Last March, a set of measures called theAI Act was voted on and approved by the European Council for the first time in the history of digital technology. This action raises a significant number of legal and ethical considerations concerning the use and regulation of AI on European territory. Since the rise of ChatGPT as an improved search engine (LLM), AI offers numerous opportunities for businesses and individuals, but has shown that it is not without dangers. This article will discuss various challenges that AI faces, and what are the solutions to remedy or limit them.
Defining AI, a task that is more difficult than expected
To effectively deal with the challenges of AI, it is important to understand this emerging technology. On what technological principles is it based? To what degree is it autonomous? All these questions need to be addressed before regulating this technology, which remains quite difficult to frame, as it remains in constant evolution. AI includes a broad portfolio of products and services, the best known being the broad language model (LLM) like ChatGPT with GPT-4.
Defining AI in a legal way therefore seems a more complex challenge than expected, since its use will be regulated in terms of its own definition.
Today, in the AI Act, artificial intelligence is perceived as one of the greatest opportunities of recent years, so the European Council has chosen to deal only with general-purpose AI models, called GPAI. These models represent the foundations that will serve as the basis for other models such as generative AIs such as ChatGPT or Mistral AI. Ultimately, GPAIs will be categorized according to their level of risk.
Legal and safety responsibilities
GPAIs, by definition, are autonomous and unpredictable at a certain level of risk. When an AI makes decisions independently, it is important to determine who is responsible in the event of an accident. Who is responsible if you lose control of an autonomous vehicle? In case of automated financial investment?
The AI Act therefore tries to answer all these questions through an in-depth study of systems and their risks. To prevent possible accidents, the AI Act establishes safety standards for users. Depending on the evolution of regulations in force, autonomous AI models will adapt accordingly.
Safety at the heart of concerns
This increased autonomy increases the vulnerability of systems against algorithmic biases and cybernetic attacks. A breach in the system combined with malicious action can have serious consequences. This is why the AI Act places a lot of importance on security standards and on the transparency of businesses in relation to the architecture of their products or services.
Ethical responsibilities
Data protection and personal information
The protection of data and personal information is a major concern in the use of AI, both for users and for businesses. Information transmitted via AI systems is often confidential and sensitive, requiring strict anonymization and confidentiality measures. The European AI regulation imposes transparency and data security obligations on AI companies to ensure that personal information is properly protected against abuse and leaks.
Some AI technologies based on facial recognition pose problems with respect for the privacy of citizens in general. In some cases, improved surveillance for security reasons represents an attack on civil rights defended by the European Union.
Is an objective and ethical AI possible?
The place of AI in daily actions has changed a lot in recent years. Many ethical considerations with regard to AI are rightly addressed by the European regulation. First, AI is likely to create algorithmic biases based on preconceived ideas and biases based on the data it is trained on.
In several use cases such as loan applications or recruitment, the use of AI risks discriminating against certain people on criteria that are not very egalitarian such as gender, age or the socio-demographic situation of the person.
Initiatives to improve the ethical and legal considerations of AI
Towards international and commercial collaboration
AI is a technology that is still emerging and is likely to change profoundly in the years to come. That is why many countries have decided to regulate AI in terms of responsible innovation. The most elaborate regulation, combining security and a pro-innovation vision, is the European Union's AI Act. In addition, other institutions such as China, the United Kingdom or the UN publish reports and draft regulations to deal with all the legal and ethical considerations raised by AI.
At the dawn of a disruptive change in processes in society, collaboration seems to be a solution being explored by the international community. Initiated by Canada and France in 2020, the”Global Partnership on AI” (GPAI) brought together 29 countries today to share its knowledge on AI and create an international collaboration to innovate around responsible AI that respects civil rights.
AutoLex, a company committed to contract analysis
The collaboration of companies with institutions is essential in order to use AI in a responsible way. By being transparent about data management, areas for improvement and privacy policies, companies have a major role in international collaboration to make AI a responsible tool that creates value for all.
Thanks to our partnership with Microsoft, we chose to host all our data in France and Europe with Azure. AutoLex's AI only uses open source and public information, such as CNIL documents, French and European regulations to train its LLMs.
At AutoLex, the security and protection of your data are essential, and allow you to do a contractual analysis of your documents safely.
If you want to try AutoLex, request a 7-day demo and trial right here!
Question/Answers
- What is responsible AI?
Responsible AI is defined by the use of AI technology to generate a positive impact on its users. Responsible AI places people at the heart of its process, evaluating both the pros and cons of using it. Therefore, it encourages transparent and legally compliant use, while respecting the ethics of users.
- What is a high-risk AI, according to the AI Act?
This new legislation classifies AIs into several categories. The most dangerous are considered”at high risk“. These AIs are determined to be dangerous for European freedoms and are subject to strict requirements and obligations before being introduced into the European Union market.
- Who is affected by the AI Act?
This legislation concerns virtually all persons in contact with the European market. Many players are involved in the development, marketing, and use of these technologies, from providers of AI systems to end users, to regulatory authorities. It is therefore essential for everyone to learn about the laws in force.
