We’ve discussed how just about any industry can benefit from integrating artificial intelligence with its business processes in “Practical Use Cases for Artificial Intelligence." Now, it’s time to look at some of the challenges associated with using it, as well as where SAP tools and platforms might help you overcome some of those obstacles.
But before we look at the nuances, hard facts, and strategy to move forward, let’s go back to basics for a moment.
Back to Arithmetic Basics
One of the central reasons why ASUG members have difficulty implementing AI advancements in their real-world operational systems is because the development of AI requires new skills and adoption of new tools. And in many cases, it requires a re-engineering of the existing installed base of software and hardware for a given use case. The skills to make all those things happen don’t necessarily exist today.
The other skills issue comes down to mathematics. Linear algebra, statistics, and probability form the foundation of the machine learning processes that go toward building AI.
Software developer and industry commentator Janakiram MSV wrote, “The availability of reusable math libraries and functions relieved developers from doing math the hard way. An average programmer doesn’t get to deal with mathematics on a day-to-day basis. Only a few gifted developers have the natural intuition to math. To master machine learning, mathematics is mandatory.”
So, we might argue that developers have had it too easy. Or at least slightly too easy given that they now need to go basic to basics.
AI Chatbots Without Coding
The skills challenge explains part of the rationale behind the way SAP is developing its SAP Conversational AI platform. Sebastien Beghelli, SAP’s Conversational AI Community Manager, wrote "How To Create A Facebook Messenger Bot Without Coding" using SAP technologies.
His vision of a working bot or chatbot is one where some of the “math-you-forgot” grunt work is bypassed through the automation layers SAP can offer in its platform. These are bots that can chat with users and, more impressively perhaps, can start to work as the admin function for cloud services. But SAP doesn’t necessarily think all bot development should be completely dumbed down and offers advice on advanced bot building patterns too.
As we’re seeing in a lot of use cases now, SAP advocates starting with a build concept hinged around the creation of a minimum viable product (MVP). Once that is live, you can start to develop something more sophisticated.
Garbage in Equals Garbage Out
Let’s go back to the challenges with AI and machine learning again. Another key reason it’s so tough to build these new systems is the fact that to achieve good AI, we need good data.
They say garbage in equals garbage out for a reason. When we say that AI needs good data we mean “good” in the sense that it should be:
- Accurate (in terms of veracity)
- Timely (and potentially time-stamped depending on the use case)
- Deduplicated and verified
- Big enough (in terms of the volume of the dataset)
- Complex, variable, and diverse enough for the job at hand
- Valuable enough to fit into the business use case for AI
Looking at ASUG members’ specific needs, this is perhaps one of the toughest areas for SAP to help with. While it’s true that SAP has data cleansing and deduplication tools in SAP HANA and SAP Agile Data Preparation is a product offering in its own right, the individual data teams must get their own houses in order, or at least speak to an SAP (or other external) consultant to help make that happen.
Scaling the Heights
Another key challenge is scale. It’s complex enough to think about building new software functions in your live IT stack that will benefit from AI. We’ve already mentioned the MVP approach and the need to build one brick at a time. Taking those AI-enriched MVP products upward, pushing them live to production, exposing them to live data, and then expecting that to scale upward to whatever level the business needs is an even bigger ask.
As Google Chief Decision Intelligence Engineer and AI guru Cassie Kozyrkov wrote, “Pulling off automation at massive scale is hard. That’s the stuff I make all kinds of fuss about, but it’s not an AI-specific problem. Doing anything at global scale is always a beast of complexity, from making burgers to delivering search results. However, at an individual scale, it can be pretty easy, unless the tools aren’t user-friendly.”
It’s safe to suggest that SAP has most (if not all) of these challenges on its radar. The question for ASUG members reliant on SAP salespeople, technical engineers, and consultants for operational success is: Just how will SAP prioritize its approach to automating which things first? We said in our first piece on SAP that the company is aiming to bleed AI DNA throughout its entire product line. Even if that were possible, it wouldn’t manifest itself at the same level in each product group at the same time with the same level of functionality.
The Neural Bottom Line
If AI is tough, AI at scale is tougher, and crafting AI into specific business use cases is even tougher than that. This is especially true if you have to do it while also ensuring your data is clean, deduplicated, and appropriately time stamped. Accomplishing this while remembering enough college-level linear algebra to perform data functions might just be the toughest part of all.
What truly makes AI especially difficult to pull off is that it seeks to replicate the neural functions of a human brain, a piece of engineering that we don’t even fully understand yet. That said, we hope these hurdles don’t put you off and that ASUG members will continue to embrace AI systems, tools, and functions as they emerge through SAP platforms and beyond.
Interested in learning how to use predictive maintenance and machine learning in the Enterprise Asset Management (EAM) space? Register for our on-demand webcast and learn where to start.