By: Jihad Dannawi, APAC General Manager, DataStax

Understandably, many are preoccupied with leveling up the customer experience, elevating efficiency, and empowering employees to be their best, most productive versions of themselves. No wonder, then, that many companies are looking to leverage artificial intelligence (AI).

In Southeast Asia, for example, Bain and Co. finds that AI played a significant role in raising the region’s revenues from the digital economy to USD100 million last year. These numbers strongly suggest rising adoption, which is also borne out elsewhere in Asia. For instance, IDC notes high hopes for GenAI outcomes in product development/design, research and development, customer engagement, sales, and marketing in most of the Asia Pacific. However, how many business decision-makers actually consider the ethical implications from all angles? After all, AI also comes with its own share of risks – ranging the gamut from deepfakes to copyright infringement. 

Generative AI, specifically, has also shown itself to be prone to making up answers – or ‘hallucinating’ – to respond to prompts it doesn’t know how to answer. There’s also growing concern around biased data and its implications. While that problem is a question of data quality, ensuring data security is also a key imperative for implementing AI. Ensuring data privacy and security is crucial, especially for sectors like finance or healthcare, where organisations hold onto vast troves of sensitive information. 

Let’s also not forget that people are central to AI success. That means equipping employees with the skills to leverage AI tools. In sum, businesses are facing a proverbial minefield. Truly leveraging AI, therefore, hinges on a holistic strategy that prioritises ethics on par with the bottom line and efficiency.

Tackling bias via robust governance

One thing that is fundamental to approaching AI is to recognise that these machines only know what their algorithms and data sets allow them to. If those same algorithms and data sets have not been vetted for biases to begin with, AI does not have the wherewithal to recognise and counter them. And this unconscious perpetuation of inaccuracies is a critical problem today. A NIST report, for example, highlighted that despite systemic institutional and social factors fuelling AI bias, the issue itself was being overlooked. 

Overcoming that, then, rests on expanding perspectives to extend past machine learning pipelines. That begins with AI governance policies, which IBM notes, encompasses the following:

• Ensuring compliance so that AI solutions align with industry regulations and legal requirements.
• Prioritising data protection via trustworthy AI to safeguard customers’ interests
• Acting transparently by shedding light on data is used and what processes underpin AI-powered services
• Adopting methods like counterfactual fairness to ensure equitable outcomes regardless of sensitive attributes
• Instrumentalising “human-in-the-loop” systems to ensure people still oversee AI and maintain quality assurance standards
• Reinforcing learning to overcome human biases and discover innovative solutions

Account for the emergence of Dark AI

You may be familiar with the Dark Web, and in these rapidly evolving times, we’re now approaching an era of Dark AI. As the name suggests, these large language models (LLMs) are essentially being used by malicious actors who want to harness the power of AI for crime, misinformation, and so on. Make no mistake, its emergence is poised to bring about disruption in societal and/or business realms. They currently can take the shape of automated agents, fuelled by open-source and unmonitored LLMs. These Dark AI tools hold the potential for extensive misuse, and could be utilised for a spectrum of nefarious activities ranging from financial fraud and organised crime to bioterrorism and terrorism. 

Furthermore, LLMs introduce a new avenue for exploitation by malicious entities, including the creation of highly convincing and personalised phishing emails, production of “deepfake” videos, automation of illicit financial transactions, or dissemination of detailed plans for unlawful or terrorist acts. Yet, not all is lost. LLMs can also be leveraged for good, and there are avenues for LLMs to be used to bolster cybersecurity. By analyzing vast amounts of data, LLMs can identify patterns and detect anomalies that signal cyber threats, thus enhancing security measures.

However, to effectively harness the power of LLMs for cybersecurity, organizations across Southeast Asia and beyond must equip themselves with scalable data solutions. Scalable data solutions enable the processing and management of large datasets efficiently, providing the necessary infrastructure to support the advanced capabilities of LLMs. This is ultimately the only way to have a comprehensive solution for AI use cases – as it delivers extensive data resources and provides a fully furnished infrastructure characterised by high relevancy and minimal latency. These are core ingredients to leveraging the powers of AI, such as retrieval augmented generation (RAG). That, then, positions organisations to build ethical AI with a strong governance structure, ensuring organisation-wide understanding of AI limitations that will help prevent misuse down the line.