Can artificial intelligence surpass human intelligence?

Artificial intelligence is one of the most disruptive technologies of the 21st century. It has been defined as the new electricity and singled out as one of the three most promising areas by Bill Gates himself. It is the key ingredient of the fourth industrial revolution and every day we wake up to the news of its impressive advances, which in many cases surpass human experience.

According to the Royal Spanish Academy, artificial intelligence is defined as the scientific discipline that deals with creating computer programs that perform operations comparable to those performed by the human mind, such as learning or logical reasoning. But can artificial intelligence surpass human intelligence?

The birth of artificial intelligence

The goal of artificial intelligence has always been to emulate human intelligence. Although it is a field that has become popular in society in recent years, its birth dates back to the middle of the last century.

In 1950, British mathematician Alan Turing published his famous paper Computer machines and intelligence, in which he described how to create intelligent machines. From this article, the Turing test was also born, which is still today the reference test for determining whether an artificial system exhibits intelligence. If a person has a conversation with a machine and another person and is unable to discern which of the two is truly a machine, the machine is said to be intelligent.

The other event linked to the birth of artificial intelligence as a discipline, and where the term was officially coined, was the Dartmouth Summer Research Project in 1956. This summer school, organized by Marvin Minksy and John McCarthy, brought together researchers who they would later be considered the founders of artificial intelligence. The goal of this project was based on the assumption that any aspect of learning or any other characteristic of intelligence can, in principle, be described with such precision that a machine can be built to simulate it.

After Dartmouth Summer School, the story of artificial intelligence has gone through a series of ups and downs. It alternated between periods of success (called springs) and periods in which funding for further field research was cut (called winters), often due to high expectations being created and not being met. For example, Marvin Minksy claimed in 1970 that within three to eight years it would be possible to develop a machine with the general intelligence of an average human being. More than 50 years have passed and this claim has not yet been fulfilled.

the new spring

Today we are witnessing the most successful period of artificial intelligence. Since the beginning of the 21st century, various factors have come together to form the perfect broth for the advancement of artificial intelligence.

The first of these is the emergence of the Big Data phenomenon, which feeds large amounts of data to artificial intelligence algorithms. The second trigger is the increase in the computational power of the machines which allows more complex experiments to be carried out. Thanks to the appearance of these factors, deep learning (or deep learning, in English). Deep learning bases its power on so-called neural networks, made up of layers and layers of processing units.

Thanks to this subfield of artificial intelligence, the greatest advances in the discipline have appeared, often related to the challenge of surpassing human intelligence or ability.

Thus, one of the great milestones in the history of artificial intelligence occurred in 2015 when AlphaGo, a program developed by Google DeepMind, managed to beat the world champion in the board game Go. Go is a Chinese game which, according to experts, it is substantially more complex than chess. Two years later, its successor AlphaZero reached a superhuman level in several board games (chess, shogi and go) in just 24 hours, also defeating previous programs and human champions.

Another field where deep learning has brought a quantum leap is natural language processing. Translation engines like Google Translate or DeepL have improved dramatically in recent years. Artificial intelligence has recent successes in specific areas such as medicine, autonomous vehicles or virtual assistants.

But if computers are so smart, why can’t they read a book?

While it may seem like AI is replacing people and surpassing their intelligence, that is not the case. We are currently in the era of “narrow” AI. This means that we have systems that can perform very complex specific tasks, but they are far from the general intelligence that people have. This fact is directly related to Moravec’s paradox, who stated in 1988: “It is relatively easy to make computers display adult-like abilities in an intelligence test or when playing checkers; and very difficult to get them to acquire the perceptual and motor skills of a one-year-old.

Later in 1994, Steven Pinker stated that “the main lesson of thirty-five years of artificial intelligence research is that hard problems are easy and easy problems are hard.”

Scientists Gary Marcus and Ernest David explain how, despite impressive recent advances in natural language processing, artificial intelligence still can’t read a book, understanding to read understand its content.

While there are powerful tools like GTB (Google Talk to Books), these are based on giving answers by imitating sentences that a person would say, but is unable to understand what they read. When people read a story, we have to follow a set of inferences that are implicit in it and that require the use of general knowledge.

The current approach of artificial intelligence is based on the representation of probabilities, trying to guess which words tend to appear in a sentence or in a context. This makes it possible to produce text or sentences that appear to be spoken by people, but does not imply domain knowledge or that a text has been understood.

To trust or not to trust, that is the question

Even in cases where artificial intelligence algorithms achieve superhuman results, all that glitters is not gold. Deep learning techniques have the peculiarity that it is very difficult to interpret how the system reached a certain decision or prediction on the data (the so-called black box model). This fact is incompatible with the need for artificial intelligence to be ethical and trustworthy, as required by the European Union’s ethical guidelines for trustworthy artificial intelligence and the Spanish national strategy for artificial intelligence.

Recently, there have been a number of cases where AI systems have behaved in a racist or sexist manner. Since intelligent systems are created by people, it seems inevitable that human biases will be transmitted to the algorithms themselves or through data.

Amazon had to withdraw its staffing algorithm after it was discovered that it was sexist and only considered male candidates. Google has apologized after discovering its image recognition algorithm was labeling a thermometer a “gun” when the hand holding it belonged to a black person.

One of the solutions to mitigate the bias that appears in AI systems and increase their trust is to have different development teams that exhaustively monitor the results obtained.

People play an irreplaceable role in the development of artificial intelligence. Not only in its creation, but also in its supervision and use. As the ethical guidelines for trusted AI postulate, intelligent systems must be people-centric, with use in the service of humanity and the common good, with the goal of increasing human well-being.

It is necessary to ensure that people are able to make informed decisions based on the results obtained by an intelligent system. Furthermore, there must always be people overseeing the operation of artificial intelligence programs to prevent fundamental rights from being violated.

Artificial intelligence and man holding hands

There is no doubt that AI is a major revolution with the potential to transform our society in ways never seen before. But we can’t forget that artificial intelligence doesn’t understand, it learns. We need to see this technology as a tool that can help us with a multitude of tasks, but which is not the enemy of people.

The dream of creating intelligence has come true. Only time will tell whether we will have generalist or conscious AI. But one thing is clear: there will be no artificial intelligence without human intelligence.

Follow us on

The Google News office

Leave a Comment

Your email address will not be published. Required fields are marked *