Manuel Alfonseca – Emeritus Professor at the Autonomous University of Madrid

64 years ago, the multifaceted scientist Alan Turing proposed a practical way to determine one of the most tantalising questions of our age – whether a machine can be considered as intelligent as a human being.

One common formulation of this is that a machine can be considered intelligent if it can deceive a human into believing that it is itself human.

The idea was succinct, pragmatic and influential, launching a new field of study focused on intelligent machines – a field for which, six years later, John McCarthy coined the formal name ‘Artificial Intelligence’.

Early successes – such as programs that could prove basic theorems or play games like checkers – led to rash predictions in the late 1950s. Scientists predicted that within a decade, machines would beat world chess champions and translate languages accurately.

They were wrong.

The progress was much slower and more complex than expected

The problem wasn’t ‘brain’ power – it was complexity. Chess, for instance, turned out to be vastly more complex than checkers. Similarly, human language doesn’t follow a simple, logical model – a single sentence can have multiple meanings depending on context, grammar, or broader knowledge of the world. Without such global perception and subtle appreciation, machines struggled.

However, despite considerable early setbacks, progress stumbled on. Advances in neural networks, expert systems, and genetic algorithms gradually grew the field. It nevertheless took three decades before a computer finally defeated a world chess champion.

Today, machine translation and autonomous systems have improved significantly, though they still require human oversight.

The limits of imitation

Even now, the Turing Test remains all but insurmountable for machines – it generally doesn’t take long for them to be unmasked when interacting with humans.

For instance, in one well-known case, a chatbot named Eugene Goostman pretended to be a 13-year-old boy, and convinced 33% of judges that it was human.

While impressive, this result exposed an important weakness in the Turing Test. As technology journalist Evan Ackerman pointed out:

“The problem with the Turing Test is that it doesn’t prove whether a program can think—it only shows whether it can fool humans. And humans are quite easy to fool.”

This shifts the question:
Is imitation really the same as understanding?

The Chinese Room challenge

In 1980, John Searle proposed a thought experiment to challenge the Turing Test:

Imagine a native Chinese woman conversing with a hidden computer. The computer is programmed to a degree that its responses convince the woman that she is talking to a fluent Chinese speaker.

Now imagine a non Chinese-speaking human taking the place of the computer, but following the program’s rulebook. He also holds a convincing conversation – although he does not understand a word he is saying. However, the difference is that he is aware that he didn’t understand the conversation.

So while the system passed the Turing Test, it has actually demonstrated no understanding – only an ability to follow rules.

This, Searle argued, leads to a deeper distinction:

            •Weak AI: systems that can simulate intelligent behaviour and pass the Turing Test

            •Strong AI: systems that genuinely understand and have self-awareness

An open question for the future

These debates remain unresolved. Can machines ever truly understand? Can consciousness be coded? Some, like Ray Kurzweil, believe that one day it could.

Others, including Professor Alfonseca, remain doubtful that this future will ever arrive.