Can AI replace humans? New research shares insights

Artificial Intelligence (AI) is making breakthroughs in science, medicine, and technology, but can it fully replace humans? A study by researchers from Johns Hopkins University (JHU) shows that humans still far surpass AI in understanding social interactions. Even the most advanced neural networks can’t match the human brain when it comes to analyzing short video scenes. Here’s why AI falls short, what its prospects are, and what this means for our future.

Experiment: Humans vs. Machines

The JHU team conducted an experiment comparing how humans and AI interpret social interactions in videos. Volunteers watched three-second clips showing people interacting (e.g., greeting each other), performing actions nearby (walking on a street), or acting independently (reading a book). Their task was to assess the scenes based on social cues: who was interacting with whom, their intentions, and the context.

Over 350 AI models, including language models (like GPT-4), video models (like CLIP-ViT), and graphical models, analyzed the same clips. The results were surprising:

• Humans: Almost always provided consistent interpretations, accurately identifying who was interacting and who wasn’t.
• AI: Models struggled. Video models couldn’t describe actions, graphical models had trouble distinguishing interactions, and language models, while better at predicting behavior, still lagged behind humans.
• Example: In a scene where someone waves, humans understood whether it was a greeting or a signal, while AI often misjudged the context.

The best performers were language models predicting behavior and video models assessing brain responses to scenes, but none came close to human accuracy. “AI doesn’t yet understand what’s happening in real life,” concluded lead author Leyla Isik.

Why Does AI Lag Behind?

The issue lies in the architecture of neural networks. Modern models, like xAI’s Grok or ChatGPT, are inspired by the brain’s visual cortex, which processes static images. This enables AI to excel at recognizing faces, objects, and text, but social scenes involve dynamics that require understanding time, intentions, and context. In the human brain, the temporal lobe handles this, an area not yet replicated in AI.

“Real life isn’t a set of photos. It’s movement, emotions, and interactions unfolding over time,” explained co-author Katie Garcia. For instance, AI might not grasp whether a pedestrian is about to cross the road or just chatting with a friend. This is critical for technologies like autonomous vehicles, where errors can be deadly.

Where Is AI Already Strong?

Despite its shortcomings in social scenarios, AI outperforms humans in other areas:

• Data Analysis: In Xiong’an, AI processes terabytes of climate data, as seen in the Nature Communications study on temperature swings.
• Creativity: xAI’s Grok Vision analyzes images via iPhone cameras, suggesting recipes or translating signs.
• Science: AI aided in modeling the Krebs cycle in space (PNAS) and analyzing lunar soil from Chang’e-6.

However, in tasks requiring intuition, empathy, or nuanced understanding, humans remain unmatched. For example, Xiaomi Mijia S800 smart scales provide health recommendations, but humans must interpret and act on them.

Can AI Catch Up to Humans?

Scientists believe AI can improve, but it requires changes:

• New Architecture: Models must mimic brain regions responsible for dynamics and social cues, not just static vision.
• Video Training: Current networks train on vast text and image datasets, but video data is harder to collect and process.
• Context: AI needs to learn cultural and emotional nuances, as humans do.

Leyla Isik emphasizes: “AI has leaped forward in face and object recognition. The next step is teaching it to see the world as a story, not slides.” X users agree: “AI is cool, but without the human spark, it’s just a calculator.”

What Does This Mean for Us?

The JHU study shows that replacing humans with AI is currently impossible, especially in fields requiring social sensitivity, like medicine, psychology, or management. However, it doesn’t rule out collaboration:

• Work: AI automates routine tasks, as in NTT’s lightning-catching drones, while humans tackle creative challenges.
• Safety: For autonomous vehicles like Tesla, hybrid systems combining AI and human input are essential.
• Ethics: The Grok scandal, where it suggested “executing” Trump, highlights the dangers of AI without human oversight.

In the future, AI may approach human capabilities, but for now, it’s a powerful tool, not a replacement. It’s like the iPhone 17 Air: sleek and smart, but just a piece of metal without a user.