The Murky World of Open AI: What Does Open Even Mean?

2024-10-30

The concept of “open source” has revolutionized software development, but its application to the burgeoning field of Artificial Intelligence (AI) is creating confusion. While everyone seems to be on board with the idea of openness in some way, there’s a lack of consensus on what it actually entails. This ambiguity can hinder innovation and raise ethical concerns.

The Open Source Conundrum

Open source traditionally refers to software where anyone can access, modify, and distribute the code. This transparency fosters collaboration and rapid development. However, AI models are more than just code. They’re intricate mathematical structures trained on vast datasets. Understanding, reproducing, or modifying them requires access to both the code and the training data – a point where things get messy.

Opening the Door (a Crack)

Companies like OpenAI and Meta have released some open-source models, but with limitations. OpenAI’s ChatGPT, despite its name, isn’t fully open-source. Meta’s Llama releases come with the model and weights, but not the training data. The Allen Institute, however, took a bolder step by releasing not just the model and weights, but also the full training data and pre-training code.

Shades of Openness

Experts like Moez Draief from Mozilla.ai emphasize that sharing an AI model isn’t the same as sharing traditional software. Issues like privacy, bias, and liability complicate the picture. Draief proposes a spectrum of openness for AI, where complete closure isn’t the only alternative. Platforms like Hugging Face advocate for a community-driven approach to define “open” in the context of AI.

Closed First, Open Later?

Some companies, like Meta, plan to open-source their models later. This “closed first” approach allows time for alignment – ensuring the AI reflects the intended values. Proponents like Zack Kass argue that AI needs to mature before venturing into the open world, just like children need to grow up before facing society’s challenges.

Openwashing vs. True Openness

Critics like Cory Doctorow accuse tech giants of “openwashing” – using open source rhetoric without true commitment. He believes they use this language to deflect regulation and criticism by portraying themselves as ethically open.

The Road Ahead

The lack of clarity surrounding open source AI requires attention. President Biden’s AI executive order calls for an assessment of the opportunities and risks associated with open source AI by the National Telecommunications and Information Administration. As the AI landscape evolves, a clearer definition of “openness” is crucial for ethical, responsible development and innovation.

References:

Initially Reported By: Axios.com
https://www.digitalnomadsforum.com
Wikipedia: https://www.wikipedia.org
Undercode AI: https://ai.undercodetesting.com

Image Source:

OpenAI: https://openai.com
Undercode AI DI v2: https://ai.undercode.help