Open-Source AI: Fostering Collaboration and Responsibility

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Oct 31 2023

AI technology plays a pivotal role in propelling innovation across society, scientific domains, and economies. In a recent publication in the journal Nature Computational Science, researchers explored the significance of adhering to open-source principles in the creation of AI technology, with a focus on fostering collaboration, accessibility, interoperability, and responsibility.

Background

Computer science research has a rich history of embracing open-source principles. However, there is an increasing trend among companies to restrict access to AI innovations. For instance, OpenAI, an organization initially founded to make scientific research openly available but later restricted access to its research findings.

While this strategy aligns with a company's legitimate pursuit of financial returns, it also leads to a concentration of power, limiting access to AI technology. Over time, this concentration of power can result in growing inequality in AI education, research, and public use. The author explores the reasons why open-source AI should complement proprietary AI, focusing on key components for building AI technology: source codes, models, and datasets.

Proprietary AI technology

Proprietary AI technology presents several challenges. AI plays a pivotal role in driving innovation across various sectors, including education, entertainment, media, and management, with large language models (LLMs) such as Generative Pre-Trained Transformers (GPT)-4 serving as the foundation for text processing. The restriction of widespread access to AI poses a risk to downstream business models, innovations, products, and services. The concentration of power over AI technology impedes fair competition, future innovation, scientific progress, and the overall well-being and progress of humanity.

Moreover, proprietary AI technology can compromise responsibility and inclusiveness. When a few companies exclusively develop novel AI technologies such as LLMs, they can arbitrarily determine which countries and languages they support, potentially excluding users from smaller markets and rare languages. Maintaining a degree of openness in AI technology is essential for researchers to assess the security, safety, and fairness of AI systems. Proprietary AI systems are challenging for the public to evaluate, identify errors, and rectify them.

Advantages of open-source techniques

The advantages of open-source principles in software development are evident. Open-source software (OSS) revolves around the concept that organizations rely not only on their internal resources but also on external sources, such as bug reports, software packages, customer feedback, and published patents. Depending on the chosen license, OSS can coexist with commercialization, enabling companies to offer additional products and services for revenue.

The open-source model ensures efficient and effective software development, although it took time to fully realize its potential. Governments have played a crucial role in promoting OSS adoption, suggesting that a similar approach can benefit open-source AI.

OSS provides substantial advantages over proprietary counterparts, such as accessibility, collaboration, accountability, and compatibility. In contrast to licensed proprietary software, OSS is freely accessible with minimal usage constraints. OSS benefits from a diverse community, fostering superior software quality, rapid innovation, and increased creativity. Errors in OSS are promptly identified and rectified, making it ideal for robust technical systems. Furthermore, OSS hinges on open standards and modularity, promoting reusability and interconnectivity.

Companies derive multiple benefits from OSS. They construct cost-effective infrastructures, bolster their image through OSS contributions, and actively influence innovation by engaging in OSS endeavors. This involvement yields valuable feedback and promotes the advancement of pivotal technologies, thereby enhancing their competitiveness in the commercial realm.

Empowering open-source AI

Promoting open-source AI technology involves tailoring the approach to address its unique aspects. AI technology consists of three essential components: source codes for formalizing training tasks, datasets for training, and models storing trained weights. Open-source AI is expensive to operate and demands a lot of hardware resources. It also threatens society; thus, its development must be approached responsibly. This approach aims to enhance accessibility, collaboration, responsibility, and interoperability.

To enhance accessibility, policymakers should actively support the adoption and development of open-source AI. AI innovation demands significant capital, especially for contemporary AI models that rely on extensive data and infrastructure. Funding, public resources, and access to infrastructure and data are essential to kickstarting and scaling open-source AI.

Developing cutting-edge AI technology is typically beyond the capabilities of currently available computational resources, which are typically restricted to academics and do not include non-profit groups.

Collaboration in open-source AI technology can reduce development costs and address scalability issues. Collaboration is facilitated by creating research communities across public institutions, businesses, non-profits, and academics. Funding large-scale initiatives for open-source LLMs can further promote collaboration and inclusiveness.

Responsibility for open-source AI technology involves establishing clear barriers against misuse. Access control, like existing norms for open data, is necessary to ensure responsible use. Access control should vary according to the type of data, source code, or models. Licensing for open-source AI should prevent malicious practices and define permissive and restrictive use cases.

Privacy and security threats associated with open-source AI should be addressed responsibly. Interoperability is vital for AI technology's future. Standardized and modular building blocks within software libraries should enable easier adoption and customization in downstream applications. This standardization helps reduce dependence on specific AI technologies, promote competition, and mitigate lock-in effects.

Conclusion

In summary, this paper discussed the significance of open-source principles in AI. It highlights the challenges posed by exclusive proprietary AI technology, emphasizing the benefits of open-source models such as accessibility, collaboration, and accountability. The author advocates for a responsible and inclusive approach to open-source AI to bridge gaps in innovation and promote societal welfare.