What human abilities can compensate for the deficiencies of AI? Here's what MIT has to say...
Editor's Note:
Today, AI has been deeply integrated into our lives: some use Kimi to write code, while others use Doubao to generate images... Technology has broken down all barriers and popularized knowledge to the public. However, concerns have also arisen: when AI masters all recorded human wisdom, will human thinking and creativity become obsolete and useless?
Marshall McLuhan, a master of communication theory, once proposed the theory that "technology is an extension of the human body" - cars are an extension of legs, and telephones are an extension of voices; Steve Jobs compared computers to "bicycles for the mind," which expand the scope of human thinking, problem-solving, and communication. By analogy, AI might be a "rocket ship for the mind."
More people believe that in essence, whether it's a computer or AI, they are just "tools" for the mind, not "replacements." We shouldn't confuse "creativity" with "the ability to recall facts" - AI can indeed retrieve facts accurately, but it can't replicate the most core functions of the human mind: processing information, interpreting information with unique cognition, constructing personal narratives from known things, and conveying ideas and arousing resonance through "storytelling."
The Shortcomings of AI
What impact will AI and emerging technologies have on our lives? Debates on this topic can usually be divided into two categories: one explores how AI can enhance human work capabilities, and the other focuses on how AI-driven automation will disrupt human work or even replace humans.
The research team from the MIT Sloan School of Management stepped out of this framework - the question they posed was: "What human abilities can make up for the deficiencies of AI?" Their research reveals that although current AI can impact "human-like cognitive abilities" such as creative writing and brainstorming, it still faces five challenges that are difficult to overcome, which also make the unique value of humans more prominent.
The Dilemma of "Small Data" Reasoning
AI relies on massive amounts of data for statistical learning. Once the data volume is insufficient or the data quality is flawed, its logic will fail. Even if the data volume is extremely large, if the data generation process itself is biased, flawed, or unbalanced, the value of 5 million observation data may be less than that of 30 randomly sampled data.
In reality, 96% of enterprises face data problems, and 40% of enterprises are not even confident about ensuring data quality. That's why data scientists spend almost twice as much time on data cleaning and organization as on model training, selection, and deployment.
The Shortage of Extrapolation Ability
Current mainstream machine learning technologies have an underlying assumption: the training data and the prediction results must come from the same data distribution. This leads to a problem: when it comes to predicting or generating results "beyond the scope of the training data," it becomes very unreliable. Moreover, the farther the extrapolation range is, the worse the prediction quality will be, and the robustness and credibility will also decline.
This shortcoming limits AI from replicating two core human thinking modes: Convergent Thinking (finding a single correct answer) and Divergent Thinking (dealing with open-ended questions and requiring flexible application of knowledge across scenarios). For example, in natural language processing, AI often struggles with ambiguous sentences; it's even more obvious in artistic creation. Most of the content such as poems generated by AI relies on existing works, lacking diversity and originality.
The Challenge of Multiple Solutions Coexisting
Facing problems without a "unique solution" (such as moral dilemmas and resource allocation), AI often can only output one "possible solution" and may even forcefully "determine" complex problems, thus covering up other equally reasonable options and leading to biases in selection. Taking moral dilemmas as an example, the facts of such problems are vaguely defined, and there is no consensus on the solution criteria, so they simply cannot be disassembled with AI's code logic. Even if the technical obstacles are overcome, the opacity of AI decision-making and the unclear attribution of responsibility will trigger a trust crisis, making it difficult to handle complex and dilemmatic situations.
The Problem with Interpersonal Relationships as the Outcome
In many scenarios, the end goal is not an "answer" but "human connection." This requires AI to break through the limitations of the "Theory of Mind" - understanding social signals and implicit communication, and more importantly, having real empathy. To truly establish a two-way relationship (and it can't be us forcibly "anthropomorphizing" the machine), AI also needs to have the ability of empathy and understanding.
Inability to Understand the Influence of Subjective Beliefs
People sometimes make decisions that go against the data conclusions - but this might precisely be an embodiment of breakthrough thinking. Throughout history, the decisions that truly changed the world often stemmed from the belief of challenging mainstream cognition. The driving force behind change has never been cold data but the firm belief deep in people's hearts that "the status quo must change." After all, the difference between subjective cognition and sample data is not necessarily a mistake. Sometimes, it's precisely this difference that can bring more fair and forward-looking judgments.
Human Core Competencies
EPOCH Abilities
To clarify the human abilities that can make up for AI's shortcomings, the MIT research team constructed the "EPOCH Ability Framework," which covers five major categories of core human traits. They are:
· Empathy and Emotional Intelligence: AI might be able to recognize emotions, but only humans can establish meaningful emotional connections and truly empathize with others' feelings. Professions such as social work and education are typical examples of this ability.
· Presence, Networking, and Connectedness: Professions such as nursing and journalism can reflect the importance of "being physically present" - it is the key to establishing connections, inspiring innovation, and collaborating with colleagues.
· Opinion, Judgment, and Ethics: Humans can handle open-ended fields such as law and scientific research with ease, while AI struggles to understand concepts such as responsibilities and obligations.
· Creativity and Imagination: As the researchers said, a sense of humor, improvisational ability, and "the imagination of possibilities beyond reality" are still unique human abilities. In fields such as design and scientific research, the value of these abilities is particularly prominent.
· Hope, Vision, and Leadership: Tenacity, perseverance, and initiative are the core manifestations of the human spirit. This means that even when the probability of success is slim, people will still face difficulties head-on - such as starting a new company.
Relying on this framework, the researchers used the O*NET database to analyze nearly 19,000 work tasks - focusing on the potential of these tasks in terms of automation and AI enhancement, as well as their association with human abilities.
To overcome the limitation that it's difficult to identify and classify similar tasks in different professions and positions, the research grouped all tasks into 750 "task clusters." For example, one cluster includes similar tasks related to "building a sales website" - this type of task has similar requirements in various industries; another cluster covers tasks related to "design review," involving multiple fields such as game design, sculpture, and digital imaging.
The researchers then assigned three scores to each task cluster - the score for the risk of being replaced by automation, the score for the potential of being enhanced by automation, and the EPOCH score. The EPOCH score is used to measure whether a task requires relevant EPOCH abilities.
Figure: Distribution of EPOCH scores for major occupational categories and relative rankings of sub - professions by EPOCH scores
The research found that jobs that require EPOCH abilities will become increasingly difficult to do without humans, both in terms of specific tasks and the growth of employment numbers. All task groups related to EPOCH abilities are associated with employment growth. Among them, the human ability of "sense of belief" has the greatest impact, followed by "opinion and judgment."
These findings further confirm the view that the core of an AI strategy should be to "empower workers." At the same time, it also provides a "roadmap for employee skill improvement" for corporate leaders - especially pay attention to abilities that highly reflect human traits, because these traits are easily overlooked when training employees for an "AI - driven future."
An Evergreen Skill: Storytelling
The essence of the human mind is to process and interpret information based on its unique cognition - it's not a machine for collecting facts but a machine for generating ideas. A person's mind will try to find a consistent personal narrative among all known things and then convey this idea, hoping to arouse others' resonance.
That is to say, AI can give you all the information you want, but this is not creativity. Creativity is to find unique connections among these facts and convey the results to others - and this can only be achieved through the skill of "storytelling." Moreover, "storytelling" is the concrete manifestation of EPOCH - it is both a carrier of human creativity and a way to practice empathy and judgment.
The key is that AI cannot imitate personalized stories - it doesn't have human personal experiences and minds, and cannot construct narratives starting from "issues that one cares about" like humans. The core of all stories stems from your interests and curiosity. As long as your story starts from this core, it can resist the impact of AI - generated content. No one will have exactly the same combination of themes as you, and no one will present these themes in the same way as you. As long as this remains unchanged, no large - language model can accurately convey the subtle differences you have experienced and felt.
The starting point for cultivating storytelling ability is to explore the issues you truly care about. You can go through the excerpts and notes on your bookshelf, sort out three issues that you deeply care about, and create a "problem log" - the problems in life will change, but this log can help you find your unique narrative identity. On this basis, develop a "theme perspective": when you see an idea, break it down into "problem + inspiration." Any sentence that resonates with you hides such a theme, and they are exactly the starting and ending points of a story.
AI is a "rocket ship for the mind," but it can never replace the mind itself. When the value of information approaches zero, what is truly precious is the human ability to process information, construct narratives, and establish connections - whether it's empathy and belief in the EPOCH framework or the "evergreen storytelling," they all originate from the core of human experience and are rooted in the need to "connect the world through thinking."
AI can generate information but cannot generate "meaning"; it can output answers but cannot output "connections" - and these are precisely the ultimate irreplaceable values of humans.
This article is from the WeChat official account "Sequoia Capital" (ID: Sequoiacap). The author is Hong Shan. It is published by 36Kr with authorization.