Unequal Distribution In South of AI Benefits and Risks

In the rapidly advancing world of artificial intelligence, the distribution of benefits and risks remains a contentious issue. The environmental costs associated with AI development are a significant concern, especially given the exponential increase in capabilities and resources allocated to frontier AI models in recent years. As AI continues to evolve, this trend is expected to persist, raising critical questions about who stands to gain and who may suffer as a result.

The Compute and Data Storage Conundrum

At the heart of AI development are compute and data storage, two of the most resource-intensive inputs. The global distribution of these resources is uneven, leading to material risks related to the unequal sharing of AI’s harms and benefits. Developing countries in the global South are particularly vulnerable, as they may miss out on the economic advantages of AI due to substantial capital costs while facing environmental degradation and labor disruption.

More compute typically leads to better predictions and more robust machine learning systems. Although algorithmic efficiency is crucial, increasing compute has proven to be a reliable strategy for scaling AI models. This allows models to leverage larger training datasets and become more intelligent. However, this comes with greater energy demands.

The Energy Dilemma

The scale of compute used for model training has grown exponentially, enabling the development and deployment of large language models. These advancements assist researchers and financial institutions worldwide in operationalizing ever-larger multi-variable datasets. Financial firms, for instance, use text-to-data AI to convert qualitative inputs like social media posts into structured data, enhancing their decision-making processes.

Despite the impressive increase in data center computing capacity, which rose by 550% between 2010 and 2018, energy consumption only increased by 6% due to improved algorithmic and hardware efficiency. However, AI energy consumption is still on the rise. Massive global investments in AI suggest that energy costs will continue to grow.

Both the training and inference processes contribute to this increase in energy consumption. Historically, training costs have received more attention, but recent estimates indicate that the energy needed for the inference process may be higher. For example, Google reported in 2022 that 60% of its AI-related electricity use was linked to the inference process alone. This has heightened concerns about the technology’s impact on global emissions, as seen in studies suggesting that implementing an LLM assistant for Google browser searches could lead to annual energy consumption comparable to that of Ireland.

Impacts on the Global South

AI advancement is often justified by the potential benefits for economic growth and improved standards of living, such as enhanced healthcare services. However, the disproportionate impact of climate change on the global South poses significant risks for already vulnerable populations and regions.

There are three primary reasons why the economic benefits of AI are less likely to accrue to the global South compared to the global North:

1. Concentration of AI Development: AI development and industry are highly concentrated in major Western cities. This concentration means that most frontier models are created in the global North, leading to growing disparities in capacity and opportunity. Although there are examples of AI-based technologies supporting communities in the global South, such as computer vision for identifying crop diseases, the overall distribution of AI economic benefits remains highly concentrated.
2. Infrastructure and Connectivity Challenges: Advanced AI requires significant electrical and physical capital. Many global South countries have limited connectivity, with only 35% of people in developing countries having access to the internet, compared to 80% in developed economies. Sub-Saharan Africa, in particular, faces low electricity capacity (80% of the population) and internet access (36%). This limited infrastructure makes it challenging to introduce AI technology broadly, potentially exacerbating global inequalities.
3. Automation and Labor Displacement: AI may lead to the automation of labor, causing work reshoring and depriving digital sectors in the global South of foreign capital and income. This could impact IT sectors in countries like India, where heavy investments in IT skills may not be sufficient to counteract the potential outflow of capital and employment to the global North.

As AI continues to advance, it is crucial for institutions and collaborations to ensure that the global South can capture more of the benefits while minimizing risks. Without such efforts, developing countries may face the brunt of AI’s negative impacts, including elevated global emissions, without reaping the transformative benefits of this groundbreaking technology.

Source: OMFIF