Future of AI With Open Source

While open-source and free software becomes widespread, particularly in mobile and server operating systems, fears and biases against open-source software have given their place to increasing demands. Also, when the licences varied, open-source software reaching a high level of the audience have delivered great opportunities at low cost. AI projects have undoubtedly received their share from the increasing number of projects that gained momentum with low costs and community support. Being frequently spoken about in almost every part of society lately, it was also inevitable for machine learning models to accelerate thanks to open source. Besides, Google's Tensorflow and Facebook's Torch libraries are free and open-source, and these are the most prominent proofs that the progressions are in this area.


The increasing number of machine learning libraries have forced academicians to release machine learning models that would remain on paper and unsupported under normal circumstances; by releasing these models, academicians or some developers who read the article created evidence and provided an opportunity for other developers to use it. Moreover, these enhancements have sometimes been made for general use regardless of an academic paper. Thus, having sufficient programming knowledge, an individual has become able to use it by taking a ready-to-use model, making certain adjustments, and fine-tuning without diving into training costs and calculations. In fact, it has become possible for even those who don't have programming knowledge to learn and make implementations with Kaggle, a platform renowned by people working in artificial intelligence and data science.


Lately, Natural Language Processing (NLP) models created to interpret human language have been the most outstanding open-source projects. The increase in demand for processing online meetings and phone calls, especially during the pandemic period, has brought about the need for studies in this area. Thanks to open-source projects such as Huggingface and Spacy, the accessibility has distinctively increased and enabled institutions and organizations worldwide not to delay their desire to work in this area anymore. These libraries offer models whose training costs sometimes exceed 60,000 USD, free of charge and ready-to-use and enable many companies from the world's leading brands to start-ups to get ready for their dream developments. Thanks to open source and free software supporters, the relevant models didn't also remain available only in English, furthermore, models have been published in many languages, including Turkish. For instance, you can click here to check out the model and data set numbers shared only via hugging face.


Among the publications, there are models which are easy-to-use and support multiple languages. As an example, we have created simple content in which you can conduct tests at this link. As seen in this example, applications for simple sentence classifications have, in fact, decreased down to a few lines by these libraries and almost become usable by the end-user. Also, with Google's collab platform you have used in the link, people have had the opportunity to work in servers having powerful hardware free of charge and, undoubtedly, with certain time limits.


To sum up, the issues we have discussed, today's most excellent AI libraries and models are substantially developed as open-source and put into use for free. Depending entirely on the scientific developments and discoveries, the advances in this area have not been monopolized yet because nobody could reach open source' speed in imitating human intelligence, detecting the problems in this area, and solving them.


Hüseyin Erdem

Data Scientist at Maydanoz

Group 7.png