There has been quite a bit of press lately about the bursting of the AI hype bubble. (About time, right?) “The AI hype bubble is deflating. Now comes the hard part,” read the Washington Post in April. “AI hype is not a replay of the dotcom bubble, but it’s a remix,” blared the Financial Times in July. “Google’s Monopoly Money + Is the A.I. Bubble Popping?” was the theme of an August New York Times podcast.
However, there’s actually nothing new about the rise and fall of the AI hype cycle. It’s been going on for decades.
That’s the word from Paul McDonagh-Smith, senior lecturer of IT at MIT Sloan School of Management. “While AI technologies and techniques are at the forefront of today’s technological innovation, it remains a field defined — as it has from the 1950s — by both significant achievements and considerable hype,” he told Forbes.
In the days since AI was first formulated as a way to inject intelligence into machine routines in the 1950s – all the way up to the 1970s, AI pioneers, including Alan Turing, Stanford University’s John McCarthy and MIT’s Marvin Minsky, “set high expectations that were later chilled in the AI Winter of the 1970s and 1980s,” McDonagh-Smith related.
During that initial ice age, “computational and data limitations meant the gap between theoretical potentialities and practical applications resulted in reduced investment and a growing skepticism and sense of AI overhype,” he said.
Machine learning, developed in the 1980s, revived AI’s momentum. “AI applications, fueled by increased compute power and availability of digital data, expanded into areas including expert systems and natural language processing,” McDonagh-Smith continued. These evolved through the 1990s and early 2000s, “creating the pathways for AI’s evolution to deep learning, enabled by breakthroughs in neural networks, and exponential improvements in the capability and performance of AI in tasks such as computer vision, natural language understanding, and complex decision-making processes.”
The bottom line is that “the question of whether AI is overhyped or is rising to meet its expectations has been a constant over the last 70 to 80 years,” he said.
The key to avoiding another AI Winter is to address the last mile between machine and human capabilities. “We will benefit from applying last-mile engineering where AI’s potentialities are converted to sustainable business value by connecting them with human capabilities — such as creativity, curiosity, critical thinking and compassion.”
Data quality and availability are also currently the major issues that are slowing down or inhibiting AI from living up to its promise. “AI systems are fundamentally reliant on the quality of the data they are trained on,” said McDonagh-Smith. “I see many organizations struggling with data silos, inconsistent data formats, and complex privacy concerns that span geographies and jurisdictions.”
Add to that a significant skills gap, “with a shortage of professionals proficient in not only AI and data science but critically the creative skills to convert AI potentialities to business impact,” he added.
To mitigate these challenges, “leaders in our organizations need to ask themselves and their businesses three simple questions: What’s my enterprise data strategy? What’s my enterprise people strategy? What’s my enterprise AI strategy?”
McDonagh-Smith also advocates “a dual-speed approach to AI strategy where we conduct fast experiments that capture data that then informs the slower longer-term strategic AI trajectories to be explored and mapped.” He defines longer-term as being at least 18 to 24 months.
“Enterprises making best headway with AI today recognize the value of investing in robust data governance frameworks with a strong focal point of leadership presence to ensure data integrity and compliance,” he continued. “These companies are also working to understand the implications of the exponential growth of computing power we will see over the next one-three-five years.”
Successful companies “are adopting compound innovation where they are combining AI technologies — such as deep learning and genAI — with compute capabilities — such as cloud and edge — and a commitment to exploration and engagement that creates opportunities for business model innovation and a recalibration of business methods.”
