Open-Source Framework

Anyone who builds a house doesn’t cast every brick themselves. In the same way, programmers don’t write every line of their software from scratch. They fall back on a framework: a collection of ready-made building blocks and rules that takes over the same recurring groundwork. “Open source” means that the program code is openly viewable and anyone may use, modify, and pass it on free of charge. An open-source framework is thus a freely available foundational structure for software, on which often thousands of people worldwide work together. Its counterpart is closed software, whose inner workings are known only to the manufacturing company.

Why half the AI industry stands on the same building blocks

Without such foundational structures, the rapid development of recent years would hardly have been conceivable. A research group can try out a new idea within a few days instead of spending months programming the basics. What emerges in one lab immediately runs elsewhere on the same foundation. This shared base is a main reason why results in AI research spread so quickly.

For companies, open code is also a matter of independence. Anyone using closed software from a single provider is dependent on that provider’s prices and decisions. If the provider discontinues the product, one is left with a pile of shards. With open code, one can, if necessary, continue the work oneself or commission experts to do so.

A common misconception is that “free of charge” also means “without conditions.” Every open-source project operates under a license, i.e., a usage agreement. Some licenses require that one’s own extensions also be disclosed. Others permit almost anything, even sale. Companies scrutinize these conditions very closely before adopting a framework.

From the repository to your own computer

The code of such a project resides in a public storage location on the internet, usually on platforms like GitHub. There, anyone can view it, download it, and report bugs. Anyone wishing to contribute an improvement submits a proposed change. A core team reviews this proposal and either adopts it or rejects it. In this way, the project grows from many small contributions.

In one’s own program, the framework is then integrated as a building block, and its ready-made functions are called upon. With an AI framework like PyTorch, for example, it takes over the elaborate computational work on the graphics card. The developer only needs to describe the structure of their model and the training data. Everything below that is handled by the foundational structure.

It is important to distinguish this from an open AI model. A framework is a tool, a model is a result. One can very well train a completely secret model using an open-source framework. Conversely, there are freely available models whose training data nobody discloses.

From the school app to the corporate press release

Almost every app on your phone contains code from open projects. The Android operating system is built on the open kernel Linux. Many websites run on open servers and databases. You notice none of this in everyday life because these building blocks are buried deep inside.

In business news, the term usually comes up when large corporations give away a framework for free. Meta released PyTorch as open source, Google previously did the same with TensorFlow. This sounds generous, but it is also strategy. Whoever sets the standard on which everyone works shapes the industry and finds it easier to recruit specialists who already master the company’s own tool.

The downside becomes apparent when a security vulnerability turns up in a widely used building block. Then millions of systems are suddenly affected, because everyone uses the same code. Such cases regularly make headlines. They are one reason why authorities and companies now precisely document which open building blocks are embedded in their software.