In the Age of AI, to Whom is Power Shifting?

Worse still, the system’s decision-making capabilities fall short of human intuition, ultimately eroding Bianlifeng’s profitability. Stripped of store managers’ frontline experience, the company has suffered basic errors in restocking fast-moving items like bread and beverages. The algorithm fails to accurately anticipate consumer demand or meaningfully improve inventory turnover rate. According to franchisees, the rigid system routinely results in overstocked hot food, leading to significant waste. Consequently, what appears highly intelligent to outsiders has earned the derogatory nickname “the idiot” among staff.

Zhuang Chenchao readily admits that the algorithmic system is prone to errors. Yet, as a staunch advocate of technological progress, he argues the answer does not lie in bolstering managers’ decision-making authority or relying on human intuition to patch the system’s flaws. Instead, he believes the path forward is to keep refining the algorithms, fortify the machine’s “brain”, and enable it to self-correct through continuous learning.

In short, he maintains that machines only appear “foolish” because the technology has not yet matured.

III. Humans Have Their Place

Yet human insight remains invaluable. Seasoned Bianlifeng managers possess a wealth of practical strategies for running their shops. They intuitively understand how to adapt product ranges and shelf layouts in response to shifting seasons, temperature fluctuations, weather patterns, and holidays.

Yet the company’s system maintains a stranglehold on decision-making, providing not even a formal channel to ‘deign’ to consider a manager’s input.

In truth, from the earliest days of mechanisation, human roles have almost always involved tidying up the mess left by machines. During the dawn of the Industrial Revolution, machinery boasted overwhelming advantages in power and mechanical drive. Highly efficient as they were, they were nowhere near as “intelligent” as modern systems. They invariably required workers to step in, adapting on the fly to handle unexpected glitches.

Once mechanised spinning was introduced, manual spinning became obsolete, yet female workers still had to stand ready to splice broken threads the moment they snapped. Victorian-era textile mills routinely employed child labour to crawl beneath moving machinery and sweep away waste cotton.

Contemporary AI may appear highly intelligent, but the unglamorous task of error-correction and crisis management still falls squarely on human shoulders. Even Baidu’s Apollo Go autonomous taxis require safety officers to manage emergencies, proving that when it truly counts, human intervention remains indispensable.

IV. Can Algorithms Truly Account for Workers?

Meituan also claims that the “real-time intelligent dispatch system” it developed for food delivery services is exceptionally powerful.

The data highlight its computational power: during peak order volumes, it can execute 2.9 billion route calculations per hour. Within tens of milliseconds, it computes the optimal matching of orders, riders, and routes, while simultaneously accounting for dozens of variables such as weather conditions, traffic, rider availability, and restaurant preparation speeds.

According to sources within Meituan, the “Real-Time Intelligent Dispatch System” adjusts delivery windows based not only on real-time variables—such as the locations of riders, customers, and merchants, along with weather conditions—but also on historical data from previous deliveries.

When delivery riders run red lights or cycle against the flow of traffic on a given route for fear of exceeding their deadline, the time they save at the cost of their own safety is absorbed by the algorithm, establishing a new benchmark. In this continuous loop, delivery times for every route are ultimately squeezed to the absolute limit.

Thus, the rise in efficiency is not simply because the system has become better at “calculating”, but more because it can monitor and control countless delivery riders, absorbing their desperate, self-driven “calculations” into the system.

The cost of this heightened efficiency, however, is the riders’ lives and safety.

According to a 2023 survey, during peak delivery hours at an intersection in Beijing, 53 riders passed through in half an hour; of these, 37 were cycling against traffic, cutting diagonally across roads, or running red lights. Shanghai has also published stark figures: in the first half of 2017, a delivery rider was killed or injured every 2.5 days on average. In 2023, the total number of traffic offences committed by Meituan delivery riders in Shanghai reached as high as 6,500 in a single week.

I have experienced this myself: on a rainy night, anxious about running late, I slammed heavily into the central guardrail. My electric bike was wrecked, but the order could not wait. I had no choice but to complete the delivery on a shared bicycle.

In retrospect, I have often imagined: even at the very least, the platform could simply allow riders to log a traffic accident to the system immediately and grant them a delivery extension. At the minimum, I would not have been left in such a predicament, and an injured rider would have had the breathing room to deal with the aftermath.

The platform has invested heavily in optimising its system to slash delivery times and maximise profits; is it really that difficult to leave a “back door” open for riders caught up in accidents? Is the inability to engineer a flawless system down to a lack of technical expertise, or is it the designers’ own notions of power dictating the outcome?

Public discourse surrounding rider accidents has persisted for years, yet we have seen no tangible improvements to the system. Whether delivery platforms are genuinely willing to take action speaks for itself.

We can never be certain whether “technical constraints” are merely a pretext for refusing to change. Ultimately, the power to shape algorithmic rules rests entirely with the corporations, while labourers remain perpetually locked out of the decision-making process.