Rethinking AI: Why industry leaders must take a balanced approach

Image: Shutterstock

Heralded as the next wave of transformative technology, AI has catalysed the business landscape. Its rapid adoption across the industry has been marked by its transition from a tool for improving operational efficiencies to a guide to company strategy. However, the skewed focus on Generative AI (GenAI) has led many to overlook alternatives, causing IT leaders to inflate the feasibility of AI. Since many companies’ first interaction with AI is GenAI, there is the risk of mistaking it for all of AI itself, leading to misplaced expectations.

Although GenAI models excel in content generation, its content could be inaccurate due to sampling biases and the lack of representative datasets. Additionally, the model may not perform well in environments for which it has not been trained or training data is inadequate. To mitigate the risks of discrimination, user testing becomes vital to product development. Research has shown that getting representatives from diverse groups to test AI products before a public launch fosters inclusivity in the product’s design process. This would mitigate the risks from over-fitting to training data (when the model cannot accurately predict values in the larger population and only works effectively on the training dataset).

On one hand, GenAI models are proficient in content generation, knowledge discovery, and conversational interfaces, but on the other, they often prove to be unreliable as autonomous systems. Current GenAI models are not robust enough to be autonomous systems and require continuous human monitoring. It remains too risky for organisations to rely completely on GenAI’s outputs while making critical decisions involving people management, supply chain management, fiscal management, and strategic planning.

While GenAI has a lot of potential, using it solely due to the hype could lead to risk for its application, where it may not be an optimal fit. There is a need to evaluate whether GenAI is the right option for your use case. Moreover, there are other proven techniques in the AI landscape.

1) Nongenerative machine learning: Nongenerative machine learning techniques trained on appropriate historical data may be more suitable than GenAI for category predictions. High-value use cases such as irregularity detection, personalisation systems, consumer churn forecasts, and predictive maintenance are a few areas in which it may be a better fit than GenAI.

2) Optimisation: Optimisation aims to maximise benefits while managing trade-offs between various business objectives. It does so by allocating the optimal combination of resources within given limitations. Optimisation is key for planning use cases such as pricing strategy, financial portfolio optimisation, inventory management, and budget allocation. It can support strategic decision-making, evaluate alternative paths of action, and assist autonomous systems.

3) Simulation: Simulation aids in testing various scenarios and dynamic changes without real-life implications. It can produce artificial data to train other AI models for financial modelling, supply chain management, strategic scenario analysis, workforce planning, and manufacturing process simulations. It is a noteworthy alternative to GenAI for use cases in forecasting and decision intelligence since it can generate content in a controlled and rational manner.

4) Rule-based Systems: Rule-based systems are easier to implement because they are easier to interpret than GenAI. This makes them a more effective choice for sensitive use cases like risk assessment, medical diagnosis, quality control, fraud detection and loan approval.

Combining GenAI with other AI flavours can mitigate the limitations and risks of using only GenAI. The synergy could yield superior results and improve user interaction.

Another reason why GenAI might not align optimally with a use case is that some risks cannot be fully mitigated. These include output inconsistencies and threats to cyber security, data privacy, and intellectual property. With the rapid development of AI, there is also a risk of aggravating inequalities or misuse.  For responsible development and deployment of AI, there is a need for legislation that addresses bias in AI, protects the privacy of individuals, mitigates broader societal impacts (such as environmental sustainability) and fosters trust and accountability in AI systems. The UNESCO recommendations on the ethics of AI serve as a well-rounded guideline. As AI continues to permeate into various facets of business, discussions among policy makers and business leaders to implement regulations must take place to affirm societal well-being.

As GenAI reshapes industries, leaders must adopt a strategic and discerning approach to its implementation. While it holds significant promise, it should not be seen as a universal solution. The true potential of AI lies in understanding when and how to leverage GenAI alongside other established analytical techniques.