We recently sat down with our CTO, Bruce Kirchner, to discuss legacy modernization and system transformation in today’s rapidly evolving technological landscape. This is the second of a 2 part series discussing the state of application modernization, the current and potential impact of artificial intelligence (AI) and modern technologies, and how organizations can devise an AI implementation strategy and move forward with AI adoption while finding value amongst the hype.
At EvolveWare, we’ve been working with AI and machine-learning (ML) technologies for over 15 years. As these intelligent algorithms lie at the foundation of our Intellisys platform for application modernization, we’ve learned a thing or two about where to invest AND where to be cautious when it comes to using these technologies in a safe, secure, and value-driven way. Here are my top 3 considerations for developing an AI implementation strategy within an organization.
AI is not ready to be the solution for everything
A leading figure in the AI space, Rodney Brooks, recently discussed the limitations of AI stating that “people see it as so capable they even want to use it for applications that don’t make sense.” It is important to understand both the benefits and constraints of AI technologies today and to plan your AI adoption strategy wisely. With the current state of technology, purpose-built models are really where you’ll derive the most value.
If we look at AI models today, they provide a lot of efficiencies when it comes to tasks like summarizing and pulling out useful pieces of information from a large dataset or corpus of text. For example, most application modernization projects require an introductory assessment phase of the organization’s software portfolio or at least of the systems in question. To determine where to take these applications and how to plan the project, you must first understand what they are currently doing, how they are structured etc. If this kind of task were to be done manually with a human poring over millions of lines of code, not only does the task become extremely time-consuming but it is also harder for any person to “see the forest for the trees” and determine information important for the project. AI and ML technologies reduce time, cost, and risk for these types of tasks, and at EvolveWare, purpose-built ML algorithms for application understanding were the foundation of what is now the Intellisys application modernization platform.
That being said, there’s been a lot of chatter about trying to use the new generalized large language models (LLMs) to quickly develop an understanding of large applications or other specialized use cases in order to further reduce time needed and improve explainability. But at this point, the technology does not give the types of results needed for custom use cases. While there are some techniques such as retrieval augmented generation (RAG) where external datasets are referenced to help models improve their accuracy rate, ultimately size of the dataset needed for a use case must also be taken into consideration. For example, the context window is not large enough for these LLMs to process an entire program or application that is likely hundreds of thousands or millions of lines of code. And trying to break up that code into chunks and piecing the results back together leads to disjointed results. So it’s essential to evaluate all the characteristics of a use case against the actual capabilities of AI technology today before assuming that AI will improve the manner in which that use case is performed. While we fully expect the technologies to mature, that will take improvement of the hardware, software and costs around it which will take time.
Data – especially data important for your business – is power
It’s paramount to use data related to your specific business to train AI models for your use case, not just generalized data that is openly found on the internet. This is why techniques such as RAG have become such a standard in enterprise Gen AI implementation strategy.
When it comes to application modernization, this business specific data is also stored in your application code. EvolveWare’s automated documentation generation capabilities can help organizations uncover that data by giving you an understanding of the various components of the application. This information could be used to build an AI solution to help organizations gain specialized knowledge about their specific applications as they look to maintain these applications or plan and implement a modernization initiative.
At the same time, organizations need to take into account privacy related concerns with the data that they think about using when training models. For example, healthcare organizations in the US have to be careful with data that would fall under HIPAA (Health Insurance Portability and Accountability Act). This is because information leakage is the biggest security concern when it comes to AI technologies. It is important to keep data that would fall under a restricted category separate from data that models are being trained on to avoid ethical and costly repercussions.
Let AI shine as the intelligent assistant, not the ultimate decision-maker
As we discussed above, AI technologies in their current form have both strengths and weaknesses. The important component is to recognize these strengths and use them to complement what we do well as humans to generate maximum efficiencies. At this point, AI can help accelerate processes that involve poring over thousands if not millions of data points and summarizing or making sense of that information to help with decision-making. But the ultimate decision is still best left to the human.
The recent mishaps related to “hallucinations” are a prime example of this. A hallucination is when an AI model misleadingly presents a response that is false as true based on the data it was trained on and how its predictions are designed. Some famous examples of these are Google’s first Bard demo where the model stated that the James Webb Space Telescope took the first pictures of a planet outside our solar system (this actually happened 16 years prior) and when Microsoft generated an article that listed a food bank as a top tourist destination in Ottawa, Canada.
Now had a person been involved in fact-checking or editing this content prior to publishing it, these errors could have been caught. And that is why we have and will continue to focus on building human-in-the-loop processes as we incorporate AI/ML into the Intellisys platform. Human-in-the-loop processes as we think about them are where humans review outputs and can provide feedback to the system. For example, one of the prime areas where we believe AI can add value for our product is in further speeding up the processes involved in the extraction of business rules. These include understanding and renaming cryptically-named variables and summarizing and categorizing rules automatically. However, once these “suggestions” are made, we do believe that there should be a person involved to accept or reject the suggestions prior to moving forward – i.e. the person uses the information presented to make the final decision on the variable rename, rule category, rule summary, etc.
Partnering for the future
There is no question that AI/ML technologies will continue to reshape the internal and external experiences that we build. And as hardware and software advancements are made and costs decrease, tasks that the technology cannot appropriately handle today will possibly become routine. However, as we think about product roadmaps and business decisions for the near future, it’s essential to be pragmatic and play to the technology’s current (and near future) strengths. While powerful, AI remains limited in crucial ways – lacking true understanding, common sense reasoning, and adaptability to novel situations. The most effective approach is viewing AI as an intelligent assistant rather than an autonomous solution. By leveraging AI’s strengths in processing, pattern recognition, and prediction alongside human expertise, creativity, and judgment, we can achieve far more than either could alone.
EvolveWare has been recognized as a Representative Vendor in the Gartner Market Guide for Mainframe and Legacy Systems Migration and Modernization Tools. To learn more about EvolveWare, our AI implementation strategy, and AI adoption within the Intellisys platform, please contact us.
