AI: commercial tools or open-source algorithms?

Most organisations of any size are exploring how they can integrate artificial intelligence (AI) into their current operations. The benefits can seem enormous, almost magical. But getting started in any way beyond using readily available tools such as ChatGPT is very challenging.

The first question that organisations need to explore is: “How do we hope to benefit from the use of AI?” While that is an essential question to answer, it’s not one that will be addressed here in detail – beyond saying that AI can be used in a wide range of ways, depending on an organisation’s circumstances. These can include increasing productivity, identifying market trends, suggesting and designing new products and enhancing security.

Once an organisation has identified why they want to use AI and what sort of goals they have in mind, they need to address the tricky question of “How are we going to get there?” The easy answer is paying for an existing commercial service – and that’s not a bad answer, especially if you have deep pockets. After all, commercially available tools have generally been well-tested and should have some good customer support.

However, many organisations may question the value that established (and not-so-established) commercial providers can deliver. Many organisations, especially larger or more technical ones, prefer an alternative approach: building their own AI tool from open-source algorithms.

While this can sound daunting, there are numerous open-source AI resources that can help, including TensorFlow, PyTorch and Hugging Face Transformers. These resources mean that building a new tool from scratch isn’t just an option for huge corporations; it’s something that even relatively small organisations can aspire to if they have the right technical resources and skills.

The DIY approach has several important advantages, including cost, customisation, control and community. 

Cost-effectiveness

Open-source tools aren’t free to develop and operate, as a cost must be attached to the internal staffing costs incurred by an organisation. However, open-source software is often free of any licensing costs. When fees are charged, they are typically very low cost: for example, an enterprise licence for Hugging Face starts at $20 per user per month. This can significantly reduce licensing costs, making AI available even to small businesses or startups.

In contrast, commercial AI platforms can charge very high prices and often include limits on use as well as hidden fees such as usage fees, costs for customisation or extra charges for certain features. 

Customisation and flexibility

Open-source tools can be highly customised, often with the limit to customisation being the organisation’s skill level. This means that you can tailor the AI solution specifically to your needs and requirements rather than those that a third-party vendor thinks are right for you (and which, in practice, are right for them).

Open-source tools allow for the modification of algorithms and models to suit a specific use case, whether in terms of performance, accuracy or functionality. Algorithms can be fine-tuned and models retrained on your own data. Functionality can be extended to fit specific business needs, something that may be limited (or a very expensive option) with commercial tools. 

Control over the development process

With an open-source tool, you have full control over the features you want to use and the deployment timeline you want to work to without being dependent on an external provider’s updates or limitations.

There is no vendor lock-in: you are not tied to a proprietary ecosystem that limits your choice of other tools to use. And because you are not trapped by a contract, it will be easier to find a different solution if the open-source tool you are developing proves unsatisfactory.

Transparency is also a benefit. You will have complete visibility into how the AI tool is built, including the pros and cons of the algorithms you decide to use. This can make debugging, auditing and improving the tool easier. It also increases the explainability and understandability of the tool, meaning that the people using it will have a clearer idea of why particular outputs are being delivered and whether they are appropriate to a particular set of circumstances. This will help ensure your AI solution adheres to privacy or ethical guidelines.

Finally, security concerns are minimised as you can manage your own data without handing it over to a third party.

Community and collaboration

Many open-source AI tools have strong developer communities that contribute updates, offer support and add new features, as well as discuss potential new use cases and issues such as unwanted bias or low levels of accuracy. An organisation that is an active part of such a community will not only benefit from a tool that is improved with continuous innovation but also will be strengthened by the association with creative problem-solving and future-orientated thinking – benefits that are significant in their own right. 

Disadvantages of the DIY approach

Of course, building your own tool also comes with disadvantages. Chief among these is the resource-intensive nature of an AI development project. Developing and maintaining an AI tool can require a significant investment of time and money.

In addition, you need skilled governance specialists, data scientists and DevOps professionals to develop, train and deploy the AI models effectively, safely and ethically. If the right talent isn’t readily available, it will slow down a process that is already slower than simply buying in a service from a third party. This can mean extra delays to product launches or service updates.

Another issue to consider is ongoing maintenance. If you build a tool yourself, you are responsible for maintaining and updating it, including bug fixes, model evaluation and enhancement, and responding to any issues faced by internal and external stakeholders. Even where there are open-source communities to provide help, it’s not the same as having the dedicated support you’d get from a company when licensing a tool.

Tooling up for open-source AI

This means that any organisation embarking on a DIY AI course needs to take several considerations into account – considerations that may require some planning before any real development work is started.

The most important consideration is skill levels and expertise. Even though open-source resource tools offer a middle ground, using them requires a team with some technical knowledge in machine learning and data science. While they simplify a lot of processes, expertise is needed to fine-tune, customise and maintain the models.

And it isn’t just the engineers and scientists involved in developing the new AI tool that will need skills. Governance is a very important part of any AI project, and this means that people at the top of the organisation must be sufficiently knowledgeable to understand the potential hazards, navigate any legal and ethical requirements and provide the resources needed for success.

There will also be a requirement for maintenance and support once the new AI tool has been implemented. Open-source tools don’t offer dedicated support, so you may need to rely on community forums to handle troubleshooting. This means developing a positive relationship with the community and being seen as an active contributor rather than merely someone who is looking for the expertise of others.

Another important consideration is integration. While open-source tools are generally very flexible, integrating them with existing systems can sometimes be more complex compared to plug-and-play commercial tools that are frequently designed to allow simple integration with other commonly used commercial tools. Appropriate skills, patience and the support of the open-source community will be essential. 

Choosing open-source AI

Opting for an open-source AI tool may not be simple, but it can often be better than signing up for a commercial solution. While a system provided by a well-known company may be considered a low-risk choice, this is often a superficial judgement: it is likely to be more expensive, especially in the long run, and less flexible. It may also be harder to use effectively as the way the model makes decisions is harder to understand than if the model had been developed internally.

Opting for open-source AI tools might seem like a forbidding challenge, and any organisation that does so will need employees with appropriate levels of skill. However, the benefits in terms of cost, customisation, control and community are considerable.