Where Is the Future of Artificial Intelligence? From Sci‑Fi Visions to Real‑World Challenges

Author: Wang Hui (Senior Education Product Manager at Hujiang) This is an original translation; please point out any inaccuracies and cite the source when republishing. Translated from TechCrunch, original author Vasco Pedro.

The concept of artificial intelligence (AI) has existed for a long time, but where does its future truly lie?

AI in visual media: many familiar movie images belong to AI

From HAL 9000 in *2001: A Space Odyssey* to C‑3PO in *Star Wars* and Samantha in *Her*, AI characters have populated the works of every major sci‑fi writer—from Philip K. Dick to William Gibson to Isaac Asimov. Almost every sci‑fi author has touched the theme.

Although many predictions from sci‑fi have become reality, true artificial intelligence is still far from those portrayals.

In reality, AI’s development level is far below what movies and literature have achieved. When Siri first appeared, many thought it represented AI’s future. It can answer simple questions and perform basic interactions, but over time users became disappointed because Siri, Google Assistant, Cortana, etc., are mostly used for simple searches or tasks like setting timers.

The main reason these AI programs fall short is the language problem—natural language processing (NLP). AI can grasp the literal meaning of words but does not understand context the way humans do.

If you are interested in AI, you may have heard of the Turing Test. Alan Turing was the first to take AI seriously; he proposed that if a human cannot distinguish between a machine and another human in conversation, the machine has reached human‑level intelligence.

The real Turing Test is more complex, but it remains a benchmark for NLP. In other words, if a machine can think and handle language like a human, it can be called AI.

Will we ever create a robot that can write poetry?

Think of Samantha in *Her*—an AI that smoothly understands language and can instantly incorporate new information, much like a human mind.

In *Blade Runner*, the replicant Roy Batty also demonstrates poetic language, showing a deep grasp of human emotion.

These sci‑fi AI types have existed for decades, but current technology cannot achieve them. The more we learn about true AI and NLP, the more we realize how little we understand about the human brain, which limits our ability to create machines that think like us.

We can distinguish three levels of AI, as described by Tim Urban in *Wait But Why*:

Level 1: Artificial Narrow Intelligence (ANI) – also called weak AI. It can excel at a single task (e.g., beating a chess champion) but cannot do anything else.

Level 2: Artificial General Intelligence (AGI) – also called strong AI, capable of performing any intellectual task a human can. Creating AGI is far more difficult and has not yet been achieved.

Level 3: Artificial Superintelligence (ASI) – defined by Oxford philosopher Nick Bostrom as intelligence that surpasses the best human brains in virtually every field, including scientific creativity, general wisdom, and social skills.

Human abilities we take for granted—such as rapid mental calculations for physics or geometry—remain mysterious, and we still do not know how our brains perform these unconscious tasks.

Donald Knuth once said, “AI has succeeded at tasks that require conscious thought, but it has not yet mastered the things we do without thinking.” The biggest unknown is how the brain processes language.

Many companies—Google, Palantir, startups like X.ai, MetaMind, Feedzai, SignalN, Lilt, and others—are researching ways to overcome these obstacles.

Imitating Evolution

Some firms try to replicate natural selection in AI through evolutionary algorithms (genetic algorithms). Machines are run through many trials, and successful solutions are combined. However, this approach has largely been abandoned since the 1990s because progress is slow.

Learning from Nature

Because the brain is a biological neural network, many researchers build artificial neural networks (ANNs). While ANNs are only rough mathematical models, they represent our best attempt to mimic the brain’s learning through repeated trial‑and‑error.

An amusing example: a developer fed the entire *Friends* script into a learning neural network, which then generated its own (often nonsensical) dialogue mimicking Chandler’s style.

The developer used Google’s open‑source TensorFlow library to build this script generator. TensorFlow powers many Google products (Photos, Search, Gmail, etc.) and is often called “Google’s Siri.”

Deep learning holds huge potential to push AI to the next stage, though other solutions are also being explored.

Giving Machines Their Own Language‑Learning Ability

Designing self‑learning machines that improve by studying and correcting themselves—much like Samantha in *Her*—could eventually create a new form of AI that evolves without direct human programming.

Moore’s Law—computing power doubling roughly every two years—has driven the exponential growth that enables deep learning breakthroughs, even though the rate of increase is now slowing.

Facebook’s “M” service, a machine‑learning‑based personal assistant, illustrates how AI can augment human work, though it still relies on human contractors for many tasks.

Facebook is also tackling NLP challenges, aiming to move beyond word‑level understanding to true comprehension of paragraphs and context.

Microsoft’s Skype instant‑translate feature shows another practical NLP application, providing near‑real‑time translation during calls.

AI offers massive opportunities for global business: eliminating language barriers could boost productivity, enable cross‑cultural collaboration, and open markets for small enterprises that cannot afford large translation teams.

Removing language obstacles will make the world more open, especially benefiting people from less‑developed regions.

We still have a long way to go before computers can truly understand human language, given its nuance, dialects, slang, emotion, tone, and context. While tools like TensorFlow and CNDK have made great strides, more human‑AI interaction is needed to advance further.

We will eventually have AI with such capabilities, but it may take at least 15 years. The iconic AI characters from *Her*, *2001: A Space Odyssey*, and *Star Wars* will likely be realized in some form, even if not identical to their fictional counterparts. Until then, combining human expertise with existing AI is the most effective way to leverage current technology.