Decentralized AI will be the key to unlocking global development | Opinion

Today, AI is omnipresent, influencing everything from how we boost productivity at work to how we resolve emotionally-charged personal matters. And while innovation in this sense has its benefits, it lacks the capacity to make a tangible impact in the most disadvantaged corners of the world. 

The United Nations Development Programme has been relentless in its pursuit of the 17 Sustainable Development Goals to eradicate poverty, drive climate action, and support equitable growth by 2030. Given its existing worldwide use cases and applications, it would be natural to view AI as the key to promoting inclusivity and global development. However, the current centralized architecture of AI is plagued by setbacks, including data privacy concerns, high costs, and limited accessibility, which severely undermine AI’s capacity for good.  

This centralized nature of AI only reinforces existing power imbalances, preventing AI from realizing its potential for good in the Global South through inherent biases, stripping communities of data sovereignty, and a lack of transparency. For AI to be applied effectively as a tool to drive global development, there must be a shift from the corporate, centralized architecture to one founded on inclusion, sovereignty, and accountability. Decentralized AI is that solution.

The centralization paradox

While AI has already been applied to solve challenges from climate change to healthcare, the reality is that its development is largely centralized, dominated by a handful of tech giants whose systems are practically and unethically unsuitable for the unique contexts of the 17 SDGs established by the United Nations. But the crisis is not one of technology, but of governance. The conventional model of AI development creates three distinct obstacles for genuine development impact.

Centralized models are overwhelmingly trained on data sourced from a handful of developed regions, excluding regions across the Global South. Studies show that when deployed in diverse contexts, such as diagnosing diseases or predicting financial risks, these models become fundamentally inept in their intended purpose. The absence of appropriate training can lead to systematic misidentification, the denial of critical services, and the reinforcement of socioeconomic disparities, posing a threat to SDG 10, which aims to promote the social, economic, and political inclusion of all.

These systems also demand the aggregation of highly sensitive local data, from patient records to financial or criminal records, onto distant corporate servers, which are susceptible to hacks due to their centralization. The practice of data extraction strips governments and institutions of their right to data sovereignty, threatening SDG 16, which advocates for the right to peace, justice, and strong institutions, and jeopardising the security of the data aggregated away from local servers. This practice has also led to the proliferation of sovereign AI technologies emerging in countries such as Singapore and Malaysia, in an arms race to preserve data sovereignty. 

But the most important consideration is that when an opaque, poorly understood AI makes a critical error or generalization on policies that could affect millions of lives, who is held responsible? The “black box” nature of centralized AI systems, including their ownership and mechanisms, makes auditing decisions, such as aid distribution or risk modelling, and assigning accountability, concerningly difficult and ethically unacceptable for high-stakes and development work. This lack of transparency has the potential to undermine all 17 SDGs.

The only way to reconcile the power of AI with the ethical requirements of international development is through a fundamental shift away from corporate, centralized AI training to mechanisms grounded in the principles of inclusion, sovereignty, and accountability.  

Decentralized AI: A two-pronged solution

Decentralized AI, anchored by federated learning and blockchain technology, is emerging as the solution to the conundrum. The SDG Blockchain Accelerator Programme, strategically led by the UNDP, supported by partners including Blockchain for Good Alliance, Stellar, FLock.io, and EMURGO Labs, further validates this by pioneering decentralized AI initiatives that empower, not hinder, communities in the Global South.

Federated learning works by training shared models across multiple decentralized devices while preserving local data. Projects in the Latin America and Caribbean region use the technology to collaboratively train predictive AI to forecast climate risks accurately while keeping local financial and demographic data secure in local servers. This infrastructure supports efficient and equitable payouts to climate-vulnerable farmers and female-led enterprises, delivering on both SDG 13 (climate action) and SDG 5 (gender equality). 

The operations supported by federated learning are complemented by blockchain technology, which replaces a single corporate intermediary with an immutable and transparent governance system. This creates an infrastructure built for collaboration and restores necessary accountability mechanisms. In Liberia, smart contracts and decentralized AI are being deployed to facilitate transparent distribution of payments and aid, while in Kenya, decentralized AI eradicates payment discrepancies for local companies, boosting the economic growth and confidence in public institutions outlined in SDG 8 (decent work & economic growth) and  SDG 10 (reduced inequalities).

Additional applications of decentralized technology to support the SDGs include the development of a blockchain-based NFT by Cambridge University and UNDP Rwanda to support mountain gorilla conservation, and secure hospital records in Africa that give patients the autonomy to grant or revoke access to their patient records, delivering on SDG 15 (life on land) and SDG 3 (good health & wellbeing).

A call for architectural responsibility

As a technology, AI has immense potential, but its fundamental challenge is one of governance. The centralized, proprietary model fundamentally undermines the principles of inclusion, sovereignty, and accountability that the UNDP SDGs embody, but current efforts demonstrate that a viable, ethical, and scalable alternative exists. 

The challenge now is for the global development community to prioritize funding the adoption of open, decentralized AI infrastructures over corporate tools that stifle development. It’s time to shift our mindset from that of passive consumers to custodians of intelligence that drive a sustainable future for the most disadvantaged corners of the Earth.