In his book Age of Discontinuity first published in 1969, Peter Drucker wrote about how the explosion of new technologies is a key factor in shaping disruptions that can accelerate economic expansion. Today, as the world faces economic uncertainty due to a black swan event, businesses are increasingly scaling their use of digital technologies such as Artificial Intelligence (AI) to remain adaptable.
Technology deployments though can be successful only when there is trust amongst all stakeholders—service providers, customers, governments, and regulators. More notably, trust can become a business differentiator and provide a competitive advantage for companies. Trust is particularly important for AI, as there may be many questions surrounding its fairness, explainability, and security. If businesses cannot explain to the end-users how and why their AI system is making certain decisions, it may lead to unfounded fears of mysterious algorithms that do not evoke trust. The stakeholders can only trust algorithms when they see how they were created and how they work. Thus, the question arises how can companies build trust?
The fundamentals of trust
Companies need to move beyond the ‘build for performance’ approach and focus on the five fundamentals—fairness, explainability, robustness, transparency, and privacy—to enable trust in their AI models and business outcomes across the entire AI lifecycle.
There is scope for an AI system’s output to be unfair owing to the technical limitations of its design (algorithm) and the data used to train and test the system. Bias is likely in instances when an AI system is used in contexts that were not anticipated or to make decisions about communities that developers may not have considered during development. Hence, businesses must leverage diverse development teams and include the viewpoints of external organisations that work with different vulnerable communities. Fairness is an essential aspect to consider as a well-calibrated AI system can assist businesses in making fair decisions while mitigating biases to ensure equitable treatment of all individuals and groups.
An AI system that is not perceptible is not ethical; its operations and functionalities should be explainable to the people for whom it has implications. Businesses must be able to contextualise how and why their AI system arrived at a particular conclusion with detailed documentation that provides information on the levels of procedural regularity, confidence measures, and error analysis. Companies should remember that a good design never sacrifices transparency.
As AI systems are widely used to make important decisions, they must be robust and secure. They must be able to thwart malicious attacks and not cause unintentional harm in the process. AI-powered systems must be able to mitigate security risks thereby enabling confidence in system outcomes.
The end-users should be able to trace the AI decisions that affect them back to the data source and the model. A transparent AI system makes the information on data collection, usage, storage, data processing, available. Transparency reinforces trust and therefore businesses should rely on disclosure to build trust.
Businesses must ensure their AI systems prioritise and safeguard the privacy and data rights of consumers and inform them on how personal data may be used and protection mechanisms that have been put in place. AI systems must collect and store only the minimum data necessary for operations, and ensure that consumer data is not repurposed for unauthorised usage. Companies must make sure their AI systems allow consumers to decide on the collection, storage and usage of their personal data through clear and accessible privacy settings.
Most importantly, businesses need to look at trustworthy AI not as an afterthought to the AI lifecycle but as a strategic and long-term investment/consideration. Moreover, today businesses need AI to transform and tackle the economic uncertainty that is accelerating change and fostering disruption. They cannot scale their AI deployments without building trust in the AI systems amongst the various stakeholders. Building trust is a must — lack of it can be the biggest obstacle to the widespread deployment of AI.
About authors: Gargi Dasgupta is Director for IBM Research India & CTO IBM India/South Asia, and Sameep Mehta is Distinguished Engineer at IBM Research India
The thoughts and opinions shared here are of the author.
Check out our end of season subscription discounts with a Moneycontrol pro subscription absolutely free. Use code EOSO2021. Click here for details.