AI Is an Exciting Opportunity, Not a Threat

• Pessimistic forecasts that AI will replace large numbers of human workers are overblown—instead, AI is more likely to help human employees improve their productivity.
• AI can only generate ideas based on existing information, not imagine radically new ideas. This makes it perfect for taking over the most predictable and repetitive tasks.
• Once students learn to use AI to assist in their everyday responsibilities, they can spend more time on meaningful and creative activities that provide greater value to organizations and to society.

 
Artificial intelligence dates back to the 1950s, when the first mainframe computers were introduced. But it has only been over the last two years that rapid advancements in AI have inspired alarming predictions, with some experts forecasting that by 2030 machines will replace large numbers of human workers. The McKinsey Global Institute estimates that 30 percent of human work hours could be entirely automated, while the International Monetary Fund predicts that 60 percent of jobs could be lost in advanced economies.

Such gloomy forecasts stem mostly from recent developments in deep learning and generative AI (GenAI) technologies that allow machines to identify patterns in massive amounts of data. By using those patterns to reproduce decisions, AI can generate results that match those of professional experts in many contexts. That means that a total replacement scenario looms for any workers whose tasks are repetitive and easily replicated by algorithmic predictions.

But most professions are not based solely on repetitive tasks. In fact, AI is likely to help most human employees significantly increase their productivity. For example, AI can analyze how skilled call center workers have answered customer questions and then recommend appropriate answers to less-skilled workers to improve the handling of customer queries. If human workers learn to apply AI effectively in their jobs, they can use its capabilities to enhance their decision-making, develop their expertise, and maximize their own potential.

In other words, we must encourage students to view the rise of AI not as the harbinger of massive job displacement, but as a fantastic opportunity to make unprecedented gains in human productivity. Here I share four things that students should know about AI—and the reasons I think this technology will benefit workers, not replace them.

1. AI Is Not Intelligent

There are two types of AI. Narrow AI is trained to be task-oriented—it performs a specific activity in a structured environment. This might include, for example, classifying whether bank customers qualify for loans based on their credit scores. By contrast, General AI is trained to be goal-oriented—it identifies the optimal method to attain a specific goal. One such task might be counseling bank customers on the most appropriate investment strategies given their personal and social circumstances.

Ongoing research will continue to improve the performance of AI algorithms over the next decade. But truly reliable and wide-reaching General AI that can completely replace human workers is still a faraway aim, if it is possible at all.

This is true for three main reasons. First, machine learning algorithms need a vast amount of data, human-led training, and human supervision to perform moderately complex tasks, as Kate Crawford points out in her book Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence.

This idea is reinforced in the book Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins by Garry Kasparov. At one point in the book, Kasparov discusses how an early chess program trained with machine learning lost games very quickly. Why? Because the program had analyzed past game data and observed that sacrificing one’s queen often leads to victory. However, it failed to understand that queen sacrifices were an elaborate strategy of chess champions and only suitable in a narrow set of conditions.

Truly reliable and wide-reaching General AI that can completely replace human workers is still a faraway aim, if it is possible at all.

Second, AI can make such errors even when it is thoroughly trained on an enormous amount of data, such as those tools that use large language models (LLM). Gary Marcus, a psychology professor and prominent critic of ChatGPT’s creator Open AI, once showed that older versions of ChatGPT could not respond to many simple user requests. For instance, they could not generate text with a certain number of words or rank sentences by word count. Here, the program failed because it does not understand language. Rather, it draws from a complex database of knowledge to predict which words are most likely to be statistically associated. (That’s why, when ChatGPT answers a prompt, the words appear slowly in sequence, as the algorithm predicts which word is statistically most likely to come next.)

Finally, while AI is good at intrapolation, it is very bad at extrapolation. Given that a machine learning algorithm identifies patterns in existing data for predictions, it performs best in cases that are known and coded in the training dataset (intrapolation). However, in cases where AI is asked to make predictions for which the outcomes are unknown (extrapolation), its outputs are severely unreliable. The quality of its predictions can depend only on the amount and the quality of the trained data.

Therefore, AI is good at predicting when people with certain credit scores are good loan risks, because past trends are reliable indications of future results. But AI is bad at predicting presidential elections because of the limited amount of data available and the unreliable nature of opinion polls.

2. Human Creativity Is Not Replaceable

Albert Einstein once said that “the true sign of intelligence is not knowledge but imagination.” We should keep his words in mind as we evaluate GenAI tools.

One good example is Midjourney, an AI-powered text-to-image generator. One might be fooled into thinking that the illustrations this platform generates are new creations. However, Midjourney generates images from a known repertoire of existing images; it creates “new” faces only by combining features of existing faces.

AI algorithms powering Midjourney and other platforms are built in ways that favor knowledge accumulation and deployment, but their outputs can potentially constrain human imagination. For instance, a 2023 study shows that, because GenAI’s suggestions are derived from existing knowledge, its collective use for marketing new products might result in very similar visuals and slogans over time.

Humans, on the other hand, perform far better than machines on tasks that require unstructured creativity or “human” attributes, such as humor (in advertising), empathy (in counseling), and appreciation of cultural differences (in management).

Humans also have a natural ability to divert from known paths almost instantaneously. For example, if a recruiter asks 10 job candidates to make a case for why they are the best choice for a position, it is likely that they each will try to respond in creative ways that stand out from the crowd, rather than rely on conventional tactics such as pointing to their degree specializations or grades.

3. AI Cannot Build on Relationships

Another 2023 paper highlights the claim that AI cannot access what some researchers call relational expertise. This term refers to the assumption that we work in a social web of relations, in which our expertise exists only because it is recognized by others and developed through interactions with others. That is, we are important in our roles only because others trust our expertise enough to follow our guidance and decisions.

If this idea holds true, there are many jobs where humans could never be replaced by machines, simply because their expertise can be developed and maintained only through relations with other human beings.

4. Humans Prefer Human Interaction

To build on that last point, many studies show that most humans prefer to interact with other humans. In job recruitment interviews, candidates prefer to speak to a human because they believe that AI is less easily influenced and will not grasp the nuances of their motivations and personal histories. In medicine, patients still want to deal with a human doctor, even if they acknowledge that machines can outperform doctors in diagnosing illnesses and prescribing treatments.

We must urge our students to be skeptical of the gloom-and-doom predictions surrounding AI—indeed there is little to fear.

While some experts predict that AI will take over some professions within the decade, the belief that only humans can understand other humans is likely to endure far longer. In healthcare, patients want medical personnel to understand their pain. In the banking industry, customers want employees to understand the social contexts of their financial distresses. In the business world, employees want their work reviewed by human managers who can best assess their progress in abstract areas such as soft skills.

Sentient and empathetic AI that can replicate the nuances of human interactions will remain a myth for decades to come. In the meantime, in most contexts, we humans will prefer to deal with humans.

How Should We Train People to Use AI?

In the last decade, AI has allowed humans to automate entire task workflows. If we look at human resources management as one example, we can see that many activities have been or are being automated, such as selecting job candidates, handling expense claims, training and assessing employees, classifying and extracting information from documents, and collecting and consolidating data.

But for other tasks that require human judgment and abilities, such as management, AI can be a fantastic tool to provide unmatched assistance for decision-making. That’s what we must teach our students, so that they learn how to ask AI to provide reliable and relevant information that will help them make good business decisions. When we offer them opportunities to use AI tools to sort through vast amounts of data, students can learn how to gain insights that they might not otherwise discover solely with their human cognition.

As educators, we also must urge our students to be skeptical of the gloom-and-doom predictions surrounding AI—indeed, there is little to fear. We can train students to use different AI tools in ways that help them thrive in their future managerial roles. Using AI effectively will give them more time for meaningful activities that bring more value to their organizations and society at large—it might even help them feel greater satisfaction in their work.

Adapting to a New Future

As humans, we have unique adaptive capabilities, and we have quickly learned to incorporate past technological advancements into our work and lives in many beneficial ways.

Here, I offer one last example: Some experts claim that LLM translators will end the careers of human translators—and even the need for anyone to learn foreign languages. It is true that LLM translators might displace those who work only with technical, unambiguous texts that hold the same or similar meaning in other languages.

But natural language processing systems that have been trained on words and sentences alone can only approximate the meaning of the original texts. Humans will always outperform machines in translating ambiguous or newly written prose or works whose meanings are deeply rooted in relational, cultural, and historical contexts. The rise of sophisticated AI-powered translators only means that translators now can focus their time on more interesting texts.

The same is true for business professionals. Students, new graduates, and seasoned managers should view the rise of AI as a privileged opportunity to develop new skills and to discover ways to spend more of their time on satisfying and rewarding activities that offer greater benefit to their organizations and society. This is an exciting moment in history!