By Beatrice Kayaga |

According to the United Nations Educational, Scientific and Cultural Organisation (UNESCO) recommendations on the Ethics of Artificial Intelligence, 2022, Artificial Intelligence (AI) is  the ability of machines to perform tasks in a manner that mimics intelligent human behaviour, often involving elements such as reasoning, learning, perception, prediction, planning, or control. At its core, AI encompasses several key fields, such as Natural Language Processing, which enables machines to understand and generate human language, making applications like chatbots, translation tools, and virtual assistants possible.  Machine Learning on the other hand, is a component of Artificial Intelligence that enables it to learn from data and improve over time without being explicitly programmed.

Deep Learning is a subset of Machine Learning which uses artificial neural networks to model complex patterns in data. Its functions can be identified in  speech recognition, autonomous vehicles and  image classification appliances or devices. Lastly, the Large Language Models like ChatGPT, Llama and Claude among others are designed  to process and generate human-like language . These broadly trained systems can be fine-tuned for specific tasks, and user needs.

Developing these models follows a Collect, Train, Evaluate, and Tune structured pipeline. Initially, vast amounts of general data are collected to train a foundation model, often using significant computational resources. Once trained, the model’s performance is evaluated, and it is then fine-tuned with domain-specific data to enhance its accuracy and relevance in specific applications.

However, this process is not without challenges. Bias and transparency, as AI systems can inadvertently perpetuate or amplify societal biases embedded in the data with which they are trained is a major concern.  For instance, Amazon’s job recruiting was found to favour men over women mainly because it was trained on predominantly male-dominated resume data.  Furthermore, the lack of transparency in so-called “black box” algorithms has sparked concerns around explainability and accountability, making it difficult for users and regulators to understand how decisions are made.

Privacy is another critical dimension of AI. Since AI systems often rely on massive datasets including data, the risk of privacy invasion looms large. For example, while facial recognition offers advantages such as improved security, social media photo tagging, contact tracing during public health crises like COVID-19, and personalized marketing, its deployment by law enforcement and surveillance agencies has sparked intense ethical and legal debates concerning data rights and risks like consent, data retention, and potential misuse. AI can also facilitate the spread of misinformation and disinformation by creating highly realistic yet false content such as deepfakes that make people appear to say or do things they never did and distribute it instantly through digital platforms. When such content spreads rapidly, it can fuel the explosion of fake news that potentially undermines democratic processes, manipulates public opinions and erodes trust in the media. It also violates fundamental rights and freedoms such as freedom of expression and information because AI moderation systems designed to detect and remove harmful content sometimes censor legitimate speech due to algorithmic errors or biases. These tech practices tend to have a chilling effect on digital rights and at times force people into self-censorship due to fear of repercussions.

As artificial intelligence is becoming deeply embedded in society, the question of   trustworthy, secure, privacy-preserving AI that is aligned with human rights has become critical. These concerns have prompted the development of legal and regulatory frameworks to balance the drive for AI innovation with societal safeguards. For example, in the European Union, the EU AI Act takes a risk-based approach towards regulating AI where the framework categorizes AI systems by their potential harm. The United States, by contrast, uses export controls and sector-specific regulations in order to regulate sensitive AI technologies as a way of protecting national security. In regions like Africa where there are no dedicated AI frameworks, governments are adopting existing legislation such as data protection laws, AI strategies, procurement regulations and  access to information statutes to regulate AI. However, the lack of specific AI laws  leaves significant challenges related to algorithmic bias, privacy, accountability mechanisms, intellectual property rights protection and the wide procurement and deployment of AI in largely unregulated spaces.

Trustworthy AI can also be achieved by integrating a human rights-based approach (HRBA) in its systems. The HRBA would ensure the placement of protection of fundamental rights and human dignity at the center of AI development, deployment, and governance. This approach ensures that AI systems are designed and operated in ways that respect fundamental rights and freedoms such as the right to privacy, freedom of expression, non-discrimination, and equality among others. Therefore, by embedding human rights principles into AI design, developers can mitigate these and create systems that serve all individuals fairly and equitably

Lastly, mandatory impact assessments should be conducted prior to the deployment of AI systems. The measures should evaluate potential social, economic, environmental, and ethical implications, and ensure that adequate safeguards are in place. Together, these measures will create a comprehensive governance structure that not only mitigates risks but also promotes trust, transparency, and accountability in the use of AI.

As AI systems continue to influence critical aspects of society, it is important to  ensure that they are based on transparency, fairness, accountability, and ethics. This would in turn build public confidence and safeguards against unintended harms. Achieving trustworthy AI requires a multi-stakeholder approach requiring collaboration among researchers, developers, policymakers, and civil society to align technological progress with societal needs and rights that underpin just and equitable societies.