The ethics of artificial intelligence

Perhaps there is no better example to illustrate the importance of ethics in AI than the Olympics. This recurring event brings together people from all over the world, with very diverse skills, genders, ages, and races.

Many artificial intelligence systems are used to study data and make decisions. However, it is crucial to consider people’s circumstances before making these decisions.

Imagine we study the long jump athletes from the Olympics, build an artificial intelligence model based on their data, and then use this model to ‘decide’ the ideal long jump distances for schools worldwide. This doesn’t seem very reasonable. After all, is it right to use the best athletes in the world as a benchmark for young people globally? And, more importantly, how would it affect these young people to have unattainable goals? The answer is obvious, and it leads to another question: if athletes knew their data would be used for this purpose, would they have given their best in their jump?

These issues and questions form the foundation of ethics in AI. An ethical application of artificial intelligence should ideally meet several requirements:

• The collection of data must be approved by the data owners for the stated purpose. If an application user has given permission to use their data for a commercial purpose, it cannot be assumed they have also consented to using that data to build artificial intelligence models. We can see how ignoring this point has caused much controversy and debate regarding ‘copyright’ and authors’ rights with the new generative AI models.
• Even when permission is granted to use the data, it must be stored anonymously and securely. An AI model does not need to store sensitive data in insecure locations.
• An AI model should never be extrapolated to solve problems in contexts different from where the original data came from. If the model is built on data from people aged 20 to 30 in Spain, it cannot be extrapolated to speak for the entire country or make decisions at a European level.
• AI models consume resources to function: the more powerful the model, the more energy it consumes. It is crucial to build models that increase productivity and help society reduce environmental harm, not increase it.
• Finally, an ethical AI application must be able to explain its decision-making logic to users so that they can understand and correct the suggested decisions if necessary.

At WonderBits, we always strive to follow all these guidelines. All data is handled anonymously and securely, using encryption in our applications and databases when necessary. We conduct thorough studies to ensure the data is accurate and sufficient to build a fair model that doesn’t favor any specific group of users. To guarantee this, we use technologies that allow us to study the models and their responses, ensuring they are safe and provide the correct results we seek. Finally, we always assess the possibility of building models at the necessary scale, using cloud services or smaller models stored on our servers to reduce environmental impact whenever possible.