Artificial general intelligence is probably an unnecessary term that confuses people and fails to encapsulate the basic concept of “intelligence”.
While almost everything in this article is debatable, it seems obvious to me that there isn’t much that is “general” about intelligence – in fact, intelligence is about the exact opposite, as far as I understand it. Intelligence is the ability to make a very specific choice – decide on the correct course of action or the right thing to do.
That requires learning at least something about what you are deciding on. Which is what other forms of artificial intelligence already do, don’t they?
Machine learning and deep learning are basically differentiated by the volume or breadth of data they ingest and the capabilities they have in analysing that data and making decisions based upon that data.
ML acquires broad-based data by itself and makes decisions based on parameters programmed into it, while DL takes a deeper look at a specific area in order to do the same thing.
Artificial general intelligence seems to conjure up in the mind a know-it-all AI system that can arbitrarily outperform you without having to learn anything that you have spent years learning.
But that’s just our common-as-muck take on it. Let’s look into this in a bit more depth.
Do we really need more jargon?
Artificial General Intelligence (AGI) is a concept that has captured the imagination of researchers, technologists, and the public alike.
The term “Artificial General Intelligence” originated in the early 2000s, attributed to researchers such as Shane Legg and Ben Goertzel, who used it to distinguish broad, human-like intelligence from the narrower AI systems of the time.
Unlike narrow artificial intelligence (AI), which excels at specific tasks, AGI refers to an AI system capable of understanding, learning, and applying knowledge across a wide range of tasks – similar to human intelligence.
What exactly is meant by ‘Artificial General Intelligence’?
AGI is defined as an AI system with the ability to perform any intellectual task that a human can do. It is not limited to specific domains like image recognition, language translation, or playing chess.
Instead, AGI aims to replicate the adaptability and versatility of human intelligence, including reasoning, problem-solving, and abstract thinking.
The term “general intelligence” in AGI implies a system’s capacity to learn and generalise across different domains without being explicitly programmed for each.
Proponents envision AGI as a transformative technology with applications ranging from healthcare and education to scientific research and creative endeavours.
Is AGI currently available?
No, AGI is not available yet – at least, not in the form that its proponents want it to be. While AI has made remarkable strides in specific domains, creating a system with the broad and adaptive intelligence of a human remains a work in progress.
Current AI systems, including state-of-the-art models like OpenAI’s GPT and Google’s Bard, fall under the category of narrow AI (ANI), which is highly specialised but lacks the ability to generalise across tasks.
Developing AGI involves overcoming significant challenges, including replicating human cognitive abilities like common sense, contextual understanding, and emotional intelligence. Many experts believe AGI is still decades away, while others question whether it is achievable at all.
Top companies developing AGI
Several organisations are actively pursuing AGI research. Here are ten notable companies:
- OpenAI: Known for its GPT series, OpenAI aims to ensure AGI benefits all of humanity.
- DeepMind (a subsidiary of Alphabet): Specialises in combining deep learning with reinforcement learning to achieve AGI.
- Anthropic: Focuses on AI safety and developing aligned AGI systems.
- Microsoft: Invested heavily in AI research, including partnerships with OpenAI.
- Google Brain: Works on fundamental AI research with potential applications in AGI.
- IBM Research: Known for Watson, IBM is exploring the next steps in AI development.
- Nvidia: Provides the hardware and software ecosystems driving AI and AGI research.
- Meta AI: Conducts research in AI systems with potential pathways to AGI.
- Tesla: Through its robotics and autonomous driving initiatives, Tesla explores generalisable AI systems.
- Baidu: China’s leading AI company, investing in AGI as part of its broader AI strategy.
Is AGI different from existing AI?
The primary distinction between AGI and existing AI lies in scope and adaptability. Narrow AI excels at specialised tasks but cannot generalise across domains. AGI, on the other hand, aspires to perform any intellectual task a human can handle, adapting to new tasks without reprogramming.
For example, a narrow AI model trained to detect diseases in medical images cannot suddenly start writing coherent essays or solving physics problems. AGI would be able to do all these tasks and more, seamlessly transitioning from one domain to another, according to its supporters.
But how can it do anything without learning anything, which brings us back to our earlier question about what makes it different from any other form of AI.
Machine Learning vs Deep Learning vs AGI
To understand AGI, it’s essential to clarify related terms like machine learning (ML) and deep learning (DL):
- Machine Learning: A subset of AI that uses algorithms to analyze data, learn patterns, and make predictions. Traditional ML involves features manually engineered by experts.
- Deep Learning: A subfield of ML that uses neural networks with many layers to process vast amounts of data. Deep learning has driven recent breakthroughs in areas like natural language processing and image recognition.
- Artificial General Intelligence: Encompasses ML and DL but aspires to integrate these and other approaches into a single system capable of general intelligence. AGI is a broader, more ambitious goal beyond the scope of current ML and DL systems.
Is AGI just a buzzword?
Critics argue that AGI is often overhyped, with some treating it as a marketing buzzword. Given the current limitations of AI systems, terms like AGI can create confusion and inflate expectations.
For example, some companies might use “AGI” to describe advanced narrow AI systems that are far from true general intelligence.
Additionally, the concept of “general intelligence” itself is debated. Humans are not born with universal knowledge; we learn incrementally, domain by domain.
In this sense, general intelligence might simply mean the capacity to learn across different contexts. If so, AGI development might not require replicating human cognition but instead creating systems with superior learning capabilities.
Challenges and ethical considerations
The pursuit of AGI raises profound ethical and societal questions:
- Control and safety: How do we ensure AGI aligns with human values and remains under control?
- Economic impact: Could AGI lead to mass unemployment by automating complex jobs?
- Existential risks: What safeguards are needed to prevent AGI from becoming a threat to humanity?
Developing AGI responsibly requires collaboration between governments, industry, and academia to address these challenges.
A general, no less
Artificial General Intelligence seems to have become widely accepted as representing the next frontier in AI research, promising revolutionary possibilities while posing significant challenges.
Although AGI is not yet a reality, its development is actively pursued by leading companies and researchers worldwide.
By understanding the distinctions between AGI, machine learning, and deep learning, as well as the broader implications of this technology, we can better navigate the rapidly evolving landscape of AI – if only from the jargon point of view.
Whether AGI is a realistic goal or a distant dream, its potential impact makes it a subject worth exploring and understanding. As research progresses, staying informed about the latest developments will be crucial for professionals and enthusiasts alike.
But, ultimately, in our opinion, it looks like AGI could be used to create an “elite” layer of AI which gets its way over supposedly “lesser” systems. Why? Because it has the rank of “General”, with a capital G.
The current public understanding of ML and DL could perhaps be enhanced by explaining AGI as something that simply ingests more data and has more analytical capabilities, and is perhaps faster. Like comparing a regular road car to a supercar – both still cars, but obviously very different in terms of speed, power and manoeuvrability maybe.
But AGI is not being sold as that – it’s not just a better car. It seems it’s being presented as a paradigm shift that outranks and is separate from – or rather higher than – all other forms of AI.
We don’t buy it. But we could be wrong, of course.
