Font size:
Print
Democratising AI
Context:
The growing influence of Big Tech companies has become a pressing concern for policymakers worldwide.
More on News
- In response, countries like India are investing in sovereign cloud infrastructure, open data platforms, and support for local start-ups to democratise AI development.
- However, these measures might not suffice and could inadvertently strengthen Big Tech’s dominance.
Challenges of Big Tech’s Dominance
- Prohibitive Costs of Deep Learning Models: The high computational expense of training deep learning models makes it nearly impossible for smaller players to compete.
- For instance, training a model like Gemini Ultra in 2023 cost around $200 million.
- This financial barrier often forces new entrants to rely on Big Tech for computational credits. Big Tech companies, in turn, advocate for ever-larger models, solidifying their dominance and recovering costs through their cloud services.
- Integrated Ecosystems and Switching Costs: Big Tech’s competitive edge lies in offering end-to-end services, including advanced developer tools, optimised cloud infrastructure, and the latest AI models.
- These offerings simplify workflows and reduce costs for developers, making it difficult for alternative infrastructures to match their efficiency.
- This integrated ecosystem further raises the costs of switching to other providers.
- Data Monopolies: Big Tech companies benefit from continuous streams of diverse and global data, providing a competitive advantage through superior “data intelligence.”
- Smaller AI companies often end up selling to Big Tech, perpetuating their dominance.
- While public data initiatives aim to democratise access, they often fall short, as better-resourced players like Big Tech are better positioned to exploit open data platforms.
- Marginalisation of Academia: The dominance of deep learning has shifted AI research from academia to the commercial domain.
- Big Tech companies now produce more academic publications and citations, allowing them to shape the direction of AI research and development.
Rethinking AI Development
To address these challenges, a fundamentally different approach to AI development is needed—one that does not replicate Big Tech’s model but instead changes the rules of the game. A “theory of change” framework can guide this effort, focusing on causal mechanisms and purposeful interventions rather than sheer computational power and large datasets.
- Small AI and Purpose-Driven Models: AI development should prioritise domain expertise and lived experiences over statistical patterns in Big Data.
- This approach would involve creating smaller, targeted models designed to address specific societal challenges, with curated data collection to refine theories of change.
- Historical Precedents: Fields such as medicine, aviation, and weather forecasting have historically relied on theory-driven models, emphasising hypothesis testing and scientific rigour over vast amounts of data.
- Revisiting this approach in AI could offer a more democratic and efficient path forward.
Missed Opportunities and the Way Forward
- The current obsession with “bigger is better” in AI entrenches dependence on Big Tech and its exploitative model of commercial surveillance.
- The recently signed Global Development Compact, while advocating for democratising AI, fails to challenge this paradigm, relying instead on building large datasets and increasing computational access.
- To truly democratise AI and reduce Big Tech’s dominance, the focus must shift toward developing small, theory-driven AI models anchored in specific goals.
- By changing the narrative around AI development, countries can foster a more equitable and effective ecosystem that prioritises public interest over corporate control.
