Over the last few years, low-code solutions have become important to help organizations widen programming accessibility to a broader base of users and help fill development gaps. And these technologies are still gathering steam—Gartner estimated the market for low-code development technologies would grow to a total of $26.9 billion in 2023 and that 65% of applications would be developed using low-code by 2024.
However, AI-based code generation is set to further level the playing field with advanced auto-completion and AI coding assistants. Take GitHub’s Copilot, a helpful AI pair programmer, or OpenAI’s ChatGPT, which can generate complex functions given natural language prompts. Other AI tools trained on UIs, like Galileo, can derive frontends for any project in a matter of seconds. So, are low-code development platforms still relevant in this new paradigm of AI-driven software development? If so, how will low-code solutions and AI coexist?
Well, generic algorithms can only go so far in creating truly unique and innovative experiences that solve real-world business cases. And since most programmers don’t have the skills to train machine learning or deep learning models from their own datasets, this is one area where low-code solutions could lower the barrier, enabling users to tag unstructured data, generate models, run simulations and share reusable AI across departments.
I recently chatted with Johanna Pingel, MathWorks AI product marketing manager, to discuss the role of AI in low-code programming. According to Pingel, low-code solutions can lower the barrier to AI functionality, both in quickly integrating AI-driven functionality into applications and seamlessly generating new AI algorithms based on unique datasets. Below, we’ll consider some additional use cases for AI and how low-code platforms may work alongside it.
Use Cases For AI
First off, what are application developers using AI for these days? Well, AI has an increasingly wide range of uses. It began with computer image and processing, such as identifying cats and dogs or postal codes. Now, voice recognition, voice-to-text and text-to-voice are commonplace, and AI is being used to create ever more sophisticated applications. One area that Pingel is excited about is applying AI to help with battery life management, predicting the lifespan of batteries for electric vehicles and consumer electronics.
“Anywhere there’s data, there’s an opportunity to create models and make predictions based on that data,” said Pingel. This especially rings true for scenarios involving data and visualization. For example, consider visual inspection in a pharmaceutical production line to ensure there are no defects with any pills. Another use case is visual inspections of parts in automotive assembly lines. Applied to these scenarios, AI can remove grunt work and make production safer and more efficient.
ChatGPT is another example of an AI application quickly becoming mainstream, not only for content creators but for scientific research. For example, ChatGPT can spin up MATLAB code and create functions and unit tests, which can be copied and pasted into IDEs. However, code written by ChatGPT should still be checked and verified by engineers and scientists to ensure its accuracy, said Pingel.
How Low-Code Lowers the Barrier to Bespoke AI
Low-code solutions can bring many benefits—these platforms help to speed up tedious tasks and level the playing field for those with different coding abilities and desires, said Pingel. Low-code platforms can also be a great tool for engineers looking to incorporate AI into their business systems. For example, having a low-code method could be helpful in generating code for machine learning algorithms and new models, which can help companies create their own algorithms and incorporate them into their own platforms.
More people are interested in incorporating AI into their applications, but the use cases are growing increasingly more sophisticated. Businesses may start with simple needs, such as identifying faces or identifying text from handwritten letters. These rely on simple datasets, and algorithms already exist to accomplish most of these tasks. Off-the-shelf AI is helpful at the start, but eventually, companies tend to want to use their own data and settings far more specific to their applications, said Pingel.
For example, a more niche case could be as specific as identifying cracks in a windshield. Or, in a financial industry, it might be analyzing mass amounts of loan applications to determine approval rates. In these situations, using your own internal data becomes necessary to create AI that understands the situation at hand. Yet, the hard part is assembling this data in a way that is accessible for a model to comprehend. Analysts estimate that upwards of 90% of data held within enterprises is unstructured, meaning it requires a heavy amount of tagging to become usable for AI/ML.
Low-code is a perfect solution to avoid the point-and-click aspects of cleaning, cropping and structuring this data, said Pingel. This can introduce automation into the process of labeling data while still involving some human intervention to fix mistakes. Machine learning models can take a while to train and simulate, underlining the need for a fast way of iterating on models. A common platform could help companies develop custom machine learning algorithms and models and keep track of experiments. This helps to ensure that the most effective model is used for the final application.
End Benefits of Democratized AI With Low-Code
Overall, low-code platforms can provide a number of benefits that lower the barrier to unique AI creations. Low-code platforms like MATLAB and Simulink are useful for structuring and labeling data and quickly generating ML models. Low-code also makes it easier to bridge different programming languages and quickly generate code for a variety of different languages, such as C, C++ and Java.
Additionally, low-code platforms enable users to collaborate easily with other users. But low-code doesn’t mean no-code, Pingel explained. In fact, “you can learn to be a better programming through low-code tools, too,” she said. For example, it could enhance understanding of the structure and syntax of code.
In general, democratized AI is an important capability for businesses to take advantage of. Although AI-based solutions are coming to market to displace old tools and processes, there is still an argument to be made for low-code—namely in empowering non-technical users and lowering the barrier to creating custom AI-driven functionality.