CTO of Sage, Aaron Harris, explores the urgent need for coordinated ethical AI standards as 90 percent of SMBs prepare to adopt AI tools by 2030, and warns against a “Wild West” approach to implementation.
A record 28 percent of venture capital investment flowed to AI startups in Q2 2024 alone. But beneath these impressive statistics lies a challenge: how to ensure AI development serves humanity ethically rather than amplifying existing biases and inequalities.
There is an urgent need for global coordination on ethical AI standards, particularly for small and medium-sized businesses that form the backbone of South Africa’s economy. Especially considering that, according to Sage’s recent Vision to Industry report, more than 90 percent of SMBs will employ AI tools for continuous monitoring and anomaly detection by 2030.
Aaron has been vocal about the need for international collaboration on AI ethics. “As we embrace the immense potential of AI, we must proceed with caution,” he says.
“The world of AI can be like the Wild West as rapid expansion and access outpace regulation. A steadfast commitment to ethical considerations is essential moving forward," explains Aaron, highlighting the risks particular to financial services, where decisions directly impact people’s livelihoods and economic opportunities.
He draws a stark comparison between AI development and other technology innovations. “Unlike other innovative tech, the ‘move fast and break things’ philosophy doesn’t apply to AI,” he explains. “It should be a prerequisite to ensure those building AI solutions are qualified before they begin. There is an ethical risk attached to building AI. That risk can be mitigated if you have the right framework in place.”
This approach stems from AI’s unique characteristics and potential for harm. Aaron explains: “In Jurassic Park, Jeff Goldblum warned us, ‘Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should’. He was talking about cloning dinosaurs, but the same applies to ethical AI, where the temptation can be to implement the technology across the board without considering the ethical implications.”
The financial services sector faces particular challenges because AI decisions can have immediate and lasting impacts on businesses and individuals.
Aaron shares an example of how seemingly helpful AI tools could cause harm: “At Sage, we could build an AI tool that allows SMBs to rate customers on how quickly they pay. But this could easily disenfranchise struggling businesses and exacerbate the problem instead of finding a solution.”
Practical frameworks for ethical AI
Aaron emphasises the importance of early integration and comprehensive governance.
“A strong ethical AI framework should be grounded in clear governance, proactive risk management and rigorous development standards,” he explains. “IT leaders should ensure that all AI products and features strictly adhere to the organisation’s AI and data ethics principles before any development begins.”
Aaron advocates for adopting initiatives like the NIST AI Risk Management Framework to assess and mitigate risks throughout the AI lifecycle. This includes comprehensive internal assessments against ethics, legal, privacy, and cybersecurity standards before any product reaches the market.
The approach to AI development also requires a fundamental shift in thinking, “adding guardrails to ensure outputs are safe, relevant and ethical”.
Technical safeguards
“To ensure data quality and prevent harmful AI outputs, several tools and practices should be prioritised,” Aaron explains. “Robust data validation and cleaning processes are essential to identify and correct errors in training data. Bias detection tools must be used to uncover and address bias in both datasets and model behaviour.”
The technical toolkit also includes adversarial training techniques to build model resilience against malicious inputs, and Explainable AI (XAI) to enhance transparency. “These tools and innovations form a multi-layered approach to building trustworthy AI, ensuring outputs remain accurate, fair and aligned with real-world expectations.”
Aaron frames diversity in AI development teams as both a moral imperative and a technical necessity. “The impact of diversity and representation on mitigating bias is significant. Diverse teams are more likely to identify and address potential biases in AI models, leading to fairer, more equitable and more robust outcomes,” he explains.
Yet only 22 percent of AI professionals are women, and 25 percent of AI employees identify as racial or ethnic minorities. Aaron advocates for targeted recruitment strategies, inclusive hiring practices and mentorship programmes to address these gaps.
The collaboration imperative
The AI developer and data scientist community is a traditionally collaborative one, especially compared to other types of technology development. “This community can come together to prioritise and protect ethics, fostering a culture of transparency and accountability,” Aaron says.
Internally, Aaron recommends creating cross-functional teams that include AI developers, ethicists, legal experts and business stakeholders. “Encouraging open communication where team members feel comfortable raising ethical concerns without fear of reprisal is crucial,” he notes.
The external collaboration extends to open-source communities and industry forums. “Open-source models offer a unique opportunity to democratise AI development by allowing developers from diverse backgrounds and skill levels to contribute to and learn from each other…ensuring a broad range of perspectives are considered in AI development.”
“The most urgent policy or regulatory gap is the lack of alignment on coordinated foundational principles for ethical AI across different countries and regions,” warns Aaron.
While initiatives exist – the EU AI Act, G7 Guiding Principles, and state-by-state approaches in the US – Aaron argues for more unified international coordination. “AI is a borderless technology. The priority should be for governments, policymakers and countries to align on a coordinated set of foundational principles that create the high watermark for ethical AI.”
He points to the Bletchley Declaration as an example of promising international collaboration, though he acknowledges the challenge of creating regulations that can adapt to AI's rapid pace of change while maintaining ethical standards and enabling innovation.
Finance leaders appear ready for the responsibility, with 72 percent of respondents in Sage’s survey planning to establish AI-specific policies and 71 percent committed to regular ethics training for AI users. “But this reality can only manifest itself if steps are taken to create a unified approach to ethical AI,” Aaron observes.
“Ultimately, the CTO becomes a steward of ethical technology – ensuring that innovation and integrity move forward together, and that the organisation’s AI reflects not only what is possible, but what is right.”
