accuracy, enabling quick and precise responses to inquiries arising during vehicle use.

Socar’s AI capabilities have been recognized. In May, the Socar AI team presented two research papers at the workshop of ICLR 2023, the world’s most prestigious conference in deep learning. At this event, the AI team introduced cases of efficiently training data in environments with limited data resources.

Also, in June, Socar hosted a workshop at CVPR 2023, the world’s top conference in computer vision, and held the first-ever mega-scale AI global challenge in the mobility domain.

This content was produced with the assistance of AI translation services.