Hong Kong’s Low-Altitude Economy departs amid challenges

Remark: This news story is part of our experimental project using AI technology to transform the financial journalism practicum. However, all content has been reviewed by humans (our student journalist team) to maintain our high standards of accuracy and fairness before publication. Details about our AI usage can be found in our 'Making of' article.

Reflecting on their impressive experiences, Tiffany Wong, project director at Hong Kong-based drone performance company OWOWWW, recalled a major event in late 2023.

Their team orchestrated a dazzling display featuring 1,000 units for a luxury brand’s fashion showcase at the Cultural Centre Plaza—a first for the city at that scale and spent only ten days preceding the flying approval.

Wong explained that while drone flight approvals were initially challenging due to regulatory uncertainty, the process has become more efficient, reflecting growing recognition of the low-altitude economy. 

Low-altitude economy(LAE), referring “to economic activities in airspace below 1,000 metres, presents a wide array of application scenarios including rescue, surveys and delivery of goods and passengers,” according to the paper prepared by the Legislative Council Secretariat for discussion last year. 

In the 2024 Policy Address, in order to “unlock the low-altitude airspace as a new production factor”, the government set up a Low-altitude Economy Development Working Group led by the Deputy Secretary of Finance, and the group soon announced the establishment of the LAE Regulatory Sandbox, a safe zone for trial LAE concepts before full implementation, last November.

This year, on Mar. 20th, 2025, the first batch of the Sandbox including 38 pilot projects were announced and launched, covering emergency rescue, logistics, inspection, surveillance and low-altitude infrastructure, which will conduct scenario simulations through implementation to collect data and experience.

Having successfully organised nine drone shows last year with a record of nearly 1,000 drones per show, OWOWWW supported the government initiative by joining the Sandbox initiative as one of its 38 pilot projects. 

Though a promising future lies ahead for the industry, technical obstacles such as drones’ adaptations to extreme weather, consistency between software and hardware on each unit, and the unrealized benefits brought by policies to LAE SMEs are still affecting the industry’s full-space growth. 

As of October 2024, at least three drone performances in Hong Kong had been temporarily canceled due to adverse weather conditions (rain, wind speed, typhoons, etc.) and ionospheric disturbances.

Wong explained that the ionosphere can affect a drone's positioning. As it contains charged particles that refract or delay satellite signals,  the ionosphere can cause drones to receive inaccurate coordinates and interfere with the communication channel between the sky and ground control stations. Ionospheric activity also alters local magnetic fields, causing compass malfunctions and navigation drift. 

“If hundreds or thousands of drones do not know each other's location, it could lead to serious collisions," said Wong.

Now benefiting from Sandbox, OWOWWW is able to establish a safety testing base at Cyberport, where restrictions on flying drones – due to Hong Kong’s urban conditions with densely populated high-rise buildings – are erased, said Mechelle Lee, operation manager of OWOWWW.

Lee added that using Cyberport as a test base, the next step is not only to find ways to minimise the interference of weather factors on drone performances but also to explore expanding the number of units in each show.

Wong points out that drone performances in mainland China have developed quite maturely, with teams have already toured around the world to demonstrate their projects.  "Hong Kong's engagement within this industry is still relatively new, but we hope to generate more creative ideas to catch up," she added.

Despite the government’s promotion of sandbox initiatives, Ricardo Kwong, senior manager of OnTheGo Limited, a Hong Kong drone distributor and training organisation, questioned the lack of tangible financial support for SMEs. 

"When working in the low-altitude economy, many of our ideas tend to be quite imaginative and unconstrained," said Kwong. He stressed that small businesses need clear and explicit guidance on how to structure their proposals and apply for potential funding, especially to ease the financial pressure even before a project officially begins. 

Sky Yeung, chairman of Drone Association Hong Kong (China), hopes Hong Kong can learn from other regions’ successful experiences in developing LAE.

“We want to introduce proven cases from mainland China to Hong Kong for pilot testing. This is not only to demonstrate the feasibility of implementing LAE concepts, but also to build confidence among the government and citizens,” said Yeung. 

For example, Meituan’s drone delivery service in Shenzhen piloted daily deliveries. Citizens can order food deliveries  through mobile phones and receive the products much faster than normal deliveries done by humans. 

As of the end of 2022, Meituan’s drone trial routes in Shenzhen had been launched in five business districts, covering 18 communities and office buildings, and can provide delivery services to nearly 20,000 households. The deliveries included  a wide range of products such as food, beauty items, fast-moving consumer goods, supermarket products, electronics and more.

At the regulatory sandbox launch ceremony on March 20, Meituan’s drone completed its first low-altitude logistics delivery mission in Hong Kong. The countdown to the opening of the drone delivery route from Ma On Shan Park to the Pak Shek Kok Corridor in Hong Kong Science Park has begun.

“More importantly, within the entire (low-altitude economy) industry chain, we aim to leverage Hong Kong’s strength as a financial hub,”  said Chan Siu-hung, a member of the Legislative Council and Greater Bay Area Low-Altitude Economy Alliance. He added that Hong Kong could serve as a springboard and gateway port, working alongside relevant Mainland enterprises to expand the LAE onto the global stage.

However, while acknowledging existing R&D capabilities in areas like drone delivery and opportunities brought by the Great Bay area, Wong believes that life-saving applications are an area worth further exploration.  

She pointed out that the value of drone delivery remains unclear if recipients are still required to leave their homes to collect packages. Comparatively, greater investment could be directed toward life-saving applications such as emergency rescue, firefighting support, and medical supply delivery.

AI in the Newsroom: The Making of This “Low-Altitude Economy” Story

Regarding this special issue, our reporting team explores how to incorporate generative AI throughout the entire process of news production. For the “low-altitude economy” story, we used different models of generative AI to assist in various tasks such as generating story ideas, optimizing interview processes, and refining content.

Gen AI assisted in generating ideas and interview question design

The answer prompted us to present drones more broadly in our articles, beyond just food delivery. The generative AI platform mentioned the government regulatory Sandbox, as well as possible challenges such as weather conditions and life-related applications, including drone performances and medical deliveries.  Subsequently, with the help of a profile, OWOWW – a drone-performing company being listed in the newly launched Sandbox scheme – we not only revealed the logic behind drone performances but also gained insights regarding the developing future of the industry from the frontier. 

Besides the application, the segment “Challenges and Considerations for Hong Kong” is a bonus gain, which equipped us with the aspect of weather intervention, when OWOWW mentions the effects brought by the ionosphere on drones’ GPS signals during the interview. We ultimately omitted some regulatory framework content to maintain readability. 

During our interviews, we found that interviewees provided crucial details that AI could not. For example, when asked about government support for LAE businesses, AI only presented a positive and general comment by saying “the government is committed to…” However, in our interviews, OnTheGo Limited, a drone training company, said that they felt it hadn’t helped them in substance due to vague guidelines and allocations. This contradiction alerts us, making us dedicate more efforts in digging out the true story behind words shouted publicly. 

Similar patterns were observed during our interactions with AI when it assists in generating possible interview questions. 

Before interviewing OnTheGo Limited, we asked the AI for an interview outline, and it generated over 40 questions across five categories within seconds, which was broad and quick. To avoid distraction from our focus on the article, we selected key angles and expanded on the company’s growth in Hong Kong’s LAE market and Greater Bay Area opportunities, not limited to the field of drone training.

But in practice, we didn’t strictly follow the AI-assisted outline. In addition to the supplementary questions on government funding mentioned above, we also adjust questions in real time based on respondents’ responses. For example, when Ricardo Kwong (OnTheGo Limited) mentioned their involvement in Hong Kong’s Low-Altitude Economic Sandbox Program— unexpected news— we pivoted to ask unplanned questions about the policy, enriching our coverage.

In reflecting on our experience, we found that generative AI’s most valuable contributions were its fast response times and ability to suggest diverse angles when answering our prompts. However, we observed limitations in the AI’s ability to generate probing critical questions, and not being able to adapt dynamically to interview situations by generating relevant follow-up questions as the conversation evolved. While we could utilise its strength, the key lies in the reporter’s awareness of news value and adaptability to situational changes. 

AI-assisted writing and production process

For news writing

We incorporated  AI to refine our writing and observed that AI would add narrative language to enhance context, while also introducing more precise vocabulary to make the expression clearer. (See examples in the screenshots below)

However, we encountered AI hallucination problems, where the system would misinterpret our original meanings and add irrelevant details which contradict journalism standards of being truthful and fair. The following information shown in the above photo contains at least three inaccuracies about OWOWW: “a Shenzhen-based company,” it “benefits from Guangdong Province,” and “mainland China’s policies.” This information appears to be irrelevant without any supporting evidence. Besides, AI added opinion sentences at the end of the first paragraph (“This breakthrough demonstrated how bureaucratic barriers crumble…”), which was too subjective and contradicted the principle of fair reporting. 

To compare different AI approaches, we then sent the same request to a different AI model, Gemini 2.0, a language model developed by Google. We found that the generated results stayed closer to our original text. (See examples in the screenshots below)  

While the AI improved the fluency of our  article, we felt that it sometimes elevated the language in ways that made it sound unnatural. For example, terms like “deploy” and “execute” didn’t seem like natural word choices when describing drone performance. “Continued success,” “utilising close to a thousand,” doesn’t sound neutral in a news article. Additionally, the sentence structures are relatively fixed (with a repetitive gerund in nearly half of the sentences), lacking the diversified style typical of human writers. 

AI-assisted for video shooting

We used AI to help plan our camera angles before taking the shot to avoid the last-minute panic. 

During implementation, we selected and modified the appropriate scenes first, eliminating over-idealised scenes and adjusting their suggestions based on real shooting conditions. For example, Shot 2, “Close-up of a robotic arm loading,” was abandoned due to the company's prohibition on filming in logistics zones. And Shot 5 was replaced by the photo of a drone flying towards the takeaway counter. 

After repeatedly checking the shot list generated by AI, we were confident that all preparations had been well settled. We turned on the camera and placed an order for a drone delivery takeout. Minutes after, a message popped up reminding me that the food would be delivered soon. The adjusting window is so narrow, as we didn’t know the direction of the drone’s flight, which made us look around in a panic, trying to catch the suggested shot of “tracking.” This unexpected arrival made us forget the preset images provided byAI, as we zoomed in and out indiscriminately. The hand holding the camera shook violently from nervousness, significantly affecting the quality of the images. 

Through this experience, we reflected on how AI, while helpful in offering detailed, multi-angle insights, provided suggestions that we found somewhat impractical for our specific shooting situation. Additionally, relying heavily on the AI-generated shot list seemed to affect our situational awareness during the actual shoot, as we found ourselves less able to adapt quickly when unexpected challenges arose. This made us consider how we might better balance AI assistance with maintaining flexibility and responsiveness in future.