No.
Wait, you were looking for an explanation? No problem!
The Large Language Models (LLMs) that power today’s AI technology aren’t trained on enough good data, struggle to understand concepts humans naturally intuit, and can’t learn or synthesize new concepts the way humans can.
And that’s just scratching the surface.
What is AGI?
Artificial General Intelligence (AGI) refers to the kind of intelligence that can understand, learn, and apply knowledge across a wide range of tasks at or beyond the level of human capability.
It’s the holy grail of AI research, and some of the brightest minds in tech seem to think it’s just around the corner.
Prominent figures like Sam Altman (OpenAI’s CEO), Demis Hassabis (DeepMind’s CEO), Dario Amodei (Anthropic’s CEO), and Ray Kurzweil (Google’s Director of Engineering and a renowned futurist) have predicted AGI could arrive within the next decade.
When people that smart are making these claims, it’s tempting to hop on their bandwagon. Usually, I would.
But here’s the thing: after spending a lot of time knee-deep in current AI technologies, I’m not convinced.
The path these folks see? I don’t see it. In fact, I see inherent flaws in our current trajectory that make AGI seem… improbable, at least for now.
So, could some groundbreaking new technology deliver AGI? Absolutely.
But if we’re talking about today’s tech stack – the stuff powering the AI boom – I’m not betting on it. Let’s break down why.
Problem #1: Garbage In, Garbage Out (AKA Idiot-Based Training)
Current AI models learn by devouring massive amounts of data scraped from the internet. And let’s face it, the internet isn’t exactly the intellectual utopia we’d like to imagine.
For every smartly-written article, groundbreaking research paper, or BRILLIANT actionable cybersecurity blogs, there are countless poorly informed opinions, falsehoods, and nonsense.
“But Rob!” I hear you ask, “What if we only trained AI on data from the smartest people?”
That’s a nice thought, but even smart people make mistakes! Plus, humanity has a long history of being wrong about big things.
Five hundred years ago, if you’d asked an AI trained on the best available knowledge whether the Earth was the center of the solar system, it would have confidently replied:
“Yes, the earth is the center of the solar system. Copernicus? That guy’s a crackpot.”
AI is great for distilling what we already know, but it struggles with the unknown. Worse, it can anchor itself to our collective misconceptions, amplifying errors instead of delivering truth.
Problem #2: Limited Transferability of Knowledge
AGI, by definition, would need to effortlessly integrate diverse knowledge areas, but today’s AI wasn’t built for that.
All LLMs are great at specific tasks. Ask them to summarize a document like a SOC 2 report, or summarize text like the answers in a cybersecurity questionnaire and they’ll impress you.
But if you ask them to seamlessly combine knowledge from biology, physics, and ethics into a coherent strategy for solving a complex real-world problem, they’ll stumble.
LLMs can sometimes produce impressive outputs in multi-domain tasks through pattern matching, but their outputs often lack the validation and reasoning you’d expect from the human-level capability AGI is intended to mimic.
While transfer learning is improving, it’s still far from AGI-level integration.
Problem #3: No Innate Understanding of the Physical World
Humans enter the world with a basic understanding of physics. We intuitively grasp concepts like gravity, object permanence, and causality.
AI does not.
Unless explicitly trained, AI models flounder when faced with tasks requiring real-world logic. Take autonomous driving, for example. Even after training on millions of hours of video, these systems struggle with edge cases – unexpected scenarios that humans can handle with relative ease.
The inability to apply experience from limited data to novel situations is a glaring limitation.
Problem #4: Human Cognition is Ridiculously Complex
Human cognition is a complex web of synapses firing millions of electro-chemical reactions in harmony, which we experience as thought.
Current AI LLMs, by contrast, are essentially glorified text predictors.
They excel at identifying patterns in data but lack almost anything else that could be defined as part of human intelligence.
Being able to predict the next word in a sentence is a far cry from understanding the world like a human.
Problem #5: No Continuous Learning
One of humanity’s greatest strengths is our ability to learn and adapt in real time.
Current AI? Not so much. Once a model is trained, its knowledge is static. It can’t learn from new experiences or adapt to a changing environment without undergoing a costly retraining process.
Yes, some of today’s tools like ChatGPT have a “memory” feature that tries to “learn” based on what you tell it about yourself. However, the information stored in its “memory” is sent to the model as input. It’s like a context file. The model itself is unchanged.
AGI would need to continuously evolve, improving itself dynamically based on new inputs. Today’s AI systems are nowhere near achieving that.
Problem #6: Scalability Challenges
The computational and energy demands of current AI models are staggering. Big tech companies are dumping millions of dollars into acquiring the energy needed to power their expanding AI datacenters – going as far as investing into nuclear energy!
Scaling these systems to AGI-level complexity would require astronomical resources. Even if we could muster the required compute power, it’s unclear whether it would result in true general intelligence or just better pattern recognition.
So, What’s Next?
Don’t take me for some sort of AI-skeptic.
Today’s AI technologies are incredible, and they’re already transforming industries. I’ve used each of the major AI tools available and am investing in AI solutions to support Fractional CISO’s service delivery teams.
I find AGI to be a whole different beast.
The gaps between what we have now and what we’d need to achieve AGI are vast. Bridging them will require not just incremental advances but fundamentally new approaches.
Could someone invent a game-changing technology tomorrow that changes everything?
Sure.
But based on what we’ve got today, I’m keeping my expectations in check.
We can be your partner and offer expert support so you can confidently meet your compliance standards, reduce risk, and establish long-term trust with your clients. Contact us today, and we’ll gladly discuss how we can help.
