Super Artificial Intelligence: The Greatest Threat to Humanity or Hope for a Better Future?
Have you ever imagined a day when artificial intelligence would surpass human intelligence? Can a machine think, learn, and innovate in a way that surpasses the ability of the human mind? This question is not science fiction, but rather an urgent question that arises in light of the huge progress we are witnessing in this field, as leading experts in the field of artificial intelligence today confirm that we will reach this stage soon.
Masayoshi Son confirmedThe CEO of SoftBank Group, the technology investment giant, said during his participation in the Future Investment Initiative conference held in Riyadh last week that super-intelligent artificial intelligence will be 10,000 times smarter than the human brain and will exist by 2035.
As OpenAI CEO Sam Altman said last month: “It’s just a matter of time before we have super-intelligence. It could be sooner than we think, it could be in a few thousand days, it could be a little longer. But no matter how long it takes, this technology will fundamentally change the face of the world.”
OpenAI’s chief scientist and co-founder, Ilya Sutskever, created a team within the company a year ago to focus on developing “safe superintelligence,” but in June he announced the formation of a new startup called Safe Superintelligence (SSI) to pursue that goal, just a month after he left OpenAI.
But what exactly is super-intelligence they’re talking about? Does it represent a golden opportunity for breakthroughs in various fields, as experts have been claiming for some time, or does it represent an existential threat to humanity, as Nick Bostrom, a philosopher and AI expert at Oxford University, warned a decade ago?
First, what is super AI?
The term Artificial Superintelligence (ASI) refers to any intelligent system that surpasses human intelligence in all aspects, including: creativity, judgment, and the ability to learn and adapt.
This intelligence can take many forms, from systems capable of self-learning and innovation, to machines that possess self-awareness and the ability to understand the world around them more deeply than humans.
But this definition prompts us to ask about the types of artificial intelligence and what type have we reached so far?
Second, what are the types of artificial intelligence?
Among the many classifications of types of artificial intelligence, the framework developed by American computer scientist Meredith Ringel Morris and her colleagues at Google is the most clear and profound, as they presented this framework in a research paper last year.
This framework provides a precise classification of AI capabilities, dividing them into six levels ranging from simple to highly advanced systems.
The classification starts from the level of (no AI), which represents simple machines such as calculators, which perform various mathematical tasks according to previous programming, and reaches the level of (superhuman), which represents systems with capabilities that exceed human capabilities. This classification helps us understand the development of artificial intelligence and determine our current position in it.
Here are the stages of artificial intelligence:
1- Narrow-band artificial intelligence (NAI):
Narrow AI (NAI) is a specialized, highly skilled AI designed to perform specific tasks with high accuracy, and does not exceed the limits of these tasks. This type of AI learns from huge amounts of data, but this learning is limited to the narrow domain for which it was designed.
This type of intelligence is the driving force behind many of the applications we use every day, such as search engines, machine translation, image recognition, text analysis, and even autonomous driving. This type of intelligence is characterized by its ability to perform routine tasks with high accuracy and super speed, but it may have difficulty dealing with unexpected situations or situations outside the scope of its training.
To illustrate this classification, Morris gives a practical example: the chess program Deep Blue , which achieved a historic victory over world champion Garry Kasparov in 1997. It represents an ideal model of a (narrow-scale) artificial intelligence system at a mastery level, as it works very efficiently within a specific domain, which is chess, but is unable to perform other tasks.
There are also some narrow-range artificial intelligence models that have capabilities that exceed those of humans currently, including the Alphafold model , developed by Google, which revolutionized the field of predicting protein structures, and whose creators won the Nobel Prize in Chemistry this year .
2- Artificial General Intelligence (AGI):
Artificial general intelligence (AGI) refers to a powerful form of AI that is capable of performing a wide range of tasks, including things like learning new skills. Experts say that AGI is the stage where a machine can perform any intellectual task that a human can.
The idea of an intelligent system capable of performing any intellectual task that a human would perform sounds exciting, but the reality is somewhat different, as we are still far from achieving general artificial intelligence in the full sense yet.
According to Morris, large language models like Gemini and ChatGPT represent a first step toward achieving general artificial intelligence, but these models are still in an early stage of development and are considered at the emerging level compared to human intelligence, which means that they are capable of performing specific tasks well, but they lack the flexibility and critical thinking that humans have.
This is confirmed by Yann LeCun , a French computer scientist and one of the three godfathers of artificial intelligence, who is now the chief AI scientist at Meta. He believes that large language models are the key to general artificial intelligence, which reaches human-level thinking and creativity, but reaching it may take 10 years or decades, as there is no specific timeline for this development yet.
“To achieve what is called artificial general intelligence, we need machines that understand the world; machines that can remember, reason, and plan at a human-level,” Yann LeCun said at the Hudson Forum last month. “So far, the current systems we have come up with all lack these basic capabilities.”
3- Artificial Super Intelligence (ASI):
When we reach the second type, which is artificial general intelligence (AGI), we will move towards the third and final type, which is artificial super intelligence (ASI), which exceeds the capabilities of humans. This means that we are still very, very far from reaching this third type; because we have not yet reached the second type, which experts indicate will take at least 10 years to reach. But this raises an important question: Where has artificial intelligence reached now?
Third, where has artificial intelligence reached now?
Determining the level of intelligence of current AI systems is a complex and contested issue, and as Morris points out, this comparison requires accurate and reliable standards and metrics.
Depending on our criteria, an image-generating system like DALL-E might be at the genius level, because it is capable of producing complex artistic images that are beyond the capabilities of most humans, or it might be at the emergent level, because it produces errors that no human could produce, such as depicting people's hands in unnatural shapes.
There is much debate about the capabilities of current systems, with a 2023 research study claiming that the GPT-4 model showed “signs of advanced general intelligence,” meaning the ability to think and learn in a way similar to humans.
OpenAI also claims that its latest model, the O1 , which it released in September, can perform complex reasoning, rivaling the performance of human experts on many benchmarks. However, a recent study by Apple researchers suggests that these claims may be exaggerated. While the O1 model is capable of producing coherent and fluent text, it struggles to solve mathematical problems that require abstract reasoning.
This suggests that the model can mimic human thinking in some ways, but it still relies heavily on matching patterns in the data it was trained on. This raises questions about whether current models can achieve a level of intelligence comparable to real human intelligence.
Let us assume that we have reached the stage of super artificial intelligence. How will it change the face of the world, and what are the expected risks at this stage?
Fourth: Super Artificial Intelligence: The Greatest Threat to Humanity or Hope for a Better Future?
Experts confirm that the transition to the stage of super artificial intelligence will be a pivotal event in human history, as the capabilities of machines will exceed human capabilities in all fields, and then this transformation will lead to radical changes in all aspects of our lives, starting from the economy, politics and medicine to culture and society.
Expected positive effects:
- Amazing medical scientific developments: Super AI can accelerate the discovery of drugs and treatments for incurable diseases, and develop new technologies to extend lifespan and improve quality of life.
- Solving global problems: It can be used to solve major challenges such as climate change, poverty, and hunger, by developing innovative and effective solutions.
- Increased productivity and innovation: This will lead to a massive increase in productivity and innovation across all industries, creating new jobs and improving the standard of living.
- New Scientific Discoveries: Super AI could unlock the secrets of the universe, expand our understanding of reality, and open up new frontiers in science.
Potential risks:
Experts such as Nick Bostrom, a philosopher and AI expert at Oxford University, warn in his book Superintelligence: Paths, Dangers, Strategies that superintelligence could pose an existential threat to humanity. If AI becomes smarter than humans, it could become uncontrollable and make decisions that are not in the best interests of humans. Potential risks include:
- Loss of control: Superintelligent AI will become able to manipulate systems and data in ways that humans cannot understand or control.
- Inequality: Few may benefit from the fruits of AI, leading to greater inequality in wealth and power.
- National security threat: AI could be exploited for malicious purposes, such as developing self-driving weapons or launching large-scale cyber attacks.
- Loss of control: AI may go beyond human control and make decisions that could harm humanity.
Conclusion:
Artificial intelligence has made great progress in recent years, but it is still far from achieving general artificial intelligence, which is comparable to human intelligence, as the capabilities of current systems are still limited to specific fields, and face great challenges in solving problems that require abstract and deep thinking.
So the dream of reaching the stage of artificial super intelligence (ASI) is still very far away, as some experts believe that it may take centuries or may never happen.