Thinking from a technology-first perspective is a pitfall that keeps cropping up whenever new technologies become available. “We have to do something with AI” is a common reflex among organizations now that there’s so much talk about AI. The fear of falling behind quickly leads to doomed pilot projects that can’t be scaled up. “Start with the question: ‘Why?’” ‘Why’ is a far-reaching question, but actually a very simple one. Do we have a very clear understanding of why we want to do this this way and what we want to achieve? Are there other options, and have we evaluated them thoroughly?” Brekelmans asks her audience. “And let’s not forget the side effects. When you choose option A, you’re not choosing option B or C, and that may lead to other consequences. In many cases, you don’t know exactly what those consequences will actually be, but you can think about them and try to gain a clearer understanding.”
Talking Instead of Checking Off Lists
Brekelmans advises that these kinds of questions are best addressed in a conversation with all stakeholders and not left solely to an ethics committee. She points out that for many people, complying with ethical requirements is equivalent to checking off lists and going through frameworks. This turns it into an administrative burden, causing people to quickly lose sight of why the rules were established in the first place. Brekelmans sometimes even recommends choosing a different word than “ethics” because it can come across as heavy-handed.
“Ethical awareness must be ingrained in every employee, so that they realize on their own: This isn’t right because it’s problematic for such-and-such reasons.”
Brekelmans has already introduced an unconventional approach in various settings. For meetings on this topic, she brings a collection of exotic stuffed animals and asks everyone to choose one. They must give the stuffed animal a name and assign it a trait. Participants must then speak from the perspective of the role they’ve assigned to the stuffed animal. Thanks to that role, people are more likely to ask very uncomfortable questions. “‘I have this very critical bat in my hand, so I’m not asking the question—this bat is.’ The approach might sound silly, but it really works,” she emphasizes.
Make sure you’re talking about the same things
In these conversations, it’s also very important that everyone is using the same definitions. The term “AI,” for example, is very broad, and people often have very different ideas in mind, which leads to talking past each other. The same applies to concepts related to ethics. “What do you mean by ‘governance,’ what is ‘responsible,’ and when is something ‘fair’? What is bias, and how do you define ‘autonomous’?”
She refers here to a statement by the Polish-American thinker Alfred Korzybski, the founder of general semantics. In 1931, he introduced the phrase “a map is not the territory,” emphasizing that a word is not the object it refers to, just as a map is not the landscape itself. A map is necessarily a selection and a simplification of reality and is therefore a useful tool. But anyone who treats the map as if it were the whole of reality is making a fallacy.
Brekelmans also referred to the ideas of the Austrian-British philosopher Ludwig Wittgenstein, who argued that language is an activity in which people participate. “I’m speaking here now, and you’re trying to understand what I’m saying. From that, I hope, a shared understanding emerges, because only then does it make sense. Moreover, communication only works if we can question things,”she explains. People must dare to say that they don’t know or understand something; that reduces the risks of decision-making.
Using AI Proportionally
The best-case scenario is when every employee becomes aware of the consequences of AI in this way and, as a result, handles it more consciously. “At Vattenfall, we’ve made some funny videos for this, in which someone in a group of colleagues asks what the weather is like. Instead of looking out the window, one of them asks Copilot. Then, in the video, a voice-over explains that at that moment, 27 data centers are consuming that many gigawatt-hours, and we see all these sad animals because the water is being drawn from their stream. All of that just to find the answer to the question.”
Brekelmans argues that this example doesn’t mean AI shouldn’t be used, but that responsible AI use means always asking whether the use of AI is actually necessary and proportionate.
Management also bears responsibility in this regard. It is important that employees not simply be forced to use AI. “Of course, at a certain level, you can expect people to be open to new developments. But there are people who are less willing to go along with that and have a different opinion on the matter. They are valuable to the organization. After all, even among customers who, for example, contact customer service, you’ll always have people who have different contact preferences and may also purchase different products or services.”
Employees with differing opinions are therefore valuable because they reflect an organization’s customer base. “My biggest fear is that soon everything will be run solely by AI. So I hope the trend reverses and people can take back more control. Ultimately, this will likely require clear laws and regulations to ensure that choice remains possible.”
