On June 8, the Wall Street Journal published an article on Sunday, exploring when AI will fully realize its potential and have a transformative impact on the social economy like electricity and the Internet. The article stated that there are many reasons to believe that the development of AI will be slower than its most optimistic supporters believe, but faster than skeptics say.
It will take time for AI to truly transform
It has only been about 1,200 days since OpenAI launched ChatGPT. Yet if the most extreme AI optimists are to be believed, the technology is supposed to have revolutionized the business world. In other words, this change will happen at any time. At the same time, AI critics are just as easy to find, arguing that AI is just the latest tech fad that is destined to fade quickly before any real results are achieved. In their view, this decline can happen at any time.
The truth is far more complex than AI hypemen and critics say. Walk into a large company today and you will find that AI is both everywhere and seemingly nowhere to be found.Employees use it to summarize meetings, draft emails, and generate first drafts of presentations. But these efficiency gains have yet to translate into clear, economy-wide productivity jumps or fundamental changes in the way people work.
So, how long will it take for AI to realize its potential and bring about change? Answering this question requires teasing out many of the challenges facing the business world: organizational inertia, human resistance to change, limited and often quite messy data, privacy and security concerns, and the imaginative leaps required to redesign how organizations actually work.
get along with
Despite all the complaints and negative press, AI is indeed making progress in the business world. Surveys of chief information officers (CIOs) and CEOs consistently show that companies plan to increase investment in AI this year and next. A research report released by Deloitte in January and a separate study from the Wharton School of the University of Pennsylvania indicate that large companies are moving out of the experimentation phase and starting to integrate AI into core operations. Wharton research released last fall also found that three-quarters of 801 executives surveyed reported positive returns on their AI investments.
These results are gradually emerging in multiple industries. Retailers are using AI for real-time pricing and product recommendations; private equity firms have built AI analysts to integrate research information and assist investment decisions; and manufacturing companies are deploying computer vision technology to detect defects on production lines.
The area where progress has been most significant is in software development. AI has become so powerful at writing code that many software engineers can simply describe requirements in natural language and AI will do the rest.
Professor Mollick denies AI application stagnation
Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania who studies how companies adopt AI, said that given the above situation, it is completely wrong to think that "AI applications are stagnating." “The idea that we’re still stuck in pilot mode is outdated and wrong,” he said. “I’m constantly talking to companies that are getting real value from AI.”
limited impact
But in the business world, the AI revolution still faces many obstacles. First, there is a basic skepticism about all the hype: Boards of directors and investors continue to demand clearer evidence from companies that AI investments are paying off. Moreover, at least so far, AI has not demonstrated its generality enough to prove that it can transform enterprises and industries on a large scale.
Researchers have coined a term to describe this uneven capability of AI: "jagged frontier." Benedict Evans, an independent analyst who tracks enterprise AI adoption, said AI models are great at some things and surprisingly bad at others, and often it's not discovered which tasks fall into which category until companies are already using them.
For example, AI excels in clearly structured tasks, such as programming, legal document review, and financial analysis. But when it comes to tasks that are more context-dependent and take up the majority of work time, this “unevenness” is exposed. It will give wrong answers with extreme confidence and cannot rely on human factors that have never been included in the training data, such as judgmental decisions, unwritten rules, and long-term intuition.
This is an obvious "hard ceiling" for current AI capabilities. Nobel Prize winner and MIT economist Daron Acemoglu said: "Whether you are a CEO, a manager, a journalist, a professor, or a construction worker, I think your skill level is above that of existing AI." He believes that current AI tools will only have an impact on a small number of jobs.
In addition, for AI to truly work, a lot of "packaging" is needed: appropriate data, appropriate permission settings, complete security and restraint mechanisms, and clear roles defined for humans who supervise AI. Because every company's systems and workflows are different, this supporting "architecture" often has to be built from scratch. And this is much more difficult than it seems.
human disorder
But as far as obstacles go, technological problems may be easier to overcome than human problems. Simply put, a lot of people need to be convinced before the AI revolution can truly take off.
Corporate executives are faced with five-year planning cycles, depreciation schedules for procurement systems dating back three years, and boards demanding returns. In such an environment, risk aversion is not irrational. At the same time, there are also issues at the employee level: those employees who think that they are "training the AI that will replace them in the future" are unlikely to actively cooperate with the implementation of AI.
“What’s being marketed is the idea of productivity and efficiency,” said Kate Brennan, associate director of the AI Now Institute, an AI policy research center. “What that means for the people doing the actual work is rarely included in the discussion.”
Management and employees may also be hesitant to truly integrate AI into operations and not just use it for menial chores. People’s instinct is often to use AI to automate certain aspects of existing processes, rather than rethinking the entire process itself.
Take, for example, an insurance company that handles minor car accident claims. Typically, companies use AI to speed up document processing while retaining the original multi-layer review and approval process. But the real opportunity lies in completely redesigning the entire process, allowing AI to assess the extent of damage based on photos taken by the customer, then approve the claim and trigger payment almost immediately. This reimagining is difficult and threatens established hierarchies and conventional ways of working.
time for change
Finally, it’s important to remember that transformative technologies often take longer than expected to deliver the kind of profound change their advocates promise.
Electricity reshaped civilization, but it took a full four decades for its impact to show up clearly in productivity data. The Internet restructured the foundations of business, work, and global competition, but it would take ten to fifteen years to penetrate the backbone of the economy. From an internal perspective at the time, the early stages of the Internet were quite similar to the current situation of AI: promising prospects, uneven results, and the entire industry had good reason to tell you: the revolution has arrived.
"To really change an organization and achieve significant change, you need to measure time on a human scale." said James Landay, co-director of Stanford University's Human-Centered AI Institute, who has been paying attention to the difficulties companies face in trying to absorb new technologies for many years.
"My judgment is more like five to 10 years, not the next two or three years," he said.
AI will almost certainly have as profound an impact as the Internet, and will likely take almost as long to reshape the economy. Supporters are generally right about the direction of development. The skeptics are also probably right about how long it will take.
Perhaps the most valuable way of thinking for any business executive, investor or policymaker right now is to accept both judgments simultaneously.