Artificial General Intelligence: Technology, Society, and the Question of What Comes Next

Author’s note: This article began as a short essay written in April 2025 for the AI Methods course at the University of Turku. The original assignment was to reflect on artificial general intelligence, its relationship to existing AI technologies, and whether AGI might eventually be possible. I have since expanded, revised, and updated the original essay for publication here. It is also the first article on this blog.

Artificial intelligence has become one of the defining technological themes of recent years. Although AI has been under development for decades, it remained largely within academia and the technology industry for much of its history. That changed dramatically in 2022, when OpenAI released ChatGPT. Suddenly, artificial intelligence was no longer something discussed mainly by researchers, engineers, and technology companies. It became a technology that millions of ordinary people could interact with directly.

This shift has also brought a much larger question into public discussion: what happens if artificial intelligence becomes increasingly general?

In this article, I explore what artificial general intelligence, or AGI, means, how it differs from today’s AI systems, and what kinds of technological, ethical, and societal questions it raises. I also consider whether AGI may be possible in the future and what current developments suggest about the direction of artificial intelligence.

From Narrow AI to General Intelligence

Current systems such as ChatGPT are generally not considered artificial general intelligence. They are often associated with narrow or specialized AI, although modern large language models can perform a much broader range of tasks than traditional narrow AI systems. Despite their impressive capabilities, there is currently no evidence that systems such as ChatGPT possess consciousness comparable to that of humans.

Artificial general intelligence refers to something more ambitious. Although there is no universally accepted definition of AGI, the term generally refers to an artificial system capable of performing a wide range of cognitive tasks at a human level or beyond. Such a system would need to transfer knowledge between different domains, adapt to unfamiliar situations, learn efficiently, and apply what it has learned in new contexts.

Importantly, AGI does not necessarily require consciousness. Intelligence and consciousness are related concepts, but they are not the same thing. A system could theoretically demonstrate highly general and flexible intelligence without having subjective experiences or human-like self-awareness.

This distinction is important because discussions about AGI often combine several different questions. Can machines reason across many domains? Can they learn as flexibly as humans? Can they become conscious? These are separate questions, and science does not yet have complete answers to them.

Why Current AI Is Not Yet AGI

Modern AI systems are primarily based on machine learning methods, particularly deep learning. They can recognize images, generate text, translate languages, write computer code, analyze data, and perform many other tasks that would once have seemed extremely difficult for machines.

Their capabilities are remarkable, but they also have clear limitations.

Current systems can fail unpredictably in unfamiliar situations, generate incorrect information with confidence, and struggle with tasks that require consistently reliable reasoning over long periods. Their ability to generalize beyond familiar patterns and situations is still imperfect.

There is also an ongoing scientific and philosophical debate about whether modern neural networks genuinely understand the information they process. They can produce outputs that strongly resemble human reasoning, but determining whether this represents understanding in the same sense that humans understand concepts and meanings is much more difficult.

For this reason, current AI systems have not yet demonstrated the level of robust reasoning, adaptability, and generalization usually associated with AGI.

Are Large Language Models the Path to AGI?

Large language models based on transformer architectures have become one of the most important approaches in modern artificial intelligence. Their rapid progress has led some researchers and technology companies to argue that increasingly capable models could eventually contribute to the development of AGI.

However, this is far from certain.

There is no scientific consensus that simply building larger language models will produce general intelligence. Future systems may require new architectures, different forms of learning, stronger reasoning mechanisms, better memory systems, interaction with the physical world, or combinations of several different approaches.

In other words, current language models may represent part of the path toward AGI, but we do not yet know whether they represent the path itself.

This uncertainty is one of the aspects of AI research that I find particularly interesting. We are seeing systems become increasingly capable, yet even researchers working at the frontier disagree about what ingredients are ultimately necessary for general intelligence.

Could We Copy the Human Brain?

Another proposed approach is whole brain emulation, which I find very interesting.

Instead of trying to design intelligence entirely through artificial neural networks and machine learning algorithms, whole brain emulation would attempt to reproduce the structure and functioning of a biological brain computationally.

In theory, a human brain could be mapped at extremely high resolution and simulated digitally so that the resulting system reproduced relevant aspects of the original brain’s behavior and information processing.

Whole brain emulation can be viewed as an extreme form of biomimetic or brain-inspired engineering. Biological brains have long served as an inspiration for computational systems, from artificial neural networks to neuromorphic computing and newer approaches involving biological neural systems (Brown & Varghese, 2026).

Unlike most biomimetic approaches, however, which imitate selected principles found in biological systems, whole brain emulation would aim to reproduce the functioning of a particular brain at a much more detailed level.

Rather than attempting to invent general intelligence entirely from scratch, the underlying idea is compelling: biological brains already provide an example of a system capable of general intelligence. If the relevant structures and processes could eventually be understood and reproduced computationally, they might offer an alternative route toward machine intelligence.

The challenge is enormous.

The human brain contains tens of billions of neurons and a vastly greater number of connections between them. Researchers have already demonstrated brain simulations at remarkable scales. For example, Lu et al. (2024) presented a Digital Brain platform capable of simulating spiking neural networks with up to 86 billion neurons. Their work demonstrates how computational models are beginning to operate at the neuronal scale of the human brain.

However, numerical scale alone should not be confused with whole brain emulation. Simulating a network containing roughly as many neurons as a human brain does not mean that the biological structure, connectivity, chemistry, learning processes, memories, or consciousness of an individual brain have been reproduced. We still lack much of the biological knowledge and measurement capability that such an emulation would require.

Current technology therefore cannot perform a complete whole brain emulation.

Nevertheless, advances in neuroscience, brain imaging, computational modeling, and computing continue to make increasingly detailed simulations possible. Futurists such as Ray Kurzweil have argued that sufficiently advanced technology could eventually allow the human brain and its processes to be modeled in far greater detail, although predictions about when or whether this could lead to true whole brain emulation remain highly speculative (Kurzweil, 2005).

AGI Is Not Only a Technical Problem

The possibility of increasingly general artificial intelligence raises questions that go far beyond computer science.

Developing systems with increasingly broad capabilities requires advances in computing power, algorithms, data, neuroscience, cognitive science, and other fields.

At the same time, the more capable these systems become, the more important questions of safety, transparency, regulation, and control become.

The central challenge is not simply whether we can build more powerful artificial intelligence. We also need to ask what happens when we do.

Who controls these systems?

Who benefits from them?

Who is responsible when they fail?

And how should society decide where highly capable AI systems are allowed to operate?

These questions become particularly important when AI systems are used in areas such as healthcare, finance, public administration, military technology, education, or law enforcement.

The Black Box Problem

One major challenge in modern artificial intelligence is the so-called black box problem.

Large neural networks can contain enormous numbers of parameters. Even when researchers understand the architecture and training process of a model, it can still be extremely difficult to explain exactly why the model produced a particular output.

This does not mean that an incorrect AI system can never be corrected. Models can be tested, retrained, restricted, or redesigned even when their internal processes are not fully interpretable.

However, limited explainability creates serious challenges.

If an AI system makes an important decision affecting a person’s life, we may need to understand why that decision was made. Without sufficient transparency, identifying systematic errors, auditing decisions, assigning responsibility, and ensuring fairness become much more difficult.

For this reason, interpretability and explainable AI have become important areas of research.

What About the Technological Singularity?

One of the most dramatic ideas associated with AGI is the technological singularity.

The term generally refers to a hypothetical point at which technological development accelerates so rapidly that its future course becomes extremely difficult or impossible for humans to predict. In some versions of the idea, an advanced AI system could contribute to improving its own capabilities, potentially leading to increasingly rapid technological progress.

Ray Kurzweil is among the best-known proponents of the idea that technological development may eventually reach such a transformative point (Kurzweil, 2005).

The possibility nevertheless remains highly speculative, and there is no scientific consensus that a technological singularity will occur.

Still, the idea raises an important underlying question.

What happens if artificial intelligence helps technological development progress faster than human institutions can adapt to it?

Even without a dramatic singularity, AI development may move faster than legislation, education systems, labor markets, and political institutions are able to respond. From this perspective, the issue is not only about hypothetical superintelligence. It is also about society’s ability to manage rapid technological change.

Different Views on the Future of AI

Leading figures in artificial intelligence have expressed different views about the future of AI and the possibility of general intelligence.

In an address at the Oxford Union, Kai-Fu Lee discussed the limitations of contemporary artificial intelligence and the distinction between current AI capabilities and much more ambitious forms of machine intelligence (OxfordUnion, 2018). Lee argued that reaching genuinely general AI would require significant advances beyond the capabilities of narrow AI.

Geoffrey Hinton, one of the pioneers of modern neural networks, has expressed stronger concerns about the pace of AI development.

Hinton left Google in 2023 and subsequently spoke publicly about the possible risks of increasingly capable artificial intelligence. In an interview with MIT Technology Review, he described how his own assessment of AI had changed as systems became increasingly capable and warned about the difficulty of predicting where rapid advances might lead (Heaven, 2023).

Hinton has raised concerns about several possible consequences of increasingly powerful AI. These include the use of AI for harmful purposes, large-scale misinformation, disruption of labor markets, and the longer-term difficulty of controlling systems that become more capable than their designers originally expected (Heaven, 2023).

These concerns do not necessarily imply that AI development should stop. Instead, they demonstrate why safety research, governance, and technological development should progress together.

How Do People View Artificial Intelligence?

Public attitudes toward artificial intelligence are more complex than simply being either positive or negative.

According to the Finnish Science Barometer 2024, 51 percent of respondents had a positive attitude toward the increasing use of artificial intelligence in society, while 39 percent viewed it negatively (Tieteen tiedotus ry, 2024).

The results also show that attitudes depend strongly on how AI is used.

For example, facial recognition for crime prevention and border control received some of the most positive assessments. Automatic weapons and the use of AI in journalism and employee selection received some of the most negative assessments (Tieteen tiedotus ry, 2024).

This distinction is important.

People do not necessarily support or oppose artificial intelligence as a technology. Instead, they evaluate specific applications based on questions such as usefulness, safety, fairness, and trust.

That may ultimately be one of the most important lessons in the broader AI debate.

The question is not simply whether artificial intelligence is good or bad. A more meaningful question is how it is designed, who controls it, how it is used, and what consequences it creates.

Where Does This Leave Us?

Artificial intelligence is not inherently ethical or unethical. Its effects depend on the systems we build around it, including the people who develop it, the organizations that deploy it, the incentives behind its use, and the rules that govern it.

This also means that responsibility cannot be placed entirely on individual developers.

Companies, governments, researchers, users, and society as a whole all influence how artificial intelligence develops and what role it eventually plays.

AGI, if it is ever achieved, would therefore be much more than a technological milestone. It could become an economic, political, philosophical, and societal turning point.

At the moment, there is no scientific consensus on whether AGI will be achieved, what form it would take, or when it might emerge.

Perhaps that uncertainty is exactly why the topic is so fascinating.

We are developing increasingly capable artificial systems while still trying to understand what intelligence itself really is. At the same time, we are being forced to consider questions about control, responsibility, consciousness, work, inequality, and human agency.

For me, this is what makes AGI particularly interesting. The question is not only whether we can eventually build a machine with general intelligence. The pursuit of AGI also forces us to examine what we mean by intelligence in the first place, what responsibilities come with increasingly capable technology, and how quickly society can adapt when technological development moves faster than the institutions surrounding it.

The development of artificial intelligence will therefore require more than better algorithms and more computing power.

It will also require research, regulation, public discussion, and serious consideration of what kind of relationship we want to have with the intelligent systems we create.

References