Algorithms and AI: which questions do you ask?
Insight starts with asking questions. Asking questions about the source, analysis and outcome helps you to better understand data models, algorithms and artificial intelligence (AI). With more insight, you make better choices. On which theme do you want more insight into AI?
Algorithms and AI are often used in our society. Think of the timetable of public transport or the selection of your favorite music. Click below for more examples.
Spotify uses different datasets from users and is constantly updating its AI models to keep as many customers as possible happy and coming back. Spotify is thus pushing the field of machine learning to optimize its own product.
In 2022, self-driving cars will work as a combination of all kinds of different AIs. A major challenge lies in the ethical dilemmas you encounter: how do we determine whether the algorithm opts for the safety of a pedestrian or the co-driver?
Many were stunned when a chess computer defeated world champion Kasparov in 1996. AI trained as a chess computer can think through all possible chess scenarios super fast, but such an algorithm cannot play checkers.
If you train an algorithm on data that mainly contains male employees, that algorithm will also mainly select male applicants when recruiting. The algorithm makes social prejudices visible, it does not introduce them itself.
The NS timetable is a good example of the added value of AI in complex problems. Algorithms can make an efficient timetable much better than we humans can.
Algorithms and AI are often used around our health and care. Think of fitness apps on your smartwatch or research into better diagnoses. Click below for more examples.
A diagnostic algorithm, developed to detect skin cancer in photos of skin samples, turned out to be better than doctors at this. Great, but the algorithm was trained on photos of mostly fair skin and works poorly at detecting skin cancer in colored skin.
The analysis of patient data such as DNA and medical history, helps to develop new therapies or medicines. The challenge is to standardize data collection, for example at patient intake. This way, quality and reliability are guaranteed and data can be properly compared.
Analyzes show that people who often drink green smoothies are healthier. Are the smoothies the cause of better health (causation) or is it a predictive factor (correlation)? Do people who are involved in a healthy lifestyle, drink green smoothie more often?
The CoronaMelder app warns you if you have been near an infected person, provided that person also uses the app. A tricky situation; the app won’t work until enough people use it, but people won’t use it until they know it works.
When looking at data on the risks of breast cancer treatments, chemotherapy seems very dangerous; these patients live shorter lives than after radiation. Abolish chemotherapy? Rather not. It is precisely this that works in aggressive types of cancer, where patients would otherwise live even shorter lives.
Algorithms and AI are often used in the transition from fossil fuels to sustainable energy. Think of the smart meter in your home or the control of a windmill. Click below for more examples.
Natural gas free
Energy labels are valid for 10 years and updating in between is not mandatory. As a result, data on energy-efficient homes is often behind the times. How does research into energy poverty (households that spend a large part of their income on energy) deal with this?
Smart meters are well established, but how many smart devices do you have? Using algorithms and AI, use can be coordinated, so that, for example, your smart washing machine switches on when your solar panels produce a lot of power, and your car stops charging when it’s cloudy.
Algorithms and mathematical models are used to gain insight into our climate. Think of the formation of clouds or predicting how much the sea level will rise. Click below for more examples.
Below sea level
How can you find out how much it rained 200 years ago? A well-researched and tested example is the measurement of tree rings. Every year a ring is added, and in wet years the ring is thicker because the tree grows more.
Impressive scientific collaboration; the IPCC report is made by scientists all over the world, with a focus on data exchange, critical openness and a lot of complex mathematical models.
They have a huge influence on the weather, but are still one of the biggest challenges to forecast and model. Clouds reflect modelable weather effects and are therefore not included in climate models themselves.
Did you know that it is easier to predict the temperature for a large area? The smaller the area, the more difficult (less accurate) this becomes. For example, a global 1 degree increase can be modeled better than the local temperature in your home town next week.