Despite the impressive progress of artificial intelligence, AGI (artificial general intelligence), performing mental work on a human level, is still just fiction. Even GPT-4 and PaLM, which handle a multitude of intellectual tasks, don't operate on this level.

Yet, scientific successes in neural networks allow us to hope that AGI will appear soon. According to the most optimistic forecasts, it will happen in a few years when one of the next-generation language models will be released.

Let's find out what AGI is and how it will impact our society.

How AGI Differs From Conventional AI

The programs based on artificial intelligence that we use now are highly specialized. For example, AI designed to play chess cannot generate images or diagnose diseases.

So, scientists refer all existing AI variants to the class of narrow AI (ANI), sometimes called weak AI. Such artificial intelligence can solve individual tasks in narrow subject areas under human control.

It's expected that the next stage of AI development will be the creation of a universal system that will be able to solve any tasks under the control of the human mind. This neural network will belong to the class of general (AGI, artificial general intelligence) or strong AI.

The key similarity between general artificial intelligence and humans is their ability to think and act independently. Every person does a huge amount of intellectual work daily: driving a car, planning finances, talking to others, writing, reading, playing, and so on. It's assumed that a general AI will be able to learn to do this. According to some scientists, AGI should have intrinsic motivation, the ability to set goals and divide complex tasks into smaller subtasks, and other skills that humans have.

What Are the Capabilities of AGI

No one knows exactly what AGI will be capable of. After all, machine intelligence is strikingly different from human intelligence. It's almost like trying to imagine what dolphins are thinking, or vice versa.

However, we can assume that AGI will be able to:

• Learn and adaptively apply what it has learned across a wide range of tasks and domains in unknown situations. Adapt to a changing environment and learn from new data and experiences. Independently expand their skills and knowledge without requiring constant intervention from a developer.
• Understand the context and connections between different ideas and concepts, grasp complex relationships, and synthesize information to make intelligent decisions.
• Possess some level of self-awareness and awareness of itself as an entity. This includes awareness of its condition, skills, and limitations, as well as awareness of its purpose or goal.
• Show elements of creative thinking and generating new ideas. AGI will be able to offer original solutions and approaches to tasks, not limited to particular algorithms and templates.

The creation of AGI will ease our daily lives. AGI is a useful thing, but it won't eliminate the need for us to make independent decisions about our own and more global issues.

The Future of Language Models and the Switch to AGI

AGI requires the development and integration of various technologies. Besides language processing, machine vision, robotics, automatic planning, and decision-making. Improvements in hardware and the development of new approaches to programming and the architecture of artificial neural networks are also required.

Although large language models are essential for artificial intelligence, they aren't the only solution to getting closer to AGI. Co-development and integration of different technologies and approaches are needed to create a complete general artificial intelligence. LLMs will greatly accelerate this progress. It isn't AGI yet, but it's already a huge leap towards its creation.

Overall, we are still not close to AGI. However, with the release of GPT-4, there is an understanding that the emergence of general artificial intelligence is only a matter of time. If we start with the report on GPT-4 research by Microsoft specialists, the rudiments of reason and common sense are already being recorded in language models. Truly significant processes are taking place, and our society will become absolutely unrecognizable in 10-15 years.