Must successful AI implementation involve humans in the loop?

What are the critical ingredients needed to boost employee AI acceptance and success amid the Great Resignation and pandemic-linked global challenges?

While organizations in the Asia Pacific region stand out for their high AI adoption rates, what are some of the challenges in implementation?

According to one recent global survey, these include: a shortage of data science/AI talent; immature AI governance policies; and concerns over maintaining data privacy and cybersecurity.

In an interview with Lee Ming Kai, Vice President, Systems Engineering (APAC), Juniper Networks, which commissioned the survey, DigiconAsia.net’s editorial team noted some of the firm’s recommendations for best practices that can optimize AI adoption and reduce implementation challenges. 

DigiconAsia: Do organizations need to already have a data strong management framework in place before enjoying successful AI adoption?

LMK: With several moving parts that facilitate successful AI implementation, data management is indeed quite the important piece of the puzzle.

However, it is likely that data management systems will evolve with the organizational needs as employees increasingly use AI as a daily function, so leaders should not be too discouraged if initial strategies get changed or updated.

Even organizations with established data management systems or practices may need to periodically re-assess if their current system can keep up with the requirements of their AI implementation. Going in with a clear plan that focuses on the evergreen challenges in data management—as well as new ones regarding data governance—will allow organizations considering AI to have a good starting point. 

DigiconAsia: Do you foresee a time when the Human In The Loop model ceases to be useful?

LMK: Using AI in organizations usually involves rapidly evolving data sets, which means that consequent AI models must also change rapidly. HITL must remain a part of the process to keep models up to date with validation datasets from current trends.

Furthermore, it is particularly difficult to label data through automated means: the value of having human eyes to accurately label data across its different functions in AI is still irreplaceable for now. This is especially true for AI systems that provide or require human interactions.

For example, HITL remains crucial in weeding out inappropriate or offensive responses from AI data sets. 

DigiconAsia: How can organizations companies show that AI complements employees’ work and is not a threat?

LMK: Clear communication is a necessary and integral part of any change implementation, especially with AI programs involving pervasive changes. 

It would be beneficial to keep employees informed of the purpose behind each AI function, and emphasize that it exists to make employees’ lives easier, not to endanger jobs.

Our data indicates that AI leaders surveyed had indicated that the ideal use of savings from AI would be for company growth, rather than from cost cutting initiatives. This means that the benefits of AI can lead to employees enjoying higher-value work involving more room for innovation and creativity, skills acquisition and team engagement. 

DigiconAsia: Besides having a strong data management policy, an AI governance plan plus auditability—what else does an organization need to ensure a successful transitioning to AI?

LMK: Our data shows that in the Asia Pacific region, strong oversight of AI is a critical measure for success. The top risks of not having this strong oversight is accelerated hacking or AI terrorism, and data privacy challenges. The strong oversight also involves ensuring tight cybersecurity and cloud availability.

Another critical factor is the prevention and management of the loss of human agency that invariably accompanies such automation. After all, employees will always be the cornerstone of any organization, with or without AI. 

An AI strategy built on clearly defined organizational goals should be implemented on an incremental basis to build a strong foundation that facilitates deeper integration of AI, so that existing employees can see the technology as a solution and partner.  

DigiconAsia thanks Lee Ming Kai for this exchange of ideas.