The 3 Phases of AI: From Narrow to General to Super Intelligence

The Superpower of AI for Businesses 

ChatGPT already had over 100 million users in its first six months. With LLMs, businesses can use them to generate marketing content, brainstorm ideas, analyze data, generate and review source code, among other tasks. Integrating OpenAI technologies into your own Azure environment can enable the development of specialized chatbots based on your data, further automate ticket processing, and clean up raw data. 

Have we achieved AGI with LLMs, competing with human intelligence? More on that later. Let's first review the history and different phases of AI. 

Phase 1: Artificial Narrow Intelligence (ANI) 

Until the 1980s, AI systems were based on manually programmed rules. They excelled at one specific task but had limited adaptability. In the 1990s, machine learning (ML) emerged, allowing algorithms to recognize and use patterns in data. In a sense, ML is self-learning: as the system processes more data, the algorithm becomes more refined. These systems are typically specialized in one area and can solve one problem, hence the term "artificial narrow intelligence." A classic example is the chess computer Deep Blue. More recent examples include computer vision for tasks like license plate recognition in parking garages and natural language translation models (e.g., Google Translate). 

Phase 2: Artificial General Intelligence (AGI) 

While machine learning was the distinguishing feature of the previous phase, this phase is characterized by machine intelligence. That means one AI model is so 'intelligent' that it is as clever as a human in all respects. An AGI system possesses human-like cognition but can perform tasks much faster and (in principle) without errors compared to humans. 

Phase 3: Artificial Super Intelligence (ASI) 

Machine superintelligence may sound a bit unsettling to some of us. What do we mean by that? Systems that have more intelligence than humans and surpass the best human specialists in almost every domain. An ASI system could solve problems we currently consider unsolvable, make significant scientific discoveries, create art as impressive as human art, and have a deep understanding of human emotions. 

'The Singularity Is Near' 

Artificial superintelligence is often associated with the singularity. This is a (hypothetical?) moment in the future when technological progress becomes so rapid and disruptive that it leads to a radical and entirely unpredictable transformation of human civilization. The assumption behind this is that if we can build a system smarter than humans, this system can also improve itself, leading to an exponential explosion of knowledge. 

The consequences of this? No one can even begin to predict. And when it will happen? Nobody knows. The most optimistic view comes from entrepreneur and inventor Ray Kurzweil, who wrote the book 'The Singularity Is Near' in 2005 and is now working on the sequel 'The Singularity Is Nearer.' He expects that around 2035, as reported by The New York Times, 'computation will be part of ourselves, and we will increase our intelligence a millionfold.' 

Now: Transition to Artificial General Intelligence 

According to Cédric Vandelaer, data science specialist and AI expert at Cegeka, we are currently in a transition period and seem to be much closer to AGI. "The revolutionary aspect of models like ChatGPT is that through textual input, you can perform a wide range of tasks such as summarization, sentiment analysis, and format conversion. The models are even becoming multimodal, meaning you can input images and audio alongside text. This is a significant departure from the single-task-focused 'narrow AI' we had before, with sentiment analysis of text and computer vision applications being good examples," says Vandelaer. 

How intelligent are these models really? 

But are these models as intelligent as humans? According to the AI expert, this is not an easy question, in part because there is no good definition of intelligence. "We understand very well how LLMs work on a technical and mathematical level. But there is a lot of debate: do these models have a deep understanding of the world, or is it more superficial?" 

Vandelaer believes that LLM models are intelligent to some extent. "Humans have highly complex interactions between the input we receive and the output we generate. Moreover, both short-term and long-term thinking are involved. We see this in models like GPT as well. They work with high-dimensional input, complex correlations are found based on this input, and high-quality output is generated through processing. In my opinion, this is quite similar to how intelligence works in humans." 

Do LLM models have consciousness? 

Vandelaer says, "Another tricky question, because there is no good definition of consciousness. Consciousness is such a complex process that we don't even understand it ourselves. I sometimes compare it to the concept of 'life force.' In the past, people believed that humans and animals possessed life force, but not, for example, stones. And no one knew why life force was sometimes present and sometimes not. Later, it turned out to be complete nonsense: nowadays, we understand that life force is a much more complex concept consisting of different components."

The data science specialist expects that the term consciousness will blur further. And we will increasingly discuss the different components that make it up, such as sensory perception, attention, self-awareness, and metacognition ('thinking about thinking'). "A robot in a factory that needs to transport objects from A to B needs to develop a form of awareness: where am I? Perhaps a similar process has occurred in humans and animals, but our goals and environments are much more complex," says Vandelaer.

He suggests that AI systems might eventually develop more complex perceptions than humans. "Think of large-scale perceptions from many different sources, such as visual, auditory, textual, and chemical data. When AI systems can process this type of data simultaneously and continuously, it may provide a more detailed and richer picture of reality than we can perceive ourselves." 

Webinar: Integration of ChatGPT into Business Processes 

This article was created in response to the Cegeka webinar "Integration of ChatGPT into Business Processes" on June 8, 2023, as well as the articles "Three Ways to Integrate AI Tools Like ChatGPT into Your Business" and "The Unprecedented Opportunities of AI Tools Like ChatGPT for Your Business." Want to learn more? You can watch the webinar here.