Recognise the fear of cold water … And then take that plunge into the AI world
What is an algorithm? What exactly does it do? And what if machine learning is involved? Anyone who starts working with artificial intelligence has to know how self-learning algorithms work. The motto is: start small. But with the emphasis on getting started!
“A computer has a flawless memory and is incredibly fast. But is is stupid. We humans are much slower and cannot remember things very well. But we are smart. For example, we don't need to look at three billion images to be able to distinguish a wolf from a dog ”, Theo-Jan Renkema neatly summarises the differences between humans and computers. As chief IT & digital auditor at Rabobank and professor of data analytics & audit at Tilburg University, he also knows all too well that these differences regularly lead to misunderstandings.
“At the moment there is a lot of talk about artificial intelligence (AI),” he continues. “But in the majority of cases it is not about AI, but about simple algorithms that work on the basis of fairly basic formulas. No machine learning is involved at all. You can also easily check those algorithms. That is really not rocket science. ”
Theo-Jan Renkema and Mark Griep got to know each other about six years ago when Rabobank organised a meeting to explore possibilities in the field of big data. Since then they have been able to catch up with each other regularly to create AI applications. For example, they collaborated on the first attempts to develop a method for auditors to assess algorithms and AI.
Questions, questions, questions
And what about the algorithms that are a step further on and therefore fall under the heading of artificial intelligence? How do you, as a company or organisation, ensure that you keep a grip on these developments? That this kind of algorithm neatly abides by the privacy rules? That it does not adopt people's prejudices during the data analysis? That it is reliable? "Yes, those are exactly the questions that many companies and organisations immediately start discussing when it comes to the use of algorithms," says Mark Griep, country head of the Dutch branch of PA Consulting, and global head of analytics. "Because of them, many people have got cold feet about AI."
“But the longer you wait before using algorithms, the more difficult it will be to get a grip on AI and the greater the chance that you will fall behind with digitisation,” Griep emphasises. “Our motto is to start small and develop solutions for potential issues at every step. So include them by design. In this way you can create an algorithm that not only does what it has to do, but that also complies with all laws and regulations and can easily be checked. ”
Ethical Code of Conduct
“Technologically more is possible. For example, software is already being used that can control software ”, Renkema adds. "But for an algorithm to work properly, you always need human common sense." Especially when it comes to making connections and indicating what is acceptable and what is not. Computers have no ethical systems of their own. So if there are ethical rules of conduct, we really have to add them to the algorithm ourselves. ”
“Exactly,” agrees Griep. “Above all, we should not become slavish followers of an algorithm. You must not only be able to account for the type of algorithm you use, but also for everything that the algorithm does. It is therefore certainly important to thoroughly immerse yourself in the underlying technology, the wishes of the customer and the laws and regulations. But it is especially important to get started quickly and start with something small. And then develop it in a controlled manner. So not worrying about the theory of artificial intelligence, but taking a practical approach, both in terms of application and management. That will take you much further. ” (Text: Menno de Boer)