Google’s New Statement on AGI Just Shocked Everyone
Demis Hassabis, the head of Google DeepMind, said something that surprised many people. He stated clearly that today’s AI systems are still very far from AGI. AGI means Artificial General Intelligence – a system that thinks like a human.
This statement is important because it comes from one of the world’s top experts. Many people thought that AGI was almost here. Hassabis says that is not true.
His view is that current AI models are powerful, but they miss several key features. It is not enough that a system solves hard tasks well. It needs much more to be called true general intelligence.
At AI w Biznesie, we watch this debate very closely. We see how companies use AI tools every day. These tools are useful, but they need careful oversight. You cannot trust them without checking their work.
What Hassabis Says True AGI Really Means
Hassabis defines AGI as a system that shows the full range of human mental abilities. It is not just about giving good answers in a chat window. Things like creativity, planning, autonomy, and understanding the world all matter.
Current systems can be very good at one single task. For example, an OpenAI model recently helped solve a hard math problem. That sounds impressive, but Hassabis says it is not enough.
His point is simple. A model can be a genius at math but still lack general intelligence. Real AGI must work well in many different areas, not just one.
The system also must be reliable. Today, AI models often make mistakes. They can “hallucinate” – which means they make up false information. This is a big problem if we want to call them truly intelligent.
The Math Example – A Success, But Not AGI
OpenAI announced that one of its models solved a major problem in geometry. The problem was from 1946 and was very hard. Mathematicians checked the result and said it was correct.
This is a huge technical achievement. The AI system did something that requires deep reasoning. But Hassabis cools down the excitement. He says this does not mean AGI is here.
He compares it to a person who is a genius in one field but weak in others. True AGI must be good at everything, like a human. That is still ahead of us.
Think of a doctor who is amazing at heart surgery but cannot treat a common cold. That is not a complete doctor. AI is similar – it can be brilliant in one area and fail in another.
Why the AGI Debate Exploded Right Now
In recent months, many people declared that AGI had already arrived. The reason was new, powerful AI models. People use them for coding, writing, research, and managing businesses.
But Hassabis says stop. The bar for AGI is much higher. This does not mean AI is weak. In fact, it is very powerful. But that is not the same as general intelligence.
The problem is that the definition of AGI has become fuzzy. Different people understand it differently. Helen Toner, an AI policy expert, says almost everyone has their own definition. This makes the debate hard.
For some people, AGI means a smart chatbot. For others, it means a system that can replace all knowledge workers. These are very different things. So people talk past each other without agreeing.
The Fight Between Optimists and Skeptics
Marc Andreessen, a well-known investor, thinks AGI practically already exists. He says the best AI models give better answers than most experts. In medicine, law, finance – everywhere.
This is a strong argument. For a normal user, such a tool looks like general intelligence. You can ask about almost anything and get a good answer. That changes how we see the world.
On the other side, Gary Marcus points out the weaknesses. Models still make strange mistakes. Sometimes they add random words, like “goblins,” for no reason. This shows their intelligence is “jagged” – very good in one place, weak in another.
Andrej Karpathy, a former OpenAI researcher, created the term “jagged intelligence.” It describes how AI does not grow evenly. In some tasks, it beats humans. In others, it fails badly. This is different from the human mind.
For example, one AI model can write a perfect business email but cannot tell if a joke is funny. A human child can do both. This unevenness is a key sign that AGI is not here yet.
What Current AI Systems Still Miss
Hassabis lists several specific gaps. First, models are not fully reliable. They work well once, then break with a small change. A tiny change in the question can lead to a completely different answer.
Second, they lack autonomy. Answering questions is not the same as planning and doing tasks alone. True AGI must act without help for a long time. It should set goals and work toward them.
Third, memory is a problem. Current systems do not have continuous memory like humans. They use a context window but do not remember everything forever. This limits their ability to learn and grow over time.
Think of a person who forgets everything after every conversation. That is today’s AI. You have to tell it the same things again and again. That is not how real intelligence works.
Fourth is understanding the world. Models work well with text but do not feel reality. They do not know what it means to be hungry, tired, or happy. This is natural for humans, but not for AI.
Fifth is true creativity. Hassabis stresses that AGI should not only solve tasks but also create new ideas. It should ask important questions. That is still beyond today’s systems.
Does That Mean AI Is Useless?
Absolutely not. Even if AGI does not exist yet, current systems are extremely powerful. They change how we work, learn, and run businesses. AI w Biznesie helps companies use these tools successfully.
AI models write code, summarize documents, help with research, and create content. They are already essential in many fields. But you must treat them as tools, not as full intelligences.
The biggest mistake would be either to ignore the progress or to trust the systems too fast. You need to find a balance. Enjoy the possibilities, but stay careful with important decisions.
One concrete example from our work at AI w Biznesie: A client used AI to draft marketing emails. The AI wrote great emails fast. But it also added a wrong claim about a product feature. A human had to catch that error. This shows the power and the limit of current AI.
What Hassabis Himself Says About the Time for AGI
In his statements, Hassabis suggests that full AGI could appear in a few years. He often talks about a range of 5 to 10 years. That is quite fast, but not tomorrow or next year.
This is interesting because it shows that his position is not simple. He does not say “never.” He says “not yet.” We are on the way, but we are far from the finish line.
In one interview, he even said that in 2026-27, people will look back and say, “oh, that is when it started.” This means the current period is the beginning of a big change.
The difference between usefulness and full intelligence is key. This debate has huge meaning for companies, governments, and normal users. It affects how we regulate AI, what investments we make, and who we trust.
For example, if a bank uses AI to approve loans, it matters if the AI is truly intelligent or just a good tool. Mistakes can hurt real people. So having a clear view of AI’s limits is not just an academic question – it is a practical one.
In summary, AGI in the strong sense has not arrived yet. But AI is already so powerful that it is changing the world. And it is changing fast. The most important thing is that we understand both the possibilities and the limits of current systems.
No responses yet