In the digital age, governments are no longer the only entities collecting data on citizens.
More alarmingly, private companies have become more efficient than administrations at gathering information. Traditional governmental methods, telephone surveys, forms and in-person interviews, remain the benchmark for accuracy, yet their inefficiencies are increasingly apparent. In a typical OECD country, over 3,000 employees work in the National Statistics Department, doing the heavy lifting of government data collection.
Costly cycles that span years from inception to completion stand in stark contrast to the real-time analytics deployed by technology giants, whose platforms monitor millions of users daily, underscoring the need for faster, leaner techniques to complement traditional collection.
Meta, the company behind Facebook and Instagram, has collected data on 77% of all internet users, around 3.9 billion people active on at least one of its platforms (Statista, Q3 2023).
The issue is not limited to methodology. National statistics departments struggle to provide the granular, timely data modern policymaking requires. In emerging economies the data itself is often incomplete: the informal economy, which accounts for roughly 15% of the OECD's GDP, remains largely uncharted by conventional methods.
The Middle East mirrors this global trend. Data availability across the MENA region is 44% lower than the global average and 54% lower than the G20, and only about half of MENA countries are up to date with their health, labour-force and consumption surveys (Data & Policy, 2022).
Open Data Barometer, global report
At the same time, policymaking is becoming more granular. As governments design dynamic policies tailored to individual citizens, the need for timely, detailed insight becomes critical. To navigate this terrain, governments must adopt big-data and AI-driven analytics.
Transforming satellite imagery into economic insight
In assessing a region's economic health, the conventional barometer has been Gross Regional Product (GRP): GDP on a more localised scale. Where robust data is scarce, Whiteshield has turned to satellite imagery to measure regional economic performance. The process breaks satellite images into smaller, manageable tiles, which are then analysed by AI.
The current state of the art is a class of machine learning known as Convolutional Neural Networks (CNNs). These require humans to first prepare extensive, hand-labelled example images to teach the system to recognise features, a resource-intensive process, particularly across the unique landscapes of the Middle East.
The power of language models in regional analysis
Whiteshield's AI Economic Analysis Tool instead uses large language models such as GPT-4 to read satellite images, an approach prized for its versatility across socio-economic research, in contrast with expert systems built for a single task.
In a study of 72 detailed satellite images of Dubai, the two methods were compared. A CNN trained on human-labelled images produced an average error of 7.32% after several days of training. A newer approach using the AI tool and GPT-4, with no pre-labelled images, running in minutes, reached an 11.9% error rate. Slightly less accurate overall, yet markedly better at identifying buildings and roads, the elements most central to economic assessment.
Simplifying complexity
What does this mean for policymakers? The tool compresses an intricate analytical process into a matter of days. Asked to assess agriculture intensity, a task that would traditionally demand a model trained specifically for desert crops: Whiteshield loaded the dataset, defined a scoring formula with stakeholders, and delivered a precise sectoral estimate across 4,000 km² in 48 hours.
40% × Crops score + 40% × Built-up-area score + 20% × Greenhouse presence
Continued research points to an optimistic future in which LLMs like GPT-4 transform how policy is developed, with alternative sources supplementing, and perhaps replacing, the survey-based approach. The AI Economics Unit is already extending the tool with cell-tower activation, night-light intensity and real-estate signals, to give policymakers near-real-time analysis specific to their sector and at an unprecedented level of detail.
For more information about our AI solutions, contact the AI Economics Unit.
