The classic Pink Floyd song Welcome to the Machine, released in 1975, pointed listeners to the reality that, even for rock stars, something beyond the scope of personal autonomy was always in play, and control was inescapable. The idea of ubiquitous control, however, is fast becoming a genuine concern in the Artificial Intelligence (AI) world, particularly for businesses seeking to get an AI program off the ground. According to Genpact Chief Digital Officer Sanjay Srivastava, with the rise of data issues and the concerns over understanding how AI works internally, the need for machine control and increased knowledge for new AI business applications is paramount.
Srivastava, the digital leader at 70,000 employee Genpact, recently gave his perspectives on the growing needs for businesses seeking to begin using AI. Because AI is still, in some sense, fledgling, business models must account for at least three difficulties in maintaining the necessary control to make applications both functional and profitable.
First, the CDO made it clear that knowledge of how an AI application is reaching a particular decision must be built in to any properly functioning AI model. Often called the ‘black box’ problem, how AI reaches a particular decision can often be obscured and therefore, while creating a desired result, not provide the necessary knowledge for future advancements. This lack of transparency reduces the ability to maintain machine control and limits growth in functionality. Therefore, businesses seeking to create AI programs must be careful to create the necessary transparency along with the proper machine learning.
Second, all AI is data dependent. Greater levels of data create greater and more robust AI models, and data volume has never been higher. In fact, the data created in 2017 exceeds the total sum of all data created by humanity in all of human history prior t0 2017. However, for businesses seeking forays into the world of AI, data sets may be limited in scope, leaving the AI ‘thirsty’ for more information in order to proceed. Future AI solutions in business, per Srivastava, must be created with enough machine control to function in low or moderate data environments and still find helpful and forward-moving solutions. This type of AI need puts the onus on developers to create new out-of-the-box solutions.
Third, Srivastava made it clear that for AI to be functionally viable and helpful, businesses must be able to convert AI results into profitable insights. To date, even simple results from AI processes have been a challenge. Moving forward companies must be able to transition into broader applicational solutions in order for AI to be a useful business tool. Without greater knowledge and machine control, AI applications may become so complex that companies are left wondering what the results even mean, while at the same time wasting vast sums of money.
While Pink Floyd’s dystopian vision of societal control terrified a generation of rockers, the need for ubiquitous machine control in the growing AI world has never been greater. While the wonders of AI are apparent, meaningful applications for businesses seeking to start AI programs will require increased knowledge, better low data solutions, and greater internal transparency if solutions are to be gained.