Artificial Intelligence is not magic
Magic is not the driver behind Artificial Intelligence. The rumours and mystery surrounding technology like ChatGPT leaves us with the impression of an almost otherworldly intelligence. But this is when we have to remind ourselves that it is a machine, a construction, a generative language model. To the very best of its ability, it calculates how to string words together and present answers that make sense. But that does not always happen. It guesses and “hallucinates”. If it was human, we would say that it was ranting or talking gibberish.
Sine Zambach is Assistant Professor at the Department of Digitalization at CBS, and throughout her career, she has worked on making IT and programming more exciting to children, young people and women – both during her working hours and her leisure time. She is excited by the new technology. However, she points out that we should also be aware of its limitations.
She suggests that all of us, but the younger generations in particular, should quickly enter AI’s engine room or stomach or brain or whichever metaphor best conveys the message: Place your hands on the keyboard and get going! With user instructions, online tutorials, courses or whatever takes your fancy. Because a language model like ChatGPT and other forms of Artificial Intelligence perform better when humans understand and control the technology.
“There is enormous potential in Artificial Intelligence for our society. And in order to become instrumental in creating the technologies of the future, you have to start somewhere. This is particularly true when we talk about children and young people, but it concerns all of us. We simply must get to grips with what the technology is able to do and how it works. It is imperative that we all possess a minimum of understanding,” says Sine Zambach and then continues:
“You need insight in order to not be seduced by AI. It cannot do everything, but it can be helpful if you feed it directions. By way of example, you have to understand that when it hallucinates, it is a so-called ‘stochastic parrot’, stringing words together more or less randomly, which in the best-case scenario could be regarded as chatting. But it is not a real friend with a real name like George Peter Thompson (the initials GPT). It is a machine that makes mistakes,” she says before pointing out:
“Insight into the technology can provide us with a really good understanding of what its potential is and how we can use Artificial Intelligence to create something of value to human beings.”
Machine or intelligence?
Sine just recently returned from Folkemødet on Bornholm, where she taught programming to a local class. With the help of AI, the pupils were able to create their own apps that focused on edifying subjects such as democracy or biology. Zambach is the author of the book "Kvinde, kend din kode (Woman, know your code)”, and for a number of years, she has been working with Coding Pirates, a digital club for children. She is an idealist with a practical approach to new technologies, and not long ago, she was appointed to the Board of Professional Data Ethics established by the Academy of Technical Sciences (ATV). Here, her contributions will be based on her data science background, her many years of working with AI and a general interest in data ethics.
“I’m very interested in how AI can strengthen democracy, for example, by increasing equality by way of helping people neutralise the human biases that we are sometimes blind to in, for instance, job postings,“ she explains.
My advice would be to call it ‘Machine Learning’ rather than ‘Artificial Intelligence’. It would help de-mystify the technology, which is justifiable.
There is a catch, however. In order for AI to contribute to the strengthening of democracy, “a lot more people simply have to understand how AI works.” Humans automatically glorify new technology, we take it very seriously by calling it “intelligence”.
“My advice would be to call it ‘Machine Learning’ rather than ‘Artificial Intelligence’. It would help de-mystify the technology, which is justifiable. That which we today call Artificial Intelligence is almost identical to Machine Learning. Basically, AI consists of rules and models created by humans, then worked on by a machine,” she says before elaborating:
“Machine Learning has overtaken a rule-based Artificial Intelligence in terms of technology and usefulness, but from a scientific perspective, it’s interesting to retrace our steps and supply Artificial Intelligence with more logic, that is, more rules. It is a super exciting approach, wanting to learn how to control Artificial Intelligence and make the most of the technology.”
Train your algorithms
One of Sine Zambach’s most important points is that understanding how an algorithm works will also enable you to dedramatize it. She is not trying to dismiss the difficulties in creating a really good algorithm for an advanced problem, but everything starts somewhere. You can start off small by gaining a sense of how just turning a few knobs can make a basic algorithm work.
She recommends that in addition to experimenting with a generative language model like ChatGPT or some of the numerous AI-based apps that recognise everything from tunes to wine labels, we all try to train an algorithm. You can do this with a tool like Teachable Machine or Machine Learning for Kids – and “it is okay for adults to use it as well,” as Sine says with a laugh.
“These programmes help you quite easily train an algorithm to solve a task. It can, for instance, be a case of recognising specific symbols like paper, scissors and stone, so that subsequently, you can play the game by using a camera. It takes five minutes to create a game like that. It can also be used to assess whether you have put on your face mask correctly. With the help of photos, the algorithm can be taught to do this in half an hour. The algorithm can also be used to help sort rubbish, decide dog breeds, or let you know when you have applied enough sun screen. These are all exercises that can help catch our attention and that touches on the core of how some types of Artificial Intelligence work.”
From playing to progress
In addition to projects on big data analytics and natural language processing and text analytics, Zambach and her colleagues from the Danish School of Education (DPU) and the University of Copenhagen are in the process of developing so-called scenario-based didactics where Artificial Intelligence is used to help solve a societal problem. Paper, scissors and stone can quickly transform into an app that can categorise birthmarks and let you know if you need to have it checked by a doctor to find out whether it is malignant.
Sine Zambach is a firm believer in Artificial Intelligence as more than technology that can help us progress in terms of healthcare, ethics and biases, it can also make us more sustainable by optimising our energy consumption, production and logistics, for example, by way of goods that will no longer have to be “hauled several times around the world before they arrive.”
It should be “mandatory for all of us to try just a minimum of Machine Learning”.
Sine Zambach compares it to learning how to use a computer or iPad. Here, you also need a simple understanding in order to make it work. How to power it, how to connect a mouse or install an app. It is important to gain insight in order to be able to challenge and build the systems of the future based on Artificial Intelligence.
XAI, the joker
She points specifically to a helpful tool called Explainable AI, XAI, which Elon Musk also uses in the name of his latest company.
“Explainable AI is important to be able to understand how Artificial Intelligence reaches its predictions. When the police arrest someone, it is important to understand why it is him and not her. Is it merely good policing or is it based on a system? With Explainable AI, you can make the systems explain themselves and teach them how to become more precise. And what is even more important, you can use Explainable AI to find out why an algorithm has made a mistake. Has it been fed too much data or used them inexpediently?” Sine Zambach explains and then concludes:
“Once we start realising what Artificial Intelligence can be used for and help with, it will make great sense to find out what Machine Learning can do – as long as we humans take control and develop our own digital knowledge.”
For more information please contact Communications Adviser Mikael Kolby