Today’s application development platforms are becoming more automated and streamlined, relying on visual programming and low-code/no-code capabilities. This has the power to make digital workflow creation more efficient, allowing more knowledge workers to participate in streamlining productivity within their enterprise. Among business automation trends, Artificial Intelligence (AI) stands as a key strategy to get a leg up. And as AI evolves, low-code platforms are beginning to incorporate more of this logic to make applications more intelligent and proactive.
I recently met with Dave Wright, Chief Innovation Officer, ServiceNow, for an update on how AI is shaping the future of low-code development. To date, AI models have gotten good at predicting user behaviors and monitoring ecosystems for security or stability problems. But going forward, Wright foresees the next evolution of AI being more conversational and prescriptive, as in not only identifying problems but automating informed solutions in real-time.
How Knowledge Workers Can Utilize AI
By using pre-built components and a visual programming model, low-code can significantly reduce development burdens. Furthermore, utilizing AI within a low-code environment could accelerate the overall application development process. It can also make end-user-facing experiences more streamlined.
Perhaps the most obvious use of AI is to improve customer support with virtual agent technology. Typically, when an application requires a user to input data, it provides a form to fill out. With an AI-based process, it leverages Natural Language Understanding (NLU) to turn forms into conversational interfaces. For example, the city of Los Angeles recently incorporated virtual agent technology into a COVID-19 testing app. “When you integrate AI capabilities into a low-code tool, it elevates the user experience,” described Wright.
Data scientists are still required to develop the core engine that runs such processes. However, the system still requires users to design the conversation, along with what actions to trigger based on outputs. Interestingly, this is one place where citizen developers could shine at the low-code level. “The skill is not in building the conversation, but in determining what conversation to have,” Wright described. This offers a way for not only professional programmers but analysts working at SMEs to participate in refining their application’s use of AI.
Conversational App Development
Core machine learning can also infer certain things from a description, aiding categorization efforts. It’s relatively easy to build that into a system, said Wright, but when you create an application from scratch, you need a significant amount of data to train machine learning (ML) models. This is partially why he sees organizations mining what structured data they have. Sometimes this involves synthetic AI data models which repurpose data from other sources and de-risk Personal Identifiable Information (PII).
Wright also highlights current hyper-automation efforts that, again, improve user-facing experiences by linking together various AI. This could enable a system to trigger workflows based on a conversation, and simultaneously collect more information to integrate into its evolving knowledge systems. “How can we enable more AI innovation within the code that is created? How can we include that within applications we build as well?” asked Wright.
Conversant AI may not just power user experiences; it could empower the application developer themselves. We’ve seen the power of algorithms like GPT-3 to infer complex instructions from simple commands. For example, the DALL-E neural network can generate custom images based on a single sentence. Similarly, Wright foresees AI driving low-code development through the ability to verbally describe applications. This would go beyond graphically-based design to incorporate voice-to-text commands that generate new applications.
From Predictive to Prescriptive in Real-Time
Yet, current low-code platforms are still far away from constructing complex applications and workflows from verbal commands. Instead, AI within low-code is in a more embryonic state. It currently involves things like engineer behavior analysis, software generation, and code recommendations. To necessitate the new age of AI in low-code, we will require a few things.
First, we’ll need increased awareness of new AI capabilities among a new generation of programmers. Rising user expectations will also beckon more advanced workflows. Furthermore, increasing the communication ability requires the ubiquity of 5G networks since much AI-related processing will occur in the cloud, far away from the client device.
Eventually, stronger connections to remote cloud processing will open up some high-impact results, resulting in real-time AI, predicted Wright. Real-time AI, he said, involves not only training the ML model but using input data to change what those models look like in real-time. These computations require highly performant GPUs, which edge devices lack. The necessity to process real-time AI in the cloud emphasizes the importance of reliable Application Programming Interface (API) integrations and ease of network communication.
The Future of AI is Proactive
With the proper infrastructure in place, the next generation of AI will be increasingly proactive in not only identifying problems but automating solutions to such problems. “We’re already at the point of AI being good at being predictive,” described Wright. “The next stage is migration to prescriptive, where not only the problem is identified, but a solution is suggested.”
For example, consider an AI that detects a network outage. Instead of simply notifying engineers to respond manually, it could perform the next best actions to remediate the problem. This could involve automatically rebooting the system or releasing semaphore implementations. Additionally, this could be quite useful in customer service management. Instead of making a customer read through dense documentation or wait on hold for a representative, a proactive AI could intimately understand customer complaints and walk through the next best actions to remediate their issues.
The future of AI seems a bit like Data from Star Trek. When confronted with a major decision on the bridge, he’ll often advise Captain Picard with a percentage chance of success or failure for multiple courses of action. Similarly, prescriptive artificial intelligence could respond to application support issues with informed recommendations, having analyzed thousands of similar situations and calculated the likely outcomes of each. Such actions could involve human intervention at critical junctures or run completely unsupervised.
Of course, building out such a system will take time and effort to gather a comprehensive foundation of use cases, understand real-world edge cases, and extrapolate outcomes. It will also require data cleansing, said Wright, as feeding incorrect data into a system will produce inaccurate models, which could cause more hassle than its worth. Furthermore, to retain control, fully automated systems will require the option for human intervention at significant “digital circuit breakers.”
AI is set to transform how we develop applications and alleviate some frustrating customer service gaps in the process. This will be made possible through more conversational interfaces and real-time modeling. It will also be interesting to track how the future AI will become more perspective in identifying issues and making informed decisions. The next question is — how much will we allow this AI to decide on its own?