With great power comes great responsibility, a statement that could easily be applied to the rapid evolution of AI in the last couple of years. Few would dispute that AI’s potential for either positive change or harm depends on its responsible use. We need to move quickly to ensure AI systems are developed, deployed, and operated safely and ethically while maximizing their positive impact.
At the extreme, some fear that AI could ‘go rogue’ and act against the interests of humanity. OpenAI even created a Superalignment team dedicated to ensuring AI remains aligned with human values - preparing for a future in which it surpasses human intelligence.
But even if the possibility of AI exceeding human intellectual capabilities seems some way off into the future, the tech is evolving so fast that we need to prioritize AI ethics and regulation to protect people’s fundamental rights and establish guidelines to steer safe innovation.
Effective regulation can deliver business value by increasing public confidence in AI and helping companies avoid AI product failures and reputational damage.
The EU AI Act
The EU AI Act is the world’s first comprehensive legal framework for regulating AI - with the power to impose significant penalties for non-compliance (fines of up to €35 million or 7% of global annual turnover).
The act categorizes AI according to 4 risk levels, starting with those systems it deems an ‘Unacceptable Risk’ – which it bans outright. For example, this includes toys that might use AI chatbots to encourage dangerous behaviour or AI-powered social scoring by governments.
At the next level are AI systems that are designated ‘High Risk ’, which the EU suggests should be subject to strict rules before they’re allowed on the market. These applications would have to meet obligations in areas such as transparency, accountability and human oversight. AI that can put the lives and health of citizens at risk, such as systems that support critical infrastructure (like transport) as well as medical AI - like robot-assisted surgery - come under this category.
For the ‘Limited Risk’ AI level, the regulations mainly revolve around providing greater transparency. Organizations must clearly indicate when people are interacting with AI chatbots and also label AI-generated content (including audio and video), which is used to advise people on “matters of public interest”.
Lastly, there is the ‘Minimal-Risk’ AI category, which includes the majority of current AI use cases (from spam filters to AI-enabled video games). For these the EU AI Act imposes no additional AI-specific regulations beyond existing laws.
Will AI regulations be effective?
We won’t be able to properly judge the effectiveness of regulations like the EU AI Act for years to come. But there’s an ongoing debate about the extent to which regulations will encourage ethical AI or stifle innovation.
Business costs will definitely rise because companies will have to shell out money to meet compliance requirements. Regulatory assessments and approval processes could delay the development and commercial deployment of some AI systems because making detailed disclosures about AI training data is complex and time-consuming.
Some companies may not be able to respond to market needs as quickly because of regulatory constraints. Ground-breaking AI initiatives could be delayed with negative repercussions; imagine patients waiting longer for cutting-edge AI-enabled diagnostic tools due to extended regulatory approval times.
Smaller companies and startups are likely to struggle most with the burden of compliance because they’re more likely to lack the resources to meet the complex technical and legal demands. Estimates show that compliance could add 17% to AI development costs, for example. And for a small firm, the heavy fines for non-compliance could cripple their operations. This raises concerns because significant innovations can sometimes emerge from niche companies and specialist startups rather than the bigger organizations.
There are reports of businesses, particularly Big Tech and providers of general-purpose frontier models, lobbying the EU and individual member states about the regulations, with tech companies spending over €97 million annually on lobbying EU institutions. This gives larger companies with more resources a greater ability to shape regulatory outcomes, while smaller businesses and open-source initiatives have less influence.
Since the EU is the first region to bring in a wide-ranging set of regulations for AI, another area of debate is the possibility of a "two-speed AI" market with customers in the region receiving less advanced AI functionality. For instance, Google held back the release of its own AI tools in the EU for several months due to regulatory concerns and Meta has delayed the introduction of AI updates to its Ray Ban smart glasses in Europe reportedly because of a lack of clarity around regulations.
Balancing regulation with encouraging innovation
To balance regulation and innovation, regulatory frameworks need to have built-in flexibility. A context-specific approach focuses on how AI is used in specific situations rather than creating blanket rules for entire technologies.
For example, an AI image recognition system requires different levels of oversight when used for social media photo filters versus airport security screening. An adaptable framework would allow regulators to weigh the specific risks and benefits of each AI application while ensuring proportionate responses - stricter oversight for high-risk uses and lighter touch regulation for lower-risk applications.
Policymakers should also involve industry experts, academics and the public in regulatory discussions, both during rule development and on a continuous basis, to ensure that regulations remain fit for purpose as technology evolves. For example, the EU Act has established a governance structure at the European and national levels, including an AI Office, a scientific panel of independent experts and an AI Board.
Alongside setting clear, well-thought-out regulations, it is important for regulatory authorities to assist and encourage businesses and researchers to push through with AI innovations. In addition to providing guidelines and supporting compliance, regulators could allow companies to experiment with AI technologies under regulatory supervision in a sandboxed environment. Financial incentives such as grants or tax breaks should also be considered to help promote responsible and ethical AI research and development, which is especially important for encouraging smaller businesses.
The European Commission has introduced an AI innovation package to support AI startups and SMEs through a variety of measures like this. They include access to dedicated supercomputers to widen the use of AI across public and private sectors, financial support for startups and scale-ups via venture capital or equity and initiatives to strengthen the EU's AI talent pool through education, training and reskilling.
As AI evolves, balancing innovation and ethical responsibility is going to be crucial. The EU AI Act is a major step in shaping responsible AI, but ongoing international collaboration between regulators, businesses, researchers and the public will be essential to creating a responsible AI ecosystem that aligns with human values. International cooperation, particularly in areas like regulatory policy alignment, standard-setting, and joint research projects, will be fundamental to maximizing AI's benefits while protecting fundamental rights.
