Implementing AI to existing systems suffers from an image issue: it’s not as easy as it looks

This is a fact our interviewees know well. Eija-Leena Koponen is thoroughly versed in the AI world, having co-founded Renessai, a new company assisting businesses embed AI into their strategy and operations. Carl Holmqvist is a seasoned Software Developer and Data Engineer at Nitor, currently imbued in helping his client implement new AI solutions to ease day-to-day operations and speed up processes.

Both Koponen and Holmqvist credit ChatGPT, as in the user interface with a GPT model, with igniting new interest and cultivating business opportunities with generative AI and language models. The deeper question is how it will impact roles traditionally associated with data-centric work.

“The role of the Data Scientist has certainly changed. Generally speaking, a traditional Data Scientist is someone who understands how data needs to be measured and applied using different models. Someone with a grasp on mathematics and statistics, but also of the theory and structure behind technical applications. An essential part is the aptitude to translate that knowledge into business cases,” Koponen begins.

“What’s truly revolutionary about current development is that big tech is providing the productionised versions of large language models, meaning chat interfaces or easily deployable and managed endpoints for people like me.

As a result, a Data Scientist doesn’t necessarily need to spend so much time on the modelling side anymore, and large language models can also function as a medium to make technical details more approachable to a wider range of people.”

The overpromise of AI’s ease of use

For Holmqvist, using large language models and AI assistants represents a relatively small change for his daily work, but he has certainly noted a shift in attitude towards AI:

“As AI has become more accessible, most people have at least tried AI tools to see what they can create. We’ve seen awesome demos by companies showing off all the impressive things their technology can do, but that has also swayed expectations on what people assume is easily accomplished by using AI,” Holmqvist states.

“I feel expectations on what can be delivered have become a bit skewed. People see all these fancy presentations, but what’s left out of the equation is the understanding of the underlying work. Solutions tend to be very complex and require understanding what kind of data is most applicable. What is the right architecture, what steps are needed in the pipeline, how do we preprocess the data and how do we guarantee the accuracy of the responses? Just giving the data to a GPT won’t solve or produce anything.”

Koponen and Holmqvist note that while AI companies often present their technology via highly impressive showroom presentations, these demonstrations are often purpose-built for promotion and don’t reflect the current usability of AI applications. This has led to frustration on the client’s side, as real use cases are limited, and implementation turns out to be much more complex than anticipated.

“Clients might request new tech without expecting the challenges it brings. From a management perspective, there is this expectation that AI implementation is something everyone can do. Just log on to ChatGPT, get some code, and this will enable everyone to do this, that, and the third thing. But once the tools are in-hand, that’s when the realisation occurs: okay, this was a bit more difficult than expected,” Holmqvist outlines.

That’s when the real work begins, and that work is still rooted in understanding of the data, business cases, and strategic goals of the company. From Holmqvist’s perspective, businesses are still in the learning phase when it comes to AI implementation, with a deeper understanding of AI’s usability still lacking.

“If you’ll pardon the analogy, in a best-case scenario an AI solution is like a tractor in the field where we used to apply a shovel. Sure, it can enable us to forage more ground with less time, but only if the people using it understand how its functionality differs from previous solutions. That requires change management, training, and time,” Koponen continues.

Understanding the data remains a key to success

Another facet is the skewed perspective of how much AI application will cost in the long run. AI utilisation is currently underscored by a dangerous notion that by simply adding AI as an all-encompassing solution, companies and organisations will require less resources, personnel, and expertise to develop and expand their offering. However, rather than attempting to apply generative AI to everything, Koponen suggests first exploring which avenues could benefit from smaller or traditional models, or where issues can be addressed by applying automation.

“I would say AI solutions can be implemented like parts of a factory, but the higher functionality of that factory requires intricate understanding of the pipeline. Let’s say AI represents an advanced mechanical arm mounting something. If it’s not connected to a conveyor belt to carry the mounted item to the next point, the machine fails its purpose. You need the entire pipeline to work,” Holmqvist continues.

“In the consulting business that both Carl and I are in, most things are based on trust. When clients put their trust in Nitor or Renessai, they need to be willing to hear us out when we advise against AI implementation. Perhaps the issue will be sorted with automation, or perhaps a language model will fit into a smaller part of the process,” Koponen adds.

Holmqvist concludes by underlining the necessity to explore the client’s intended outcome and what components advance or hinder that goal:

“AI for the sake of AI isn’t an effective way to achieve results. We need to support and guide clients in exploring what systems are already in place, what can be tweaked and what can be improved with simple calculations, and also take a stern look into whether or not AI has a real capacity to improve things.”

Looking for innovative ways to turn data into a more successful business? Read more about our services!