Artificial Intelligence: Compliance and Regulatory Challenges
1 PRIVACY & CONFIDENTIALITY 3 GOVERNANCE 4 BIAS & DISCRIMINATION
• The expansion of AI has implications on personal privacy, AI algorithms have common features that • The AI industry is fast-moving and largely unregulated.
which is a fundamental human right. will complicate overall governance : • Developers will need to avoid “Bias deep inside the code”.
• Some of the information learned might be private or • There are no one-size-fits-all solutions. • Algorithms must not discriminate on grounds of age,
confidential even when the data mined to extract this insight • Governance is going to have race or gender.
might not be considered as PII or confidential. cross-border nature, requiring global • Ethical standards will need to be defined and enforced.
• This might be particularly serious under regulations like dialogue and collaboration. This is not going to be an easy process for regulators.
GDPR. • The AI industry is evolving quickly. • Bias and discrimination are going to be critical for
Governance and Regulatory initiatives are compliance in Financial Services, Government Services and
slow by nature. generally speaking in highly regulated industries.
2 ACCOUNTABILITY & LIABILITY 5 ALGORITHMIC TRANSPARENCY
This is both a fascinating and extremely complex issue. • If we do not know what an Algorithm is doing, we cannot
Several questions arise: hold it accountable.
• Who is going to be liable when autonomous • The transparency topic might be a serious issue in
decision-making, AI enabled, processes cause damage? particular in regulated industries.
• Will sophisticated or autonomous algorithms be subject • This is especially problematic when implicating people,
to traditional product liability legal frameworks? and it is not easy to address this problem as just because
• Is there a need to create a new treatment for “Thinking someone has access to the ML code, it does not always
Algorithms”? equal being able to explain how the software works.
• Will it be needed to create a classification and specific • Transparency affects several dimensions, including:
treatment based on their level of autonomy? • Data and Algorithms used
• Goals and Outcomes
• Influence and potential manipulation
SO cyberstartupobservatory.com Cyber Startup Observatory - Insight