Building A Responsible AI

Artificial intelligence (AI) has transformed the way we interact with each other and use technology. As a technology, AI is integrated seamlessly into our lives -– from automated routing services to voice transcription to product recommendations.

However, even as the benefits of AI that include customer satisfaction, revenue lift, and efficiency gains can be very compelling, enterprises must be liable for the actions driven by their AI solutions. Ignoring this could cost a company heavy in the form of irreversible damage to their supply chains, engagement programs, brand reputations, and bottom lines.

Besides these legal and financial implications, companies also carry a moral obligation of constantly refining AI to eliminate issues such as bias that have afflicted the technology for long.

All this boils down to having a 'responsible AI' -- one that imbibes empathy, ethics, and fairness.

Let's look at how each of these pillars can help build a 'responsible AI.' 

Building Empathy to Deliver Contextual Message

Empathy entails understanding and identifying with another person's feelings. On the customer engagement front, this implies putting yourself in the customers' shoes and looking at actions and offers from their perspectives. 

In traditional customer engagement approaches, every communication is decided much in advance and applied to a large group with absolutely no insights into what a specific customer needs. What is needed instead is the ability to make real-time decisions so that you can understand the customers' context, what they need from you at that moment, and how they want those needs to be fulfilled. 

With its ability to analyze customer data, AI can aid in making an organization more empathetic by determining the right message that needs to be delivered, at the right time. 

For example, a company can direct a customer from a hurricane-ravaged area (or even as the hurricane is approaching) to a local shelter or enable a single mother to renew her health policy online during her lunch break. It could also contact a customer proactively and upgrade the phone data plan that could help save overage charges. 

A responsible AI solution will also flag inappropriate messaging (such as irritating sales offers) or messages that aren't being favored, while persisting with those that are being well-received. As building responsible AI hinges on keeping empathetic considerations in mind, the onus lies on the AI developer to define the right guardrails and rules for it.

Bringing In Transparency to Gain Trust

In regulated industries such as insurance, financial services and healthcare, transparency assumes extreme importance. Regulations such as GDPR make it mandatory for companies to explain their decisions or face huge fines. 

There are examples when a lack of transparency in AI has led to negative publicity for some of the most customer-centric companies in the world. One such company saw itself at the receiving end when it launched a new credit card. Customers with the same profile and financial information were receiving varying credit limits. 

The company had no answer to this miscalculation. The real culprit was not an employee but the AI, which had gone rogue. The biased decisions made by AI were not easily detected, and the mishap led to a loss of face for the company. 

There are several areas -- ranging from self-driving cars to medical diagnosis -- where transparent algorithms are critical. Companies in these areas, therefore, need to be proactive in certifying their algorithms and ready in providing a clear explanation of the basis of their decisions. 

The prerequisite for AI to be transparent is that it must be explainable. The suggestions or decisions provided by it have to be understood by its developers and users.

Inculcating Fairness to Avoid Bias

AI keeps learning and evolving. However, how it evolves depends on how diverse the information is. This can result in issues around fairness.

The American criminal justice system is an example of how bias in AI can be life altering. Algorithms are used to assess a criminal defendant’s likelihood of becoming a repeat offender.

ProPublica examined the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) in 2016. The examination found the system erroneously judged black defendants to be at a higher risk of reoffending as compared to the white defendants. It also incorrectly flagged white defendants as low risk.

Algorithms in AI make decisions based on data available to them. The developers at COMPAS had no intentions of developing a biased AI. This was a result of the bias that existed within the judicial system itself. 

To build an AI on the foundation of fairness, models, data, and strategies must be included in such a way that they are balanced towards all. 

AI for Universal Good

Almost all companies have a code of ethics and a set of values that its employees follow. Why shouldn't the technology being deployed then adhere to the same standards? 

Building responsible AI is not just about reducing risk. It also strengthens customer trust, thereby amplifying brand value. Enhancing customer confidence also proves as a catalyst in depicting AI in a more positive light, which in turn accelerates its innovation and adoption.

Enterprises are at the forefront of innovation and are creating benchmarks on which all future AI will be utilized, created, and evaluated. Organizations, therefore, should develop the ability to manage and master their AI and direct its power towards enhancing outcomes both for themselves and their customers.

Suman Reddy

Guest Author Suman Reddy is managing director of Pegasystems in India. Pegasystems Inc. is a Nasdaq listed company that develops agile and strategic applications for sales, marketing, service and operations.