Training the AI personality
It’s common to talk about AI as intelligence, but our understanding of intelligence has historically been connected to individuals, and individuals have personalities. Does AI or more importantly, does each iteration of AI have its own personality? And what implications does that bring to our understanding, treatment, and interaction with different incarnations of AI? How can CIOs responsibly lead their organizations as they bring AI into their operations?
The promise of AI is that it will take on characteristics of human behavior and help people do their jobs better. A variety of machine learning techniques is bringing this promise closer to reality, and companies are putting the results to use in lots of ways ranging from chatbots that help customers get answers and perform tasks more quickly without having to wait for humans, to manufacturing and process automation. We teach AI systems to emulate human behavior, so it is appropriate to think of them in some of the same ways we think about teaching children as they grow and develop, or employees as they learn their roles.
Training the trainers
AI systems have been designed to learn from events and data and to develop a way of dealing with what they learn. Like children, they also develop understandings modeled on the people training them. Accenture reports “Raising responsible AI means addressing many of the same challenges faced in human education and growth.” As individuals learn how to learn, they develop ways to explain their own ways of thinking and decision making. Responsibly raised people then evaluate their decisions and take responsibility for them. It’s incumbent on the employees who deal with and therefore train the various enterprise AI systems to lead them to become responsible AI citizens. “As Artificial Intelligence expands further into society,” says Accenture, “the business accountability around raising a responsible and explainable AI will rapidly grow.” And the CIO is in the right position to guide the enterprise.
As CIOs and their staff bring more AI into the enterprise, they need to be mindful of five practical evaluations they can perform as their AI ‘citizens’ become more deeply embedded. In many ways, these are the same principles used to raise children or train new employees. And for technologists, it’s important to understand the difference between AI and applications. Typical IT application development projects have specific goals and end points even though they may continue to iterate for long periods. But AI systems are more akin to individuals in that they continue to learn and change. CIOs need to develop a framework that can stay focused on progress and direction their AI systems are taking and train their staff on these principles.
Five principles in raising AI
Paul Daughtery, Chief Technology & Innovation Officer for Accenture explains, “Humans have always used technology to advance their goals… but that use needs to be based on guidelines and rules that were agreed upon by people.” These five principles can be defined by and monitored by people to guide the development of the AI personality.
Accountability - Just as individuals are accountable to parents or employers for their actions, AI instances need to be accountable as well. More importantly, the people responsible for training and directing the AI need to be accountable for how they train and how different levels of decisions are left to the AI versus which are transferred to humans. The decisions about who is responsible for defining the point at which control is given to humans needs to be taken seriously.
Transparency - It is not enough for AI to deliver answers and take actions. The system needs to be transparent and be able to provide explanations for its reasoning about how its decisions are made. This critical principle provides a checkpoint to be certain the right data sets are being applied and used with the algorithms that drive the AI and keep processes in check to assure trust in the systems.
Fairness - AI systems are built by and raised by humans and can incorporate biases in the same way children learn their parents’ biases. Extra attention needs to be devoted to assuring AI systems are free from bias, so their decisions and actions are fair.
Honesty - AI systems need to follow legal frameworks and not present subterfuge. Just like individuals they can be guided to present less than truthful ideas that could produce things like fake news and refine it based on reactions. The CIO needs to instill a responsibility for producing truthful and accurate underpinnings in enterprise AI.
Support - The role of AI is to support the individuals it serves, not to replace them. They need to be trained to provide the right information and guidance at the right times and to hand off control and decision-making responsibilities that enhance those they support as their primary mission.
As AI systems grow up, the CIO needs to lead the efforts to deliver the right direction through their staff with clear instruction and continuous follow up.
Inclusive, risk-taking cultures fuel digital transformation success
Jim DuBois Former CIO Microsoft