Roadmap towards AGI

As much as AI is already changing to great extent many different industries, there is still a (possible) long road ahead towards achieving achieving a first resemblance of an AGI, however, the signs are here.

What is an AGI

AGI, or Artificial General Intelligence, refers to a type of artificial intelligence that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks and domains at a level comparable to human intelligence. In contrast, most traditional or narrow AI systems are designed for specific tasks and lack the versatility and adaptability of AGI.

Is ChatGPT an AGI?

Taking the example of the well known Language Models such as GPT, Lamma, Orca, etc and other media types of AI tools (video, image and audio editing), these are very narrow focused and operate strictly on top of an immense data pool of patterns.

Albeit world changing, this means that they lack fundamental abilities of an AGI such as:

Learning skills

Although these models are well equipped to fine tune their answers based on input and more data, this is not the same as having a learning capability. Fine-tuning is applied to solve a specific purpose and not to expand the general skill set of the AI.

In Language Models for example, we can train them extensively to become better at understanding what our words mean, to become faster and better at answering them, but no amount of fine tuning applied to a Language Model will move it closer to being able to actually submit a tax report on your behalf.

Adaptability

AGIs should be able to adapt to new and unforeseen situations, exhibiting a level of flexibility and problem-solving ability comparable to human intelligence. Regular AI systems are usually constrained by their predefined programming and may struggle when faced with tasks outside their designated scope.

Today, AIs still can't break from within the walls they were designed to operate on, neither can they use existing "knowledge" (which is just data at this point) and apply it to completely different scopes.

Generalization

The combination adapting to unknown circumstances and learn new tricks with it is the path to making an AI into a fully functional (and scary) AGI. With both these skills, an AI can technically become generic as it will have the capabilities literally evolve by trial and error, however, at an unprecedented speed compared to us, mere mortals.

Is it possible to achieve an AGI?

Our benchmark for AGI is, for a lack of better reference, our own brain and how it operates. As such, tremendous efforts have been placed to understand how our own brains function with the hope of unlocking the secret to building AGIs. But is this possible or even necessary?

One of the best references out there for the "if" and the "how" on AGIs is the amazing paper from William J. Rapaport on Landgrebe and B. Smith's "Why Machines Will Never Rule the World: Artificial Intelligence without Fear", entitled Is Artificial General Intelligence Impossible?

The writers question the possibility of successfully modeling complex systems in the world, particularly in relation to the human brain. They raise concerns about the current inability to model complex systems mathematically and emphasize the challenges in generating sufficient amounts of energy for small motors, indicating limitations in our current knowledge and methods for modeling complex systems mathematically.

Additionally, they argue that the emulation of the human brain is not the goal of AI, and there may be other ways to achieve the goal of developing a computational theory of human cognition. They suggest that the focus on emulating the whole brain may not be the most effective approach and that there are different perspectives on how to emulate desired parts of the brain's functioning, indicating a divergence in approaches within the field of AI research.

In sum, we may not need to model the human brain to achieve AGI, and trying to do so may distract us from the purpose into an unattainable search for perfection.

Final thoughts

What seemed science fiction for most of us is slowly becoming a possibility given the recent breakthroughs with AI. The hype is pushing more and more funds and research into this field which is definitely going to speed up the process.

Having said this, the road do AGIs is pretty much in its early stages. Although there is tremendous promise, there are many questions to answer as to if this will even be possible and how.

Can we expect to one day have a massive AGI modeled after the human brain; a set of smaller niche AI's that learn to cooperate between themselves that as a whole are flexible enough to be considered an AGI; or something in between? Will the industry culminate with the long feared singularity, or eventually burn through the hype, peak, and slowly descend down into "last years news" for a few years, again?

Time will tell.