Artificial superintelligence: will machines surpass humans?

Artificial superintelligence is a theoretical idea that, if materialised, would bring changes comparable to historic transformations such as the Industrial or Digital Revolution. According to Swedish philosopher Nick Bostrom, author of the book Superintelligence: ‘The development of superintelligence would be humanity's last great invention. After that, machines would take care of the rest.’ 

Cinema and literature have imagined scenarios where AI becomes uncontrollable, such as in the film Terminator (1984), where an AI threatens humanity. We are still far from this example, but fiction invites us to reflect on the impact of technology on our society. To talk about artificial superintelligence, we must start at the beginning: artificial intelligence (AI).

What is artificial intelligence?

AI is the ability of machines to perform tasks that require human intelligence. It ranges from simple tasks, like a search query, to complex systems such as virtual assistants capable of maintaining fluid conversations. 

Beyond mimicking human intelligence, IA develops new ways of thinking and decision-making based on data logic. For example, deep learning systems detect large volumes of data that humans may not perceive. This advancement raises the possibility that AI will not only imitate but also complement areas where human perception and analysis are limited. 

However, not all AI systems function the same way. Let’s explore the different types and their distinctions. 

Types of AI and their potential

1. Weak or narrow AI: this technology powers tools many of us use daily, such as customer service chatbots, facial recognition on smartphones, or voice assistants like Siri and Alexa. It is designed to perform specific, repetitive tasks faster than humans. For example, data processing or executing commands. 

2. General AI: this is the next step in AI evolution. Unlike narrow AI, it can understand, learn, and perform any task as humans do. Although still in development, it is considered a revolutionary milestone. 

3. Artificial Superintelligence (ASI): If developed, it would surpass human intellect. Though currently just a concept, artificial superintelligence would be a machine capable of thinking and reasoning. It might even develop emotions! 

This leads us to ask, ‘What are the specific characteristics of artificial superintelligence?’ 

What is artificial superintelligence?

It is a developing theory, but it is expected to exceed human capabilities in complex areas such as creativity, problem-solving, and decision-making. 

The concept was popularised by Swedish philosopher Nick Bostrom in his book Superintelligence: Paths, Dangers, Strategies. Bostrom argues that once a machine achieves general intelligence (AGI), it could improve itself and evolve into a superintelligence beyond any human capacity. 

Its arrival would redefine our understanding of technology. But how does it differ from AI? 

Key differences between AI and artificial superintelligence

At first glance, ASI may seem like just an advanced version of AI, but it goes far beyond that. 

Autonomy and adaptability

Traditional AI requires data and programming to function. For example, a facial recognition system cannot adapt to changes unless it is retrained. Superintelligence, however, would learn independently and adapt to new environments or challenges without human intervention. 

Cognitive ability

Superintelligence would be capable of understanding, reasoning, and solving problems across all domains. 

Impact on society

AI already influences sectors such as healthcare, transportation, and energy. Still, the impact of superintelligence would be unimaginable, as it would have the ability to learn on its own. We have even seen this concept in sci-fi films like Transcendence (2014), which explores a future where AI acquires infinite processing and learning capacity.

Though currently just a concept, artificial superintelligence would be a machine capable of thinking and reasoning. It might even develop emotions!

This current fictional scenario opens an ethical debate that leads us to ask, ‘What are the risks of ASI?’ 

Challenges and dangers of artificial superintelligence

One of the biggest fears surrounding ASI is that once a machine becomes more intelligent than humans, we might lose control over it. ASI could make decisions independently, while its objectives might not align with ours. 

Elon Musk, CEO of Tesla and SpaceX, has warned about this risk, suggesting that AI poses an ‘existential threat.’ In an interview, Musk expressed concern that, without proper regulation, machines could make decisions contrary to human interests. 

Another major concern is inequality. If only a few governments had access to this technology, they could control large sectors. 

Decision-making is another crucial factor. Would a machine be fairer and more ethical than a human, or would it rely solely on logic without considering values? 

However, it’s not all about risks. We must also consider the potential benefits of artificial superintelligence. 

Potential benefits of artificial superintelligence

One of the greatest benefits of ASI would be its ability to tackle global issues such as climate change, poverty, and disease. It could analyse massive amounts of data to find innovative and rapid solutions, allowing humanity to overcome challenges that have so far been insurmountable. 

In medicine, biotechnology, and physics, ASI would be revolutionary. It could develop new medical treatments, design stronger materials, and contribute to the creation of groundbreaking technologies. This would thus accelerate human progress to unprecedented levels. But when could these advancements arrive? 

Are we close to achieving artificial superintelligence? 
 

Although ASI is still a distant concept, some AI experts argue that we are closer than we think. Tech companies such as OpenAI and DeepMind are pushing the limits of AI development. In fact, OpenAI believes it could arrive in less than ten years. 

Recent advancements—such as the development of language models like GPT-4—demonstrate remarkable human-like language processing capabilities, representing a step toward general intelligence.

However, achieving true superintelligence would require breakthroughs in how AI acquires knowledge and develops metacognitive skills—that is, its ability to learn how to learn, adapt, and self-regulate autonomously. 

Yann LeCun, Meta’s AI director, believes we are still far from creating systems that surpass human capabilities. According to him, progress toward superior intelligence will be gradual, but there are currently no models that can achieve this level. 

Still, we have seen that general AI is an intermediate step before reaching superintelligence. While its materialisation may still be far off, we must prepare to manage the impact of this major technological breakthrough on humanity.