During the pandemic, leading companies continued to use AI to address urgent business priorities. With its value-expanding impact on internal financial optimization versus external customer experience, the essentiality of AI solutions became clear, both in and before the current crisis.
Today, studies show more than 70% of business owners consider artificial intelligence as a game-changer. Yet, less than four out of 10 companies currently use AI or plan to use it soon. The gap of perceived importance remains considerable. Being active in the field, I find there are five common myths or misunderstandings about AI within organizations.
Myth 1: AI is all about data and algorithms.
Up until, and during, the AI hype in the nineties, artificial intelligence was a scientific discipline that almost exclusively dealt with data and algorithms. Over the past decades however, the field has matured, and AI has become an integral part of automated decisioning systems that are at the heart of what we do as individuals and organizations. Consequently, a large portion of AI research, development, and implementation encompasses people and processes. I remember having a business conversation with a large energy provider in which we were talking about automated systems and data-driven methods that, driven by customer data and smart meters, could enhance their customers’ experience. One hour into the meeting, they suddenly asked: “This all looks very promising, but shouldn’t we also do something with AI?” While everything we had proposed was inherently enabled by AI implementations, the term itself was not used and therefore it was not entirely clear for the management representatives we were talking to. This illustrates we can “sell” AI without even having to mention it as if it were a buzzword.
Myth 2: I have a lot of data, so there must be something valuable in it.
If you have the combined luck and skills, you can probably cook a decent meal with ingredients that come from a randomly filled refrigerator. The real question, however, is: “What do you want to achieve?” In the example of the refrigerator, it might occasionally be an effective solution if you need to quickly fill stomachs and don’t have time to go shopping. But I would not advise you to start a restaurant based on this method. The same is true for AI projects, and probably even more so. From data selection to implementation, any AI project should start with a business challenge and not with the data.
Myth 3: I have hired a top data scientist, so AI is going to rock my business.
To stay within the metaphor of the restaurant, if you hire a three-star Michelin chef to cut carrots and peel potatoes, will this chef be motivated to stick around and boost your restaurant? The same is true for data scientists. AI is much more than complex algorithms and data analyses. It also involves essential supportive work such as harmonizing, cleaning, and preparing the necessary data sets. For this, you should hire a data engineer rather than a data scientist. At least as important, AI involves a variety of non-technical competences as well. A full-blown AI team, therefore, consists not only of data-scientists and engineers, but also includes professionals with a background in psychology, sociology, business, legal and ethics, as well as (of course) the end user.
Myth 4: AI is a black box, so I will never be able to understand and trust what it is doing.
Yes, there are unfortunately still a lot of AI implementations out there (and being newly created) of which the internal mechanisms behind their decisions are (deliberately or not) inaccessible for outsiders.
More importantly however, is the fact that there are ample opportunities readily available to avoid this. Within the field, we call this “explicable AI”: implementations that not only present you with their decisions, but also allow you to get a comprehensive insight into the reasons why this decision was made. For example: a medical AI algorithm might assist a doctor in setting a diagnosis (say for detecting a tumor), since it is far more efficient in scanning through thousands of patient records than would any doctor or human being. Yet, the doctor should always be able to understand why the algorithm recognizes a specific feature on a medical image as being a tumor or not. This way, the doctor can have the final verification and confirm or decline the proposed diagnosis based on his contextual and medical experience.
Myth 5: AI is being decapitated because of privacy regulations.
A number of people claim privacy regulations, such as Europe’s General Data Protection Regulation (GDPR), prevent the possibilities of working with AI. In fact, the opposite is true.
Regulations offer a framework that describes how you can collect, manage, and exchange data. So, rather than prohibiting it, they make sure it can happen with respect for the variety of viewpoints and stakeholders involved.
It’s because of these initiatives, and others like the numerous ethical committees that exist on national and internationals levels, that I am quite optimistic about the role of legislation in the context of AI. I often compare it with the automotive industry. Since the early days of cars (and still today), a lot of accidents happen. Yet, it has not kept us from prohibiting the use of cars. Rather, we are in a continuous exercise and dialogue to find the right balance between promoting the good that cars bring us (personal mobility that has contributed to unseen innovation and freedom) versus their negative impact on nature and society. For me, this would be an ideal situation to strive for also in the never-ending development of AI. Because the last thing we want, is that regulations would prohibit the use of AI.
Geertrui Mieke De Ketelaere is the AI program director at imec. She has a master's degree in civil and industrial engineering and is specialized in robotics and artificial intelligence during her studies. Over the past 25 years, she has worked for various multinationals on all aspects of data and analysis (IBM, Microsoft, SAP, SAS, etc.). In recent years she has started focusing more on customer intelligence environments and the use of personal data. With her knowledge of the new digital data flows (online, social, mobile, sensor, chatbots, etc.) and of big data platforms, She has been a guest speaker on digitization and AI at various business schools in recent years.