AI redefined: how to build faster and easier AI solutions

To unlock the full potential of AI, a fundamental shift in adopting AI within companies is required. The minimum investments required to start utilizing AI within a company are still perceived as too high. When we for example look at the current state of AI adoption in Flanders, Belgium, we notice many companies haven’t yet started implementing AI[1]. The main reason?? AI is perceived to be too unreachable and too complex.

Therefore, lowering the complexity and cost of creating an AI solution, will greatly help towards a faster and broader AI implementation within companies.?

Three steps towards a less complex and faster AI solution

Step 1. Data: use yours today

One of the most important parts of a good AI solution is data. No data, no model, no AI solution. Generally, in companies, before any data can be consumed, extensive databases and/or data lakes need to be constructed (which are both timely and costly). Even after these expensive solutions have been built, large parts of the data tend to be still unusable or inaccessible.?

However, with the impressive movement towards cloud solutions, it has never been easier to both collect and access data. Numerous data connections have been made available to easily connect to the broad spectrum of data storage locations being used in the cloud. These new cloud solutions make it both possible and easier to directly consume a myriad of data points within the company.?

So instead of waiting for that expensive data lake or data warehouse to be constructed, you can already start consuming the data that is within reach. Does this mean that those expensive data solutions have become unnecessary? Absolutely not, but with the cloud solutions of today, there should not be an excuse to wait to start building those AI solutions.

Step 2. The AI model: start as high as you can, go as deep as you need

Over the last few years, we have seen two main evolutions on the AI front which can greatly help us reduce the complexity of constructing well-working AI solutions: First, the enormous growth in freely available large pre-trained models for NLP, Vision, etc. Second, is the enormous growth in low-code modeling tools like Azure Machine Learning Studio.

More and more large pre-trained models[3] are becoming more easily available through open-source platforms like Huggingface-hub[4] and through API-based solutions made available by the large cloud providers (Azure, AWS, and Google Cloud). These solutions allow us to directly utilize captured knowledge and insights collected within these large pre-trained models, without needing to train them ourselves. This way we can avoid both the high costs in CPU time and the necessity to acquire the hard-to-find know-how to do so.

The other important recent innovation that can greatly help lower the complexity of building AI, is the rise of studio-based (low-code) environments in which AI models can be built without the need to write out all the code. These environments can enormously minimize the effort and know-how required to build your first AI models.

It is by combining the two AI evolutions mentioned above, that we can not only dramatically lower the start-up complexity of building AI models, but also the incurred training costs related to training AI models.

Step 3: No consumption, no value?

In our goal to lower the complexity of building smart end-user applications, I am convinced that low-code application platforms (LCAP) could be the perfect tool to use. It is clear that these low-code platforms are steadily on the rise[6]. According to Gartner, by 2025, 70% of application development will be low-code based[7]. Not only do these LCAPs provide an environment to easily build end-user applications, but many also include fully out-of-the-box AI components to lower even further the complexity to construct AI solutions. For example, the PowerApps platform from Microsoft has the AI Builder module to directly insert finished components that are built on AI models: a module that extracts the text from a picture, a module that can extract semantic-based keywords from a text, etc.

For now, I have been focusing on the PowerApps platform from Microsoft as the preferred LCAP. The PowerApps platform is not only a current leader in the LCAP Magic Quadrant[5], but also broadly interconnected with almost everything from the Microsoft cloud family. Furthermore, I am planning to later release an article that shows how easy it can be to build an end-to-end AI solution through the Microsoft stack.

The fast way or the highway?

Even though the main structure of an AI solution process hasn’t changed much in the last 10 years, the way each of the three subprocesses is applied, clearly has: First, data, if even collected, used to be only available through monolithic databases, difficult to access and mostly with poor data quality. Second, models needed to be completely built from scratch without any supporting frameworks to construct or publish them, and all of them had to be fully trained by yourself. Third and last, if you wanted an end-user application to consume it, everything had to be built from the ground up. In other words: it used to be a colossal and highly risky undertaking to create any AI solution. Luckily, by today things have been greatly improved and it has never been easier to build those AI solutions.

Does this mean that the age of large and complex AI algorithms and solutions is over? Far from it! After all, well constructed advanced AI solutions bring an unparalleled number of insights, unobtainable through other means. This is, at its core, still the biggest advantage of AI.?

With both methods in mind, I’m strongly convinced that the most efficient way to fully embrace the future of AI, is to apply both small and large AI solutions in parallel. This will not only allow your company to generate AI-based insights on short notice, but also generate the necessary experience and skills within your company to successfully construct those advanced and robust AI solutions which will carry the real fruits of progress.

With the technology of today, there is no excuse left that should still hold you back from starting with AI.?

1.??? AI barometer | Agentschap Innoveren en Ondernemen. https://www.vlaio.be/nl/begeleiding-advies/digitalisering/artificiele-intelligentie/ai-barometer.

2.??? The Future of Business Is Composable - Gartner Keynote. https://www.gartner.com/smarterwithgartner/gartner-keynote-the-future-of-business-is-composable.

3.??? Han, X. et al. Pre-trained models: Past, present and future. AI Open 2, 225–250 (2021).

4.??? Wolf, T. et al. HuggingFace’s Transformers: State-of-the-art Natural Language Processing. (2019) doi:10.48550/arxiv.1910.03771.

5.??? Jason Wong, Kimihiko Iijima, Adrian Leow, Akash Jain & Paul Vincent. Magic Quadrant for Enterprise Low-Code Application Platforms. https://www.gartner.com/doc/reprints?id=1-275QSBDL&ct=210813&st=sb (2021).

6.??? Woo, M. The Rise of No/Low Code Software Development—No Experience Needed? Engineering 6, 960–961 (2020).

7.??? Salesforce Is Leader in 2019 Gartner Magic Quadrant for Low Code Application Platforms - Salesforce Blog. https://www.salesforce.com/blog/gartner-lcap/.