[Paper Review] Near to Mid-term Risks and Opportunities of Open-Source Generative AI

Francisco Eiras, Aleksandar Petrov+22 more|arXiv (Cornell University)|Apr 25, 2024

Scientific Computing and Data ManagementDecision Sciences3 citations

TL;DR

This paper advocates for the responsible open-sourcing of generative AI models in the near to mid-term, arguing that open access accelerates innovation, safety research, and equitable development. Using a novel AI openness taxonomy applied to 40 LLMs, it identifies differential risks and benefits of open vs. closed models and proposes technical, ethical, and policy measures to mitigate risks while maximizing societal benefits through transparency and collaboration.

ABSTRACT

In the next few years, applications of Generative AI are expected to revolutionize a number of different areas, ranging from science & medicine to education. The potential for these seismic changes has triggered a lively debate about potential risks and resulted in calls for tighter regulation, in particular from some of the major tech companies who are leading in AI development. This regulation is likely to put at risk the budding field of open-source Generative AI. We argue for the responsible open sourcing of generative AI models in the near and medium term. To set the stage, we first introduce an AI openness taxonomy system and apply it to 40 current large language models. We then outline differential benefits and risks of open versus closed source AI and present potential risk mitigation, ranging from best practices to calls for technical and scientific contributions. We hope that this report will add a much needed missing voice to the current public discourse on near to mid-term AI safety and other societal impact.

Motivation & Objective

• To argue that open-sourcing generative AI models in the near to mid-term is essential for innovation, safety research, and equitable access.
• To develop and apply an AI openness taxonomy to assess the current state of 40 large language models.
• To compare the risks and benefits of open versus closed-source generative AI across technical, societal, and regulatory dimensions.
• To identify and recommend concrete risk mitigation strategies, including best practices and technical contributions, for open-source GenAI development.
• To provide a balanced, evidence-based counterpoint to growing calls for tighter regulation that may stifle open-source progress.

Proposed method

• Developed a three-stage framework (near-term, mid-term, long-term) to categorize generative AI development based on adoption rates and technological advancement rather than time.
• Applied an AI openness taxonomy to analyze 40 large language models, assessing openness across code, data, and model weights.
• Mapped the model lifecycle into three stages: training, evaluation, and deployment, with a focus on LLMs.
• Conducted a comparative analysis of open vs. closed source models across dimensions including innovation, safety, and regulatory risk.
• Evaluated global regulatory landscapes in regions including the EU, US, China, Saudi Arabia, UAE, and others to assess alignment with open-source principles.
• Proposed a set of technical, scientific, and policy recommendations for responsible open-sourcing, including model auditing, transparency, and governance frameworks.

Figure 1: Three Development Stages for Generative AI Models : near-term is defined by early use and exploration of the technology in much of its current stage; mid-term is a result of the widespread adoption of the technology and further scaling at current pace; long-term is the result of technologi

Experimental results

Research questions

• RQ1 What are the key risks and benefits of open-sourcing generative AI models in the near to mid-term?
• RQ2 How can openness in generative AI be systematically categorized and measured across models?
• RQ3 In what ways do current regulatory frameworks in different regions affect the development and distribution of open-source generative AI?
• RQ4 What technical and procedural safeguards can mitigate risks associated with open-sourcing powerful generative models?
• RQ5 How can open-source generative AI contribute to equitable innovation and safety research compared to closed-source alternatives?

Key findings

• The AI openness taxonomy successfully categorized 40 large language models, revealing that while many models are partially open, full transparency across weights, data, and code remains rare.
• Open-sourced generative AI models enable broader safety research, faster innovation, and more resilient systems compared to closed models.
• Regulatory efforts in regions like the EU and US often fail to account for open-source GenAI, potentially undermining its development.
• Countries such as the UAE and Saudi Arabia show strong support for open-source AI through initiatives like Falcon and Allam, indicating growing global alignment with open innovation.
• The report identifies that open-sourcing models can reduce monopolistic control and enhance transparency, but only when paired with strong risk mitigation practices.
• There is a significant gap in policy recognition of open-source GenAI, with most regulations focusing on closed models and corporate deployment, not community-driven development.

Figure 2: Model Pipeline : stages showing (1) training, (2) evaluation, and (3) deployment analyzed in the report. The component Common Benchmarks Evaluation (light gray) is included for completeness yet will not be analyzed in detail as these are standard and commonly available.