The Dilemma of Open Source AI

Pardon my mini rant here but I thought a lot about open source vs. closed source during the holiday break..

Like many in AI, I believe in the importance of open source. It is not merely a cradle for innovation; it represents a safeguard against an overwhelming concentration of power. The idea that a few entities could wield so much control over an existential technology like AI seems fundamentally unsound. And even if we trust the current generation of companies, there’s no assurance that the next generation of companies, with access to even more powerful AI, won’t misuse the technology. By acquiescing to the status quo, we set a dangerous precedent for future generations.

Yet, most technologies today are close-source. Consider cloud infrastructure dominated by AWS, GCP, and Azure. Or operating systems dominated by Microsoft and Apple. Even those technologies that started out as open gravitate towards becoming closed eventually.

The reason for this is simple: as a technology’s value increases, so too does the incentive to privatize the IP and monopolize the market. As these monopolies grow, they invariably start to control the industry’s most critical assets and bottlenecks. In AI, this means access to GPUs or top AI engineers. This serves to further precipitate the monopoly. 

But the underlying reason behind the dominance of close-source is rooted in something else: a clear incentive framework. In closed-source, there are well-defined economic incentives that govern all stakeholders – from management and employees down to shareholders and customers — facilitated through a set of instruments such as salaries, stocks, and corporate profits. This clear framework aligns all parties towards a common goal.  

In contrast, the incentive framework for open-source is more opaque. While communal engagement, social recognition, and intrinsic motivations do help propel open-source, they often can’t overcome the more powerful economic forces that propel closed-source. The recent events that unfolded at OpenAI is a testament to the formidable nature of such economic forces.

To bridge this gap, we must establish a clearer incentive framework for all participants in the open-source value chain, possibly down to the data/content creators themselves. Technologies such as identity management (via. watermarking) or distributed payments (via blockchain/tokenization) could help. Academia and policymakers must also get involved.

It would be naive to assume open-source will win just because it is the “better” path for humanity. Economic incentives and monopolistic tendencies are innately powerful constructs that should not be underestimated. Instead, we must start a discourse around new mental models for open-source and work towards a system that’s not just idealistically superior but practically superior as well.  Only then can we build an AI ecosystem that thrives on collaboration over proprietorship and participation over exclusion.