“AI systems tend to be used in more complex contexts, rely on more sensitive data, and are often less transparent than traditional IT systems,” said Professor Davis.
“It deserves special attention from the board and senior executives.”
The report suggests four areas for urgent action. Developing fit-for-purpose AI strategies. A finely tuned monitoring system that focuses on data quality and reports to the board of directors. Build in-house expertise. and focuses on the social and personal human harm that AI can cause.
The survey found widespread ignorance about the use of AI. About a quarter of senior executives and three-quarters of corporate directors were unaware or unsure of the details of AI governance within their organizations.
“We have seen people embarking on AI implementations without a full understanding of the risks and harms,” says Simon Burns, partner at the law firm Gilbert + Tobin, who helped produce the HTI paper. says Mr.
“What we are seeing is that the exponential speed of AI is causing a gradual change in what companies are experiencing, which will impact technology and organizational capabilities and risk management capabilities. This creates a considerable gap in understanding.
“So we need to close that gap and I think we can do that relatively quickly, but regulation will take longer.”
The study found significant diversity in the way Australian organizations manage and govern AI systems.
“Many of the largest and most resourced organizations are now using a combination of ad-hoc practices and existing structures to monitor AI systems,” the report said.
“Many company leaders reported that their design, development and procurement decisions are heavily influenced by a single ‘leader’ who is perceived as an AI organizational expert.”
The report notes that at least two large organizations reported that their AI-related risks were managed by simple spreadsheets separate from other IT governance systems. Many interviewees said their AI systems were approved by legal and compliance teams with little or no working knowledge of AI systems.
Various current governance models expose companies and boards to significant personal and societal harm that can occur if AI systems are not properly overseen.
“The standout model for us has been described as the Guru model, where there are pockets of expertise within the organization and AI-related governance and risk issues are referred to them on an ad-hoc basis without a formal structure around them. ,” Burns said. He said.
“This can leave a blind spot for the board, which does not always have all the information they need to put the framework in place and make the necessary decisions. .”
However, the report points to research that suggests that ethical frameworks and principles alone are insufficient and, in some cases, counterproductive.
“We need to go beyond responsible and ethical AI principles to practice change,” Solomon said.
“Empirical studies show that AI principles and codes of ethics have little measurable impact on the behavior of engineers developing AI systems,” the report states.
“The expression, interpretation, and execution of ethical principles renders them irrelevant (“But what does ‘fairness’ really mean?”) and isolates them (“They give us I don’t think it applies”), preying on the toothless (“they are”). It’s just a guideline’),” the report said.
“Furthermore, other research shows that poorly designed normative frameworks that rely on moral encouragement can be counterproductive and induce the very behaviors organizations sought to avoid.”
Australia currently has few laws explicitly directed at AI systems, the report notes. This has led some corporate leaders to mistakenly assume that AI systems are generally unregulated or that the use of AI is largely an ethical issue.
“Reform is definitely needed, but AI systems are not operating in the ‘highly regulated West,'” Solomon said.
“AI systems are subject to privacy, consumer protection, anti-discrimination, negligence, cybersecurity and labor, health and safety obligations, as well as industry-specific laws,” he said, noting that regulators have already approved AI applications. It pointed out that it is expanding the examination of
