In today’s rapidly advancing digital era, AI is at the forefront of revolutionary innovations across almost every private sector. That is why President Joe Biden announced an executive order earlier this week to impose a series of sweeping new rules and regulations on the use of AI. From healthcare and education to entertainment, the transformative impact of AI is evident.
What Is Open Source AI?
Open source AI is artificial intelligence software and tools that have their source code open and available to the general public. Developers, researchers and other interested parties can access, use, modify and distribute the software without triggering any licensing conditions.
However, as the technology continues to evolve and become more sophisticated, a rather important question comes to light: Should the core principles, algorithms and datasets that underpin AI be proprietary? Or should they belong to the collective knowledge of humanity? As an academic deeply entrenched in the AI community, I firmly believe that AI technology should be open source for three reasons.
Open Source AI Can Reduce Bias
Completely eliminating AI bias is an extremely difficult problem. However, open source AI provides several mechanisms that help address and reduce bias including transparency, audits and community involvement.
Transparency
In this context, transparency means making AI models open source and publicly available so that researchers and developers can inspect the underlying code. This makes it easier to identify potential sources of bias in the training data or the design structure itself.
Contrast this to closed source AI running at a major company that has notoriously made racist inferences, produced misinformation and even called out its creators for exploiting their users, scaring some of the non-savvy users with terminator-style stories of AI turning against its creator. Transparency would have helped mitigate all of these issues, avoiding the inconveniences and offenses to the end users.
Auditing
Making AI open source means it is also open to external parties to audit the system without needing special access, permissions or NDAs, like they might with closed source AI. This would force the creators to be more responsible in creating the systems and further ensures that biases are caught and corrected.
Community Involvement
Open source code often involves a wide range of contributors from different backgrounds. A diverse group of contributors can bring different perspectives, which may in turn help to recognize and address biases that might be overlooked with a more homogeneous group that may be suffering from tunnel vision itself.
Open Source AI Can Advance Science
The growth of artificial intelligence has significantly changed how we approach scientific research. Open source AI, in particular, offers researchers a rich repository of knowledge and tools. Platforms such as Google’s TensorFlow and Meta’s PyTorch foster collaboration, accelerating progress and enhancing the quality of AI models.
Further, open source AI repositories like HuggingFace, which provide access to open-source projects in AI and allow users to download free pre-trained AI models, have influenced an array of scientific disciplines. For example, the Transformers Library created by HuggingFace helped researchers study linguistic phenomena and syntactic structures in various languages; mine medical literature and predict protein structures; analyze climate change impact; and mine and categorize data and logs from telescopic observations, among many other disciplines.
If these researchers were not properly funded (a common problem in academia) and the AI was all closed source, then none of these papers and publications would have been possible.
Open Source AI Can Create New Standards
Open source AI can be used to influence and create new standards. The rise of AI has led to discussions about ethical considerations including bias, fairness and transparency. Due to their public nature, open source AI projects, such as Bittensor, often lead these discussions, influencing the standards around responsible AI.
Closed source AI projects, though, do not foster innovation except for internal competition between company employees. The end result is a product that may not be ethically responsible, with no way to foster a discussion around its performance or design.
Additionally, from an engineering standpoint, reproducibility and performance benchmarks will also suffer without open source AI. Reproducibility is extremely crucial in AI research, with open source AI projects often championing rigorous documentation, standardized testing environments, and shared datasets to ensure others can reproduce results, leading to higher standards for documentation and reproducibility in the industry. There’s also a need to standardize benchmarks to compare the performance of all of these models.
Contrast what we’ve learned about open source AI in this article with the recent executive order, which states that every company training an AI model must report this model to the U.S. government and obtain approval. This seems like a rather outdated and draconian measure because open source AI already resolves those concerns in a more effective and efficient manner.
Further, the executive order maintains a status quo in which only a few corporations can wield AI for their own purposes. Open source AI seeks to democratize and decentralize this power for the greater good.
