AI Serving Administration: A Revolution in Progress
Artificial intelligence (AI) is increasingly becoming the ultimate accelerator of performance, efficiency, transparency, and accessibility in public administration.
AI and E-Government: Concrete Applications
Administrative decision support, automatic detection of tax fraud or traffic violations, combating money laundering, reception robots in government offices, interactive voice bots in local languages, automatic generation of administrative reports or summaries, aerial image analysis to map displaced population camps... the applications of Machine Learning (ML1)[1], Deep Learning (DL2)[2], Natural Language Processing (NLP)[3], Computer Vision[4], Expert Systems[5], or Generative AI[6] in e-government are endless.
Case Studies: AI in Action
-
Singapore : Streamlining Traffic with AI
In Singapore, peak-hour delays have decreased by 20%, and average speed has increased by 15% thanks to the artificial intelligence-based traffic management system that uses real-time data to monitor traffic flow, predict congestion patterns, and adjust traffic lights accordingly.
- Estonia: AI at the Heart of Healthcare and Education
In e-Health, the Estonian government has developed an AI-based health information system to manage patient data and improve healthcare delivery. This system integrates data from various healthcare providers, enabling real-time access to patient records and facilitating decision-making by healthcare professionals. AI algorithms analyze patient data to identify trends, predict potential health problems, and recommend preventive measures, ultimately improving public health outcomes.
Also, as part of a unique global initiative, Estonia is providing all its secondary school students and teachers with access to ChatGPT Edu.
- France: Ambitious Projects to Improve Citizens' Lives
In France, the Government has highlighted thirty projects funded by the France 2030 investment plan, using AI to improve healthcare, agriculture, education, and more, to make citizens' lives better.
Ethical Challenges of AI in the Public Sector
However, these rapid changes raise profound ethical concerns. These stem from the potential of AI systems to incorporate biases and discrimination, contribute to climate degradation or human replacement, manipulate opinion, or threaten human rights, among many other things.
International Principles and Frameworks for Responsible AI
Adopted in May 2019 by the 42 member countries and partners of the OECD, the Principles for Responsible and Trustworthy AI constitute the first intergovernmental framework for guiding the development of artificial intelligence in an ethical and beneficial way for humanity; namely, benefit for individuals and the planet, respect for human rights, democratic values and fairness, transparency and explainability, robustness, security and reliability, and finally, responsibility.
The OECD and the G20 have aligned with these principles, giving them near-global reach. In 2020, the OECD created the Global AI Policy Observatory (OECD.AI), a platform that tracks the concrete application of these principles in each country.
In 2021, the UNESCO General Conference adopted the first "Recommendation on the Ethics of Artificial Intelligence," establishing a comprehensive framework to guide AI towards human, social, and environmental well-being.
European Regulation: AI Act
On August 1, 2024, the European AI regulation, known as the AI Act, came into force. It regulates the development, marketing, and use of AI systems, with obligations for suppliers and prohibitions for certain uses, including social scoring and remote biometric identification in public places, except in exceptional cases.
Recommendations for Responsible AI
In addition to its principles, the OECD proposes guidelines for governments to promote responsible AI:
- Invest in research and development in reliable AI.
- Foster an inclusive innovation ecosystem.
- Adapt legal frameworks to encourage transparency, security, and fairness.
- Develop human capacities.
- Encourage international cooperation.
According to UNESCO's recommendations, the goal is to put AI systems at the service of humanity, preventing harm and promoting social, environmental, and ethical well-being.
[1] Machine learning is a field of study within artificial intelligence that aims to enable machines to "learn" from data through the use of mathematical models. Specifically, it involves the process of extracting relevant information from a training dataset. The goal of this phase is to determine the parameters of a model that will achieve optimal performance, particularly in executing the assigned task. Once training is complete, the model can then be deployed in production. (Source: CNIL)
[2] Deep learning is a subset of machine learning that utilizes neural networks with multiple hidden layers. These algorithms, which contain a very large number of parameters, require substantial amounts of data for training. An artificial neuron operates in a manner inspired by biological neurons: a node within a network of multiple neurons typically receives several input values and generates an output value. (Source: CNIL)
[3] Natural language processing is a multidisciplinary field involving linguistics, computer science, and artificial intelligence. Its aim is to develop tools capable of interpreting and synthesizing text for various applications. (Source: CNIL)
[4] Computer vision is a branch of artificial intelligence whose primary goal is to enable machines to analyze and process one or more images or videos captured by an acquisition system. (Source: CNIL)
[5] This refers to a set of software systems whose problem-solving capabilities in a specific domain are comparable to those of a human expert specialized in that field. Their overall architecture revolves around a knowledge base— a memory containing representations of factual knowledge and reasoning rules within a particular domain. At the core of the system is an inference engine, which manages, according to general strategies (hypothesis generation, analogy) or specific ones, the sequence of problem-solving steps, each corresponding to the application of a rule. (Source: Larousse)
[6] A generative artificial intelligence system is capable of creating text, images, or other content (music, video, voice, etc.) based on instructions from a human user. These systems can produce new content from training data. Their performance today is comparable to certain outputs created by humans, owing to the large volume of data used during training. (Source: CNIL)
