How Soon Will AI Replace Human Employees?
-
Recently, a CEO of a well-known bank called me to discuss the prospects of generative AI. Initially, we explored various scenarios to enhance fraud detection and customer service. However, with the recent flurry of news, it became clear he had bigger ambitions. Like many industries, banking faces labor issues: a gap between the demand for skilled workers and the supply of those willing to return to offices under pre-pandemic rules.
He believes generative AI might solve this problem. These new tools can reduce costs and improve efficiency through automation, but can they also address talent shortages? Simply put: How soon will AI replace human employees?
This conversation echoes discussions I’ve had since November with executives from insurance, manufacturing, pharmaceuticals, and even Hollywood studios—where writers and actors are currently on strike. They all want to know how to create more value with fewer human resources. This question arose after OpenAI’s ChatGPT went viral last fall, demonstrating AI’s ability to autonomously generate emails, essays, recipes, financial reports, articles, and ideas. Goldman Sachs estimates that 300 million jobs could be eliminated or significantly reduced in the next decade.
Signs of disruption are emerging. 'Prompt engineers,' who instruct systems like ChatGPT to generate content, are being offered salaries of $300,000 or more. OpenAI’s GPT-4 passed the U.S. bar exam, hinting that lawyers may soon be unnecessary for transactional work. Walmart is prototyping a generative AI system (unrelated to OpenAI) to draft supplier contracts, while 75% of contract lawyers and procurement officers now prefer negotiating with AI over humans. Google’s Med-PaLM 2, a model trained on medical knowledge, answers medical exam questions at expert levels. By summer 2023, partners will begin testing apps that can review X-rays and automatically draft mammography reports—without human doctors.
The rapid development in this field is astonishing, which explains why so many executives have reached the same conclusion: within just a few years, powerful AI systems will perform cognitive tasks at the same (or even higher) level as human labor. Enticed by the possibilities of AI, concerned about finding and retaining qualified employees, and feeling insecure due to recent market adjustments or unmet analyst expectations, business leaders envision a future workplace with far fewer people than today. In my view, this is a significant misjudgment.
First, it is still too early to make definitive predictions about AI's future—especially considering that generative AI is just a small part of interconnected fields, each at different stages of development. What jobs AI will eliminate and when remain speculative. For an AI system, merely performing a task is not enough; its outcomes must be proven trustworthy, integrated into existing workflows, and managed for compliance, risk, and regulatory issues.
Second, during periods of rapid technological disruption, leaders tend to focus too much on immediate gains rather than how their value networks will transform in the future. As AI evolves, it will require us to reimagine entire business domains before we fully understand the future. Remember the early days of the public internet and web browsers when they were seen as mere entertainment? No one could have predicted how they would impact presidential elections or the creation of the world's first trillion-dollar companies. Back then, it was impossible to foresee their full implications.
To be sure, today's executives must make decisions in the most complex operational environment I've seen since the early days of the internet. Understandably, leaders worried about missing the next wave of technology are inadvertently making high-stakes bets on their companies' futures. To navigate the uncertain world where generative AI and human labor coexist and evolve in unknown ways, leaders can take the following steps.
Here lies a paradox: we need to view the workforce as evolving alongside generative AI, not being replaced by it. The workforce must evolve, and employees will need to repeatedly learn new skills over the years. Leaders must adopt a new approach to maximize AI's potential in their organizations. This requires tracking key AI developments differently, fostering a ready workforce through iterative processes, and, most importantly, creating evidence-backed future scenarios that challenge conventional thinking within the organization.
So how can leaders navigate this era?
First, temper expectations about what generative AI can and will do for business. Historically, AI has gone through cycles of breakthroughs, surges in funding, and fleeting mainstream interest, followed by unmet expectations and capital withdrawal.
In 1970, influential computer scientist and AI pioneer Marvin Minsky told Life magazine that artificial general intelligence (AGI)—AI with cognitive abilities indistinguishable from humans—would emerge within three years. At the time, the computing power required for such AI didn’t exist, and supercomputers were mostly theoretical. The same was true for personal computers. The Datapoint 2200 and its processor eventually became foundational to what we now know as PCs. Minsky and his colleagues’ grand promises never materialized, and funding and interest dried up. This scenario repeated in 1987, when computer scientists and businesses again made bold claims about AI timelines, only to hit the same wall.
Despite their power, today’s mainstream generative AI tools—ChatGPT, Midjourney, DALL-E 2—are not finished products. Soon, their novelty will wear off, and people will realize that while AI can create content, it’s not yet ready for practical application. Similarly, specialized AI tools in medicine, climate science, and life sciences remain in early stages. Much work remains to fulfill generative AI’s promised miracles of scale and cost efficiency. Remember, these tools were purely theoretical until very recently.
Executives must clarify the practical roles generative AI will play in their organizations today. They should also adopt a pragmatic view of the opportunities and risks it will eventually unlock—we’re only at the beginning of a long journey. From my observations, few leaders are developing realistic strategies to bridge today’s operations with tomorrow’s vision, socializing these plans within their teams, and revising performance metrics accordingly.
Recently, I met with executives from a multinational CPG company eager to collaborate with generative AI firms. I described a likely scenario: a chatbot asks customers about their preferences and needs, then automatically fills their online cart with a week’s worth of groceries. But the CPG’s brand isn’t in the cart—or if it is, it’s not at the top. Just as Google and Amazon invented new mechanisms and rules to optimize search engines, generative AI across retail platforms and shopping apps will pose fresh challenges for CPG companies, potentially relegating them downstream in critical decision-making value chains.
Secondly, assess what data the company is generating and how generative AI will use this data now and in the future. Business data is invaluable because once a model is trained, transferring this data to another system can be extremely costly and technically cumbersome. Emerging platforms today are designed with limited interoperability, making them difficult to integrate. Generative AI platforms are evolving into walled gardens, where the companies creating the technology control every aspect of their ecosystems. The largest AI firms are competing for market share and the vast amounts of data needed to make their models the most competitive. By marketing their platforms to companies, they aim to lock them in along with their data.
Today's AI systems are built using a technique called Reinforcement Learning from Human Feedback (RHLF). Essentially, AI systems require continuous human feedback; otherwise, they risk learning and memorizing incorrect information. The more data input, the more annotation, labeling, and training are required. Currently, this work has been automated in places like Kenya and Pakistan. As AI matures, the demand for experts with specialized knowledge is emerging. Many business leaders I've encountered have no plans to include an internal RHLF department responsible for ongoing monitoring, auditing, and adjusting AI systems and tools. (No leader wants to see an unsupervised AI system deciding how to evolve on its own.)
Even with trained personnel involved, companies must continuously develop strategies to expose risks associated with generative AI systems, especially those operated by third parties. AI systems are not static; they improve incrementally over time. With each new development, new potential risks and opportunities arise. It's impossible to rule out all potential negative outcomes in advance without rapidly testing these predictions. (It's currently impossible to create a Monte Carlo simulation that can perfectly predict the future.) Instead, there should be a dedicated team monitoring the learning of generative AI systems and related cybersecurity challenges. They should develop short "what-if" scenarios to anticipate potential errors.
Similarly, as AI advances, opportunities for new growth will emerge. This means companies should also have a dedicated internal business development team to create near- and long-term scenarios for how emerging tools can enhance productivity, efficiency, product development, and innovation.
Third, when it comes to AI, leaders must shift their focus from the front lines to the top level. This may seem counterintuitive, as many view generative AI as a way to reduce operational costs. Today's intelligent chatbots will soon give way to multimodal systems capable of addressing multiple problems and achieving diverse goals simultaneously. Imagine a property insurance company where every underwriter interacts with AI. Initially, the underwriter might ask the AI to assess risks related to an insured property. After a preliminary text analysis, she might request it to refine the results using images from inspection reports or audio interviews with potential policyholders. She might toggle between different data sources until arriving at the best quote for both the insurer and the client.
The key to leveraging multimodal AI effectively lies in understanding what and how to delegate to machines, enabling humans and AI to accomplish more through collaboration. However, delegation is a common challenge for professionals: either too much is delegated, too little, or the wrong tasks are assigned. Working with multimodal AI requires employees to master the art of delegation.
Once employees understand how to delegate tasks correctly, it becomes a force multiplier within the organization. By conceptualizing and simulating new revenue streams, identifying and acquiring new customers, and seeking various operational improvements, teams can be more ambitious in increasing company revenue.
This foreshadows the need for a different approach to skill development in the future. Most employees don't need to learn how to code or write basic prompts. Instead, they need to learn how to leverage multimodal AI to do more and better work. Consider Excel, used daily by 750 million knowledge workers. The software includes over 500 functions, but the vast majority of people only use a few dozen because they don't fully understand how to match Excel's extensive capabilities with their daily cognitive tasks. Imagine a future where AI, an even more complex software, becomes ubiquitous. How much utility will be left if business leaders approach skill development too narrowly?
Workforce transformation is an inevitable side effect of technological advancement, and leaders need a systematic way to envision their organizations post-generative AI. To this end, a simple framework can help leaders predict how and when the workforce needs to change to leverage AI effectively. The goal here isn't to make long-term predictions but to prepare organizations for the ongoing evolution of AI (see the "IDEA Framework" diagram).
In this era of change and uncertainty, the best thing organizations can do is methodically plan for the future. This requires understanding the limitations and strengths of generative AI and fostering a culture of continuous evaluation and improvement. Leaders should resist the temptation to reduce headcount and instead use strategic foresight to create a future where highly skilled employees can leverage AI, and human-AI collaboration yields greater productivity, creativity, and efficiency than either could achieve alone.
Amy Webb is a quantitative futurist, CEO of the Future Today Institute, and a professor of strategic foresight at NYU Stern School of Business. She is the author of The Signals Are Talking: Why Today’s Fringe Is Tomorrow’s Mainstream, The Big Nine: How the Tech Titans and Their Thinking Machines Could Warp Humanity, and The Genesis Machine: Our Quest to Rewrite Life in the Age of Synthetic Biology.