Ms Scherer, as the co-founder of Delta Labs, you help companies on their journey into the world of AI and have in-depth insights into the challenges they face. In which areas is the need for support currently greatest, and how has this changed since your start-up was launched in 2022?
When we started out, AI literacy was the greatest need. Initially, many companies simply wanted to understand what AI actually was. So we organised special training courses for C-level managers because they wanted to gain a clear overview. Prompt engineering was also a major requirement for lots of companies at the beginning – many wanted to know exactly what AI was and how they could use it effectively. Later on, the number of requests for specific use cases increased. Companies wanted to know how they could deploy AI practically in their work areas. Now, in a third phase, we are seeing that companies have developed a better understanding of large language models and their application. We are increasingly focusing on implementing the technology in practice. Requirements such as fine-tuning, secure setup and integration into in-house infrastructure are now becoming more important. The data required and its quality is becoming more important too.
Many companies are working on implementing specific AI projects. What different approaches have you encountered in this regard?
In our work, we come across very different approaches. The projects launched the quickest and driven forward fastest often came directly from the C-level. The message was clear from the outset: "Generative AI – we have to do something." This was then passed on to others based on the outlook of: "Guys, see what you can do with it." Although these projects started very quickly, they were sometimes difficult for us to implement because the employees were often not involved at all. The technology was introduced from the top down, so to speak, and people were suddenly faced with the task of having to do something with it without any real preparation. By contrast, other projects were set up in a completely different way, more from the bottom up – referred to as grassroots initiatives. Here teams decided: "We want to use generative AI and to understand what this means for our sector." These projects got off to a slower start, but in the end were implemented in a completely different way. You could tell the people involved were well motivated and had a clear understanding of what the technology was capable of in their field of work.
Although these projects started very quickly, they were sometimes difficult for us to implement because the employees were often not involved at all. The technology was introduced from the top down.
At many companies, AI is often introduced by external software providers, such as Microsoft or Google, by integrating it into existing software. How do companies ensure that these technologies actually create added value for everyone?
My impression is that companies which introduce Microsoft Copilot, for example, or develop and roll out internal GPT models, often initially encounter a big issue: they have a great AI model, but it is hardly ever used properly. A few AI enthusiasts are jumping on board, but the majority of the workforce remains on the sidelines. If a company simply provides a generic GPT model, the marketing or communications manager often does not know how to apply it to their specific tasks. That's why it's so important to introduce employees to the technology through personalised training programmes that are coordinated with their individual area of work. In a way, it reminds me of the launch of Excel many years ago: if you give people Excel without explaining how it works or what benefits it provides in their job, they won't use it. In the past, many employees were still using pocket calculators, and it was only targeted training that enabled them to apply the new capabilities in their routines. Over the next three to five years, we will undoubtedly see more and more of training programmes like this. We are already seeing growing demand in this area.
The problems described (e.g. lack of involvement of all employees, top-down implementation) seem to mainly affect large companies with a large number of employees and numerous teams. Do you see any differences between start-ups, SMEs and corporates?
There's a very clear difference between large companies and small start-ups. Large corporations have huge IT departments that can develop their own initial AI solutions. However, cultural change is much harder to bring about there. These companies already have complex structures in place that are difficult to align with AI. In large companies, there is also often resistance or a certain degree of reluctance, as many employees fear their jobs could be put at risk by AI. This is why they are much slower in implementing AI solutions. In contrast, it is much easier in an SME or a small start-up like ours. We grow commensurately with AI. Before we hire new employees, for example, we first check whether an AI could do this job for us. The mindset here is simply different – we integrate AI from the outset which makes our processes faster. In my view, AI can provide a great boost at the moment, especially for smaller companies.
The issue of fear over the negative consequences of AI is also apparent in our surveys. Where do you think this fear comes from? And how should companies deal with it?
I think a lot of the fear stems from the fact that people are afraid of being replaced by AI. They start to wonder whether their job is still needed and what place they have in society. Such concerns are completely understandable. That's why it's important to communicate to employees what AI can and cannot do. A good way to illustrate this is by using a simple task such as: "Give me a list of ten words containing nine letters." AI cannot currently perform this task. It shows that in many cases AI doesn't possess 'healthy human logic' and still fails at simple tasks. Understanding this helps to dispel the notion that AI can replace everything. It is important to convey to people that AI is a tool which can help to improve efficiency at work and make life easier, but it can't do everything. There is still a need for people to scrutinise things critically and to apply the 'human in the loop' approach. The people who use AI the most are often the least worried, because they understand it is a tool that improves their work, but does not replace it. In society, we need to develop a more realistic view of AI – e.g. as we have already seen with previous technologies, such as PCs and Microsoft Excel. They have simplified many aspects of our lives, but were never a substitute for humans. We also need to start training much earlier in order to teach people how to use AI and make them less anxious about it. We need to make clear what AI can do, but also what it cannot do, in order to promote a healthy, realistic perception and dispel fears.
In many cases AI doesn't possess 'healthy human logic' and still fails at simple tasks.
In your view, how long will it take before generative AI will actually deliver profound added value in companies?
This will undoubtedly take some time. We are at the very beginning of this development. It will take several years before the majority of companies and employees have seamlessly and efficiently integrated these technologies into their work processes.
More information on this topic:
The full analysis, including figures, use cases and assessments of the AI transformation in the world of work, can be found in the new GDI study entitled 'Smart and human - AI between the poles of efficiency maximisation and customer centricity' (German only). Download now free of charge at gdi.ch/smart-und-menschlich