Understanding the ethical and legal issues of AI

AI is not new. Its beginnings trace back to Alan Turing's work in the 1950s.

Today, AI technologies are numerous and have significantly advanced in recent years.

In general terms, AI refers to computer technologies capable of replicating behaviors and intellectual abilities typically associated with human intelligence, such as reasoning, learning, creativity, problem-solving, speech and facial recognition, decision-making, and natural language understanding. To achieve such results, AI applies complex algorithms to simulate or replicate human reasoning and learning abilities. This ranges from simple coded rules to sophisticated systems capable of learning from experience (machine learning).

Further information – AI, AI solutions, AI systems: what's the distinction?

In literature and practice, terms like AI solutions or AI systems are frequently encountered. Both terms refer to the application and implementation of AI technologies to achieve tangible results and solve specific problems.

In other words, while AI is the concept and foundational technology, "AI solutions (or systems)" refer to integrating AI into products, services, or systems to solve specific problems or improve existing processes.

AI solutions thus represent practical applications of artificial intelligence in real-world contexts.

Examples of AI solutions deployed in healthcare include AI-assisted diagnosis, disease prediction systems, emergency service scheduling, and surgical robotics. In education, AI solutions are equally diverse and include predictive analysis for dropout prevention, automated assessments, adaptive and personalized tutoring, and automatic translation tools.

Talking about AI in the singular and in a generic manner does not capture the diversity and technological power of AI. There are several types of AI, each with its own characteristics and applications.

Their distinctions are important for assessing and classifying specific risks they pose, and thus adapting their regulation.

They constitute innovative technologies that offer significant benefits, but the impacts of AI on society and humanity also raise concerns.

Identifying these concerns and distinguishing the characteristics of AI is necessary to develop ethical and responsible AI systems that benefit society as a whole (see open letter from the Observatory on the Societal Impacts of AI and Digital Technologies dated April 21, 2023).

Learn more – What are the main types of AI?

At the heart of modern AI are algorithms that enable AI systems to analyze data, learn from it, make predictions, or make decisions without being explicitly programmed for each specific task.

AI also uses language processing algorithms to understand and generate human language, enabling interactions such as automatic translation, text generation, and speech recognition.

Optimization algorithms are used to improve the performance of AI models by adjusting model parameters to achieve the best possible results.

AI algorithms are constantly evolving, driven by several factors including:

• Increasing computational power to handle larger and more complex AI models.
• Explosion of available data for training, enhancing AI algorithms' performance by learning from larger and more diverse datasets.
• Continuous advancement in academic and industrial research, as well as the open-source movement pushing the boundaries of what is possible with AI.
• Demand for more ethical and transparent AI leading to the development of algorithms that can be better explained and audited.
• Trend towards creating systems capable of processing and integrating multiple types of data (text, image, sound) for deeper understanding and richer applications.
• And more.

These advancements illustrate ongoing efforts to make AI more powerful, efficient, accessible, and useful across a variety of applications.

Learn more – AI and algorithms

AI is used by private and public organizations across various domains and for different purposes. Here are some examples:

Healthcare: Assisting doctors in diagnosing diseases such as cancer, heart disease, and diabetes by analyzing medical images, scans, and patient data; personalizing treatments; assisting surgeons during operations; accelerating drug discovery by analyzing large databases of chemical compounds and predicting their effectiveness.

Transportation: Measuring traffic flow based on days and times, identifying traffic issues, adapting traffic signals, enforcing speed limits, geolocating individuals or delivery trucks, providing directions, and autonomous vehicle operation.

Finance: Analyzing borrowing capacity, determining interest rates, and evaluating loan approvals.

Consumer goods: Sending targeted advertisements, conducting purchase analyses and predictions, providing personalized goods or services, recommending products or content based on user browsing history and online behavior, sentiment analysis to detect customer satisfaction with a product or service.

Security: Detecting fraud or identity theft, ensuring national security, preventing crimes, assisting in emergency situations, combating cyberattacks or cybercrimes.

Employment: Optimizing industrial processes, selecting candidates for jobs, measuring productivity, automating tasks, evaluating performance, industrial espionage, restraining or stimulating competition, etc.

Yes. The risk varies depending on the degree of functional autonomy of the AI technology being used. The more AI has the ability to learn autonomously and perform tasks without human intervention, the higher the risk. Risks may include imbalance or infringement of rights in social, ethical, economic, or political domains.

The example of the "black box" or lack of transparency of AI

Many AI systems, especially those based on deep learning, can be considered "black boxes".

This refers to an AI system that performs complex tasks, but whose internal workings are difficult to understand. It then becomes challenging to justify its outcomes. For example, a deep learning system in computer vision can recognize objects in images, but it's difficult to explain how it "thought" to make that decision. This level of opacity is particularly problematic in fields such as healthcare, banking credit decisions, or hiring processes, where understanding why an AI made a particular decision is crucial.

AI systems labeled as "black boxes" are often very powerful because they can effectively solve complex problems. However, they raise issues of accountability, ethics, and transparency because their internal functioning is hard to comprehend. Therefore, it's important to develop methods to make these systems more transparent, especially when they make decisions that impact people's lives. Explainable AI is an important area of research aimed at making these systems more understandable without compromising their power.

Example of the environmental footprint of AI

The type of AI also influences the cybersecurity risk. For instance, the more data-intensive an AI is, the greater the risks of privacy breaches in the event of a cyber-attack. Therefore, understanding the security risks associated with AI systems is crucial for developing appropriate defense mechanisms.

Certain AIs are energy-intensive and can have a larger negative impact on the environment. For example, deep learning poses a greater risk of increasing electricity demand and CO2 emissions due to its complexity. However, efforts are underway to make AI more eco-friendly, including deep learning. This involves using more energy-efficient hardware, using renewable energy to power data centers, and researching technologies that require less computational power.

Paying attention to the development and use of AI systems is important for navigating our increasingly technology-driven world in an informed and conscious manner.

The applications of AI are only limited by its technology. As technologies evolve, the applications of AI will expand further.

AI can integrate into all human activities and even replace humans in certain tasks. It promises significant economic and social advances through its efficiency, performance, reliability, and cost reduction for organizations that use it. However, its advancements raise various ethical and social questions, particularly regarding the impact of automation on employment, privacy, decision-making, and accountability in case of errors.

AI's ability to process and analyze vast amounts of data faster than humans can lead to profound changes in sectors such as healthcare, finance, and security. Moreover, the rise of AI sparks discussions on the need for training and professional retraining to prepare the workforce for a future where the coexistence and collaboration between humans and machines are becoming systematic.

Indeed, recent years have highlighted an intensification of the negative effects of AI, such as transforming the job market, increasing risks of discrimination, misinformation, cybercrime, and violations of fundamental rights.

In other words, integrating AI into our society requires thorough reflection on governance rules. The need for regulation seems undeniable.

Learn more - Ethics or law to regulate AI?

In recent years, AI ethics has seen significant development, yet it has struggled to effectively mitigate the negative effects of AI. This can be explained in several ways, starting with the fact that ethical standards are not mandatory for AI actors. Moreover, these standards are numerous and vary widely in content depending on the organization that establishes them.

In response to this, alongside ethical standards, some advocate for the establishment of a legal framework to "better" regulate AI. This means providing AI with a binding framework in which the conditions under which AI can be developed and used are more rigorously defined and applicable to all AI actors subject to the authority of the law in question.

Legally regulating AI helps establish a socially acceptable balance between the benefits of AI exploitation, its development, and the protection of citizens' rights and interests.

Ultimately, compliance with laws and regulations promotes trustworthy AI: those whose development and use are socially and economically beneficial.

Regulating AI through laws and regulations offers the following advantages:

• Imposing rules that govern the development and use of AI while considering the risks they pose.
• Prohibiting socially undesirable uses of AI.
• Defining common standards on practices and uses considered acceptable in society.
• Translating ethical principles such as the common good, social and environmental responsibility, transparency, and inclusion into legal obligations.
• Compelling developers, distributors, and users of AI to comply with current regulations.
• Enforcing penalties against non-compliant behaviors and practices.
• Building citizen trust in AI systems by demanding more transparency, accountability, and compliance.
• Establishing monitoring mechanisms and avenues for redress so individuals can assert their rights or report AI uses that harm them.
• Harmonizing regulations to promote international trade and maintain the competitiveness of Quebec businesses against foreign companies.

Several states around the world are debating the merits of adopting specific laws for AI. For some, the adoption of AI laws or regulations poses a threat to innovation, while for others, using the law can secure the AI ecosystem by safeguarding innovation and investments and promoting the development of responsible innovation.

Once this debate is settled, the legal regulation of AI presents numerous challenges for legislators who must:

• Consider the variety of AI systems and the technologies they use to assess the risks they pose;
• Contemplate the multiple applications of AI in our societies to distinguish their uses according to different situations; Continuously adapt to the extremely rapid advancements in technology;
• Integrate the international dimension of this technology.

Indeed, AI transcends national, regional, and cultural boundaries, promoting global integration and interconnection. Its use is not confined within geographic borders, which can make determining applicable jurisdiction and establishing coherent global regulations challenging.

In addition to these challenges, the development and use of AI raise unprecedented issues such as:

• Transformation of the job market and disappearance of certain professions;
• Violations of individuals' rights (privacy infringement, discrimination, infringement of freedom of expression, mass surveillance, language rights, etc.);
• Risk of misinformation and manipulation of public opinion; Security and cyber-attacks; Impact on creativity and culture;
• Preservation of diversity of cultural expressions; Technological dependency; Unequal accessibility to technology, etc.

In Quebec, there is currently no specific law that regulates the development and use of AI. This does not mean that AI solutions are exempt from the law. General laws, such as the Civil Code of Quebec, still apply even in the context of AI. However, applying these laws to cases involving AI presents practical interpretation challenges that have not yet been addressed in court.

Currently, there is debate over whether to adopt a law that would regulate the development of artificial intelligence and clarify the interpretation of existing law.

This question is being actively discussed in light of the ethical limitations of AI.

Quebec is in the stage of reflection. Public consultations on the regulation of artificial intelligence in Quebec represent a first step towards establishing an action plan to regulate AI solutions.

AI and digital data are inherently linked because data forms the foundation upon which many AI systems are built and improved. To function effectively, certain AI solutions require access to large volumes of digital data to learn, generalize, and produce more accurate and useful results. The more data available, the more opportunities AI systems have to analyze and understand different patterns, which can lead to better outcomes.

For example, in image recognition, an AI model that has access to a wide variety of images can develop better capabilities to identify different objects and contexts. Similarly, AI models designed for natural language processing, such as automatic text generation, require vast amounts of textual data to learn the structure, grammar, and linguistic context.

When the collection and use of data involves personal information, it raises significant ethical and legal challenges concerning privacy, data security, and the potential for discriminatory biases in AI systems. These issues are exacerbated by the massive scale of collecting and using personal information.

The massive collection and use of personal information are subject to laws on personal data protection that rigorously regulate how such data can be collected (purposes, consent, volume, retention period) and individuals' rights (e.g., right to access, right to rectify).

It is therefore crucial to ensure that massive data collection adheres to ethical and legal standards, and that resulting AI systems are designed and used without causing social harm.

The use of personal information to fuel artificial intelligence is widespread. Preferences, habits, and behaviors of individuals are particularly useful for developing AI systems capable of targeted advertising and providing personalized goods and services.

However, personal information is part of individuals' privacy and is protected by specific laws. The collection and use of personal information by AIs must respect the rules set out in laws on personal data protection, which rigorously govern how personal data can be collected (purposes, consent, volume, retention period) without infringing upon individuals' rights, including:

• Privacy
• Data security
• How data is used and stored
• Access to data and rectification
• Freedom of expression

These aspects will be further clarified in the upcoming second part of the Q&A box.