Generative AI is entering company processes at an ever-increasing pace, thanks to its ability to create a huge variety of content, automate the most useful and frequent daily operations, and significantly broaden the knowledge base available to human resources at all levels of the organisation.
Many organisations are still asking themselves what is the best way to implement this emerging technology. One of the first questions that arises when considering AI adoption is whether to opt for commercial (closed source) models or to prefer an approach based on open source models, in order to have full control over customisation and the handling of proprietary data.
Open source AI has both advantages and limitations, which must be evaluated with great care. Drawing on Moxoff's experience, we will look specifically at what open source AI currently consists of, the factors that characterise it, and at the same time highlight the main differences compared to closed source solutions.
We will see how organisations can currently find their way between two alternatives that often give their best when used synergistically, making the most of the main strengths of both, provided they are properly understood, customised and managed in detail.
What is Open Source AI
Open source in the field of AI includes artificial intelligence models in which the source code and model parameters are freely available for use, modification, and commercial distribution, in accordance with the relevant licence type.
Making a system’s algorithms, pre-trained models, and datasets open enables the creation of a thriving community of students, professionals, and researchers who actively contribute to technological development. This promotes innovation and experimentation even in areas where a commercial model would probably never be used, as there would be no direct opportunity for ROI.
Public use allows open source AI projects to grow day by day, accelerating the development of practical solutions that can best adapt to the real challenges companies face along their digital transformation journey.
Open source AI projects are available on platforms such as GitHub, which are used daily by millions of developers worldwide, and they now play a crucial role in the search for innovative solutions in sectors such as manufacturing, finance, healthcare, and education.
AI frameworks based on open projects are available for Windows, Linux, macOS, iOS, and Android, offering tangible support for developers of AI-based applications across all desktop and mobile environments.
By leveraging freely available libraries and tools, even small development teams can focus on the creation of customised solutions, saving time and resources compared to having to develop them from scratch.
In other words, open source AI is democratising access to AI-based technologies, speeding up the development of high-impact applications for a wide range of business use cases.
A brief history of Open Source
It was 1971 when Richard Stallman, still regarded as the father of the open source movement, joined the MIT AI Lab in his first year at the prestigious Harvard University. Although this experience did not have a great future, it laid the foundations for what would become something truly extraordinary in the history of information technology.
Over the course of about ten years, the intellectual property produced was never made public, but instead acquired by private companies that had supported its development.
Clearly unhappy with this, Stallman developed the belief that for a society to be truly free, it needs software that is not just a black box to be used, but can be adapted and improved not only for the benefit of a specific business, but for the whole community. According to Stallman, for this to happen, software must be free, which cannot occur if it is owned by a single entity.
In 1984, Richard Stallman launched the GNU Project, supported by the Free Software Foundation, to lay the foundations for the development of the operating system that would become Linux.
In 1998, the source code of Netscape Navigator, one of the first internet browsers to become mainstream, was made open, encouraging the development of many web applications, including the well-known Mozilla Firefox (2003).
Today, a substantial contribution to the culture and development of projects based on free software is provided by the Open Source Initiative (OSI), founded by activist Eric Raymond, who also inspired Netscape to make its work publicly available.
In addition to its advocacy work and providing guidelines for free software policies and licences, the Open Source Initiative also promotes a common line of discussion around open source AI.
Advantages of Open Source AI
To the question of why artificial intelligence needs to be open source, the Open Source Initiative offers the following answer: "The principles of Open Source have shown that huge benefits arise for everyone when barriers to learning, using, sharing and improving software systems are removed. These benefits can be achieved through autonomy, transparency and collaborative improvement. Everyone needs these advantages from artificial intelligence. We need essential freedoms to enable users to create and deploy trustworthy and transparent AI systems."
Open source AI therefore plays an explicit role as a catalyst for the rapid and collaborative development of artificial intelligence-based systems. Its democratising nature has opened up opportunities for developers around the world, regardless of their affiliation with large companies or research institutions.
Open source AI is accelerating the development of intelligent systems, generating significant advantages in terms of accessibility, collaboration, transparency, security, innovation and adaptability.
Accessibility and collaboration
Thanks to their nature and objectives, open source AI projects, especially models, are freely accessible to a wide range of stakeholders, including students, professional developers, researchers, and a broad array of organisations engaged in developing innovative solutions.
The main strength of open source AI lies in its ability to foster the creation of communities that are highly active and involved in advancing artificial intelligence–based projects, both at the level of pure technology, for example by progressively improving the quality of AI models, and in developing vertical applications across many business sectors.
Transparency and Security
The ability to freely access the source code allows developers to accurately evaluate how an AI model functions. This condition of transparency is essential in a technology such as machine learning, which involves many black boxes that must be clearly explained to stakeholders.
Open code encourages the spread of a culture focused on achieving truly ethical and responsible AI in its applications, especially in fields where decision support concerns critical areas such as healthcare, the administration of justice, or the granting of mortgages and loans.
Using open code naturally fosters trust, helping to overcome the mistrust that often arises in society towards closed models, where only the owner knows the specific details of how they work.
Innovation and Adaptability
Open source AI, thanks to communities made up of a wide variety of stakeholders, promotes experimentation and the continuous discovery of new solutions to an incredibly broad range of problems. The benefits produced by individual research efforts contribute to advantages that extend to society as a whole.
Researchers and developers can also openly share their experiences, including both mistakes and successes, to find new inspiration and determine the best direction to take when it comes to making AI applications more efficient.
Limitations of Open Source AI
Despite its clear advantages, open source AI also presents a number of objective challenges, arising both from its inherently open nature and from the relative immaturity of some AI techniques, such as in the case of generative artificial intelligence.
Security Issues
Open source AI, especially in the enterprise context, raises concerns about security and reliability, precisely because “anyone can get their hands on it.” This applies both to less competent developers and to malicious individuals, who could, for example, use generative artificial intelligence applications to commit fraud or launch cybersecurity attacks, as well as to conduct disinformation campaigns.
Through the establishment of the AI Alliance, companies such as Meta and IBM—as well as over a hundred other organisations that have since joined—support open source AI by freely making some of their technologies available as part of an open scientific exchange designed to generate innovation.
At the same time, brands such as Google, Microsoft, and OpenAI are more in favour of a closed and controlled approach, citing concerns regarding the security and irresponsible or improper use of AI models. The matter, in fact, remains unresolved.
Quality and Reliability
The excessive freedom of applications based on open source AI technologies, as in any other area of free software, inevitably leads to the production of low-quality content, which must be filtered out using a high level of expertise
Indiscriminate development of AI models can result in misaligned outcomes, wasted resources, and projects doomed to fail, creating a sense of distrust towards the technology, beyond any economic harm.
It is therefore essential to pay close attention when choosing which tools to use. The support of a qualified consultant with proven field experience, such as Moxoff, can help companies avoid falling into the trap of unreliable AI applications.
Lack of official support
When open source projects are not backed by solid foundations, specifically funded to constantly verify the validity of AI models, there is an inevitable risk of encountering the problems mentioned in the previous paragraph. Open source constitutes a universe of different realities, without the official guarantee provided by a commercial brand.
Open Source AI and Closed Source AI
Open source AI was created as an alternative to the traditional closed source model typical of proprietary commercial software. In the case of AI, “closed” systems involve algorithms and data that are not freely distributed in the public domain. Users typically interact with these systems via a set of APIs, without having full access to the underlying technology.
A well-known closed source AI technology is ChatGPT, from OpenAI, while Meta Llama is one of the most widely used open source LLMs. Let’s look at the main differences to consider when evaluating whether to adopt open or closed source models.
As of today, the most performant models—in areas such as NLP or LLMs, are undoubtedly closed source. However, the open source community offers some very interesting insights and progress, and is showing itself to be catching up, thanks in part to organisations like MistralAI, a French company that has based its business model on releasing open source models.
Costs and Investments
Closed source AI, much like the majority of commercial software, is distributed by an owner who manages every aspect end-to-end. Whether it is a tech start-up or a big tech company, these are private firms funded directly by their partners and investors.
Open source AI projects, in line with the principles of free software, are generally managed by dedicated foundations, which coordinate the community and release official versions of applications according to a public roadmap shared with stakeholders.
The foundations supporting open source software are funded voluntarily by their community and a series of technology partners in the initiative, which often include developers of closed source software as well.
Another common investor is the so-called big client, typically a prominent brand that wishes to customise its applications without being subject to the constraints of closed software. For this reason, they choose to finance the development of open source software to meet their own business needs first and foremost.
From the perspective of the client company, it is not guaranteed that a commercial solution will be more expensive than an open source solution. Even without an initial cost, an open AI model can sometimes be particularly costly to manage. It is therefore necessary to evaluate each case individually, relying on proven expertise, to determine which project is the most interesting and economically sustainable.
Flexibility and Customisation
One of the main criticisms levelled at closed source AI is that it tends to be overly conservative in line with commercial interests, lacking the degree of innovation that an emerging technology would deserve. This happens for various reasons, primarily due to the vendor’s desire to maintain a high level of control over data and models, even at the expense of the increasingly common demand for customisation from clients.
In this regard, open source AI is inherently more flexible and open to the customisation needed to adapt a technology to specific business use cases, especially concerning the data used for training an AI model.
In any case, techniques such as RAG can be used via prompt engineering to customise even the knowledge base of a closed commercial model, without necessarily having to retrain it.
Support and Maintenance
In the case of commercial closed source software, the vendor is responsible for keeping the model up to date and releasing new versions, following their own implementation roadmap.
For solutions based on open source AI, the foundation managing the project releases new versions and addresses any issues that are brought to its attention by the community.
Having access to the application’s source code also allows the client company to make modifications and integrations to adapt the model to its operational needs, without necessarily having to wait for certain features to be included in the public release.
Examples of Open Source AI
In the field of open source AI, there are several widely used frameworks that enable the development of models while benefiting from substantial support and resources provided by the community. These frameworks go beyond the traditional concept of a tool, allowing users at all experience levels to take full advantage of the remarkable potential of AI-based applications.
TensorFlow
TensorFlow is a very popular and widely used machine learning framework. Known for its flexibility, it supports programming languages such as Python and JavaScript. TensorFlow enables the development and deployment of machine learning models for most existing platforms.
Thanks to its outstanding community, TensorFlow offers users a wide library of models that can be used as a starting point for new projects, simplifying the process and reducing the time to market for applications. The great variety of available resources and the opportunity to engage with many experts worldwide also encourages the experimentation of new AI-based solutions.
PyTorch
PyTorch is an open source artificial intelligence framework with an especially intuitive interface, enabling simplified debugging and a very flexible approach to creating deep learning models.
It stands out thanks to its strong integration with Python libraries and support for GPU acceleration, making it ideal for training and experimenting with a wide variety of models. This is reflected in its large user base, both in professional environments and in scientific research.
OpenAI Gym
Known worldwide for ChatGPT, a strictly closed source commercial product, OpenAI was in fact originally founded, as its name clearly suggests, to support the development of open and publicly shared models. The turning point came in 2019, when OpenAI decided it was “too dangerous” to continue releasing GPT publicly, fearing its creative potential could fall into the wrong hands.
This radical transformation of its business model led to the departure of some supporters who had initially backed the AI Lab led by Sam Altman precisely because of its openness, followed by Microsoft’s growing dominance in funding the initiative.
The legacy of OpenAI’s open period is still present thanks to OpenAI Gym, which is now managed by a community of independent developers who are very active in publishing libraries and fundamental models for advancing the field of machine learning, specifically in reinforcement learning. OpenAI itself has published research papers describing its models in detail, supporting the work of its followers.
Open Source AI e Moxoff
The choice between closed or open source AI solutions depends on many factors, which must be carefully assessed and weighed on a case-by-case basis, following a thorough analysis of each organisation's business needs. In this context, thanks to its expertise and demonstrated experience in developing and customising AI models, Moxoff is an ideal partner for evaluating the most suitable solution to successfully implement AI solutions in business processes.
Moxoff's experience can help organisations understand which AI solution is best suited to their case, taking into account available budgets—not only for the initial implementation of AI, but also for the customisation and ongoing maintenance required to achieve their desired objectives.
As previously discussed, open source offers significant flexibility and customisation, but companies without major requirements or the in-house capabilities to develop AI models may find closed-source tools more convenient.
For many years, Moxoff has supported businesses of all sizes and sectors in developing AI infrastructures that often include both closed and open source AI solutions, depending also on the available data assets.
There is no single answer as to whether open source or closed source AI is better. Any decision requires a balanced consideration of all the aspects highlighted so far, with a view to developing an AI strategy that meets current needs, while also being scalable over time and easily integrated with existing solutions.
