Open source and artificial intelligence (AI) share a deep-rooted history. In 1971, while many associated AI with Isaac Asimov's Three Laws of Robotics, Richard M. Stallman joined MIT's Artificial Intelligence Lab. This was a time when AI was more than just a concept from science fiction. As proprietary software began to emerge, Stallman introduced the revolutionary idea of Free Software. This concept evolved into open source, which paved the way for contemporary AI.
Alan Turing, a computer scientist, laid the foundation for the AI movement with his 1950 paper, "Computing Machine and Intelligence." This paper introduced the Turing Test, which posits that if a machine can convincingly emulate human conversation, it can be considered intelligent. While some believe that today's AI systems can pass this test, it's evident that significant progress is being made.
In 1960, another computer scientist, John McCarthy, coined the term "artificial intelligence." He also developed the Lisp language, which became the first programming language for AI. Lisp's unique feature was its ability to intertwine data and code. Despite its potential, the AI advancements of the 1980s were limited by the available hardware and the absence of essential resources like Big Data. However, open-source projects like Hadoop, Spark, and Cassandra later provided the necessary tools for AI and machine learning, enabling the storage and processing of vast amounts of data.
Today, even tech magnates like Bill Gates acknowledge the significance of open-source-based AI. Popular AI generative models, including ChatGPT and Llama 2, have their roots in open source. While these models are not entirely open source, their foundations are built on open-source platforms. For instance, Hugging Face's Transformer is a leading open-source library essential for building machine learning models. ChatGPT, for example, heavily relies on this library. Additionally, TensorFlow and PyTorch, products of Google and Facebook respectively, are pivotal in the development of ChatGPT and other AI models.
While some AI programs, like Meta's Llama-2, claim to be open source, they come with licensing restrictions. For instance, if a program using Llama-2 garners more than 700 million monthly active users, the developer must seek a license from Meta.
Despite these challenges, the open-source community remains a formidable player in the AI arena. Contrary to the belief that massive cloud infrastructures or high-end GPUs are essential for AI, it's now possible to run large language models on devices as compact as a smartphone. With tools like the Hugging Face open-source low-rank adaptation (LoRA), AI models can be fine-tuned efficiently and cost-effectively.
In conclusion, the open-source community has always been at the forefront of technological advancements. Just as open-source operating systems challenged and eventually surpassed proprietary systems, it might not be long before a fully open AI system outperforms its semi-proprietary counterparts.
References:
- Alan Turing's "Computing Machine and Intelligence"
- Hadoop
- Spark
- Cassandra
- Hugging Face's Transformer
- TensorFlow
- PyTorch
- Meta's Llama-2
- Hugging Face open-source low-rank adaptation (LoRA)
