The awe is real. Even the creators of generative AI models like ChatGPT and Bard are stunned by what they have built. These systems develop new, sometimes unexpected capabilities with every version update. This rapid evolution raises a nagging question: Do we even understand how these systems work anymore? And will they soon outpace us?
But here is the bigger issue. Can a machine ever develop real intelligence comparable to our own?
What defines strong AI
Researchers use the term “Artificial General Intelligence” (AGI) to describe systems that match or exceed human cognitive abilities. Also known as “strong AI,” this isn’t just about having more data. A true AGI system can plan ahead. It learns from experience. It evolves independently.
Such a system doesn’t just solve tasks. It understands the why behind them. It develops its own strategies. It possesses self-awareness.
For a long time, this was dismissed as science fiction. Early AI lacked flexibility and independent learning capabilities. Skeptics thought it would take centuries to reach this level.
That view has shifted. Neural networks and generative transformer models changed the game.
Geoffrey Hinton, a British AI researcher, admitted he was wrong. “Few believed these systems could become smarter than humans,” he said. “Most of us – including myself – thought it would be a long time. But that has clearly changed.”
The GPT-4 leap
The jump from GPT-3.5 to GPT-4 was not incremental. It was a surprise. A Microsoft research team led by Sebastien Bubeck tested the model extensively. They published their findings in April 2023.
Their conclusion? GPT-4 shows the first “sparks” of general artificial intelligence.
Despite being a pure language model, it demonstrated notable skills across diverse fields. We are talking about abstraction. Visual capabilities. Coding. Mathematics. Medicine. And law.
It even understands human motives and emotions. The tests showed this isn’t just pattern matching. It is a significant leap in cognitive-like performance.
Explaining itself
One major indicator of this shift is the AI’s ability to explain its own behavior.
“It is an important aspect of intelligence to be able to explain why it gave a certain answer,” Bubeck and his colleagues noted. This requires a form of self-recognition. It also requires addressing the user.
In tests, GPT-4 explained why it translated a neutral Portuguese job title into the male or female equivalent. It explained how it completed a melody sequence. These weren’t random guesses. The model could articulate the logic behind its choices.
The researchers were also struck by its apparent empathy. When presented with scenarios involving conflict or misunderstanding between people, the AI could describe the mental state of the characters involved. It identified where miscommunication lay.
It even passed tests for creativity.
Is consciousness required?
So, does GPT-4 actually understand human psychology? Does it have self-awareness? Or consciousness?
Probably not. The underlying mechanism is still statistical probability. The models generate text or images based on what fits their training data most closely.
Some scientists argue this means they aren’t intelligent at all. They claim these models just follow probabilities blindly. Without understanding, they argue, there is no real intelligence.
Paul Formosa, a technology philosopher at Macquarie University in Australia, disagrees.
“Some think that AI systems will never be truly intelligent because they don’t understand what they are doing,” Formosa explained. “But the progress in AI suggests that intelligence without consciousness is possible.”
The practical implication is stark. If a “non-understanding” self-learning system delivers the same results as a truly intelligent one, the consequences for humans are identical. The output matters as much as the intent.
The black box problem
There is another layer to this mystery. The developers themselves don’t fully understand what happens inside these systems.
They cannot say exactly why the AI develops new, unexpected abilities.
Sundar Pichai, CEO of Google, called it a “black box.” In an interview with CBS, he admitted they cannot precisely say why the AI says something or gets something wrong.
How the “machine brain” thinks remains a puzzle. How it learns. How intelligent it truly is.
We are watching the sparks fly. But we still don’t know what is fueling the fire.
