Open-Source AI

In the AI Story

The structural significance of open-source AI is not primarily technical. It is economic. A digital economy entirely dependent on a handful of proprietary model providers is a digital economy whose landlords can raise rents at any time, change terms at any time, and demolish the buildings where experimentation happens. Open-source models function as the continuously available alternative that disciplines pricing and terms across the entire market — not because most builders will run models locally, but because the option of doing so sets a floor under what proprietary providers can extract.

The analogy to Jacobs's old buildings is precise. Old buildings in healthy cities are not always the preferred space for every enterprise; newer, better-appointed buildings attract tenants who can afford them. But the old buildings remain available, and their availability means that experimental enterprises can start, that established enterprises can threaten to leave if rents rise too high, and that the overall ecosystem maintains the diversity that concentrated landlords would otherwise suppress.

The current state of open-source AI is genuinely ambiguous. On one hand, the capability gap between open models and frontier proprietary models narrowed dramatically between 2023 and 2026. Llama 3.1 and its successors, DeepSeek-R1, and Mistral's releases demonstrated that capable general-purpose models could be produced outside the largest American labs. On the other hand, the compute cost of frontier training continues to rise, and the organizations that can afford such training remain a small set — some commercial (Meta, Alibaba), some partially public (European efforts, academic consortia), with the long-term stability of each uncertain.

The Jane Jacobs volume's argument is not that open-source AI will inevitably succeed, but that its continued existence is necessary to prevent the catastrophic money dynamic from producing a fully captive digital economy. The institutional support required is specific: public investment in non-commercial frontier research, regulatory frameworks that prevent proprietary providers from foreclosing open alternatives, and professional communities that maintain tooling, expertise, and infrastructure for open-model deployment. These are the zoning codes that determine whether cheap space remains available.

The analogy also illuminates a specific risk. In physical cities, the supply of old buildings naturally renews: today's new buildings become tomorrow's old buildings, as construction costs amortize and neighborhoods shift. The digital equivalent is less automatic. If the open-source ecosystem cannot sustain the compute requirements of frontier training, the stock of cheap alternatives may erode rather than renew. This is the specific institutional challenge the framework identifies: maintaining the continuous renewal of cheap space in an industry where the cost of "construction" is rising rather than amortizing.

Origin

The open-source AI movement traces through early releases of BERT (Google, 2018), GPT-2 (OpenAI, 2019), and the broader HuggingFace ecosystem, but the decisive inflection came with Meta's release of LLaMA in 2023 and subsequent open-weight releases through 2026. The Jane Jacobs volume reads this ecosystem through her framework for renewable cheap space; related framings appear in Morozov's analysis of data infrastructure and in Ostrom's framework for commons governance.

Key Ideas

Economic function before technical function. Open-source AI's primary role is to discipline proprietary pricing and maintain the supply of cheap space, not necessarily to serve as the default option.

Disciplinary effect. The option of open alternatives sets a floor on what proprietary providers can extract, even when most users remain on proprietary platforms.

Renewal is not automatic. Unlike physical old buildings, the open-source stock does not renew itself; sustained public and community investment is required.

Institutional support is structural, not charitable. The zoning codes that maintain open-source viability are public funding, regulatory protection, and professional infrastructure.

The capability gap is narrowing but compute costs are rising. Whether open models keep pace with frontier capability depends on institutional choices made now.

Debates & Critiques

The open-source AI debate operates on multiple axes. Safety researchers disagree about whether widely available frontier-capable models increase or decrease catastrophic risk. Commercial AI companies argue that proprietary development is necessary to sustain the investment frontier capabilities require. Open-source advocates respond that the cost of capability suppression is structural dependence. The Jane Jacobs framework does not resolve these debates; it specifies that whatever trade-offs are accepted, the absence of a viable open alternative produces the specific pathology of fully captive economies that her urban research documented in detail.