Artificial Intelligence: How close are we to the singularity?
Some individuals think it will occur before 2060.
These days we see a large number of publications about tools based on Artificial Intelligence. Mostly due to the momentum generated by ChatGPT. It is already known that there will be a paid version and other similar tools will surely continue to appear. The majority of them are highly tailored to each need. This is where the big difference with human intelligence relies on. We can accept that AI is outperforming us in specific tasks but something more general is still far away. Are We sure about that?
The term "singularity" refers to the time when Artificial General Intelligence—one that can be fully compared to that of a human being—will first exist. It was formerly believed that this would be established after a machine passed the Turing Test (1950). However, in 2022, both Google and OpenIA asserted that their models had accomplished it in one way or another. Without a doubt, this is an important milestone, but we cannot conclude that we have attained "singularity" based on the current expectations.
In 1989, the mathematical physicist Roger Penrose was already considering this question from a very demanding point of view. It does not matter if the machine can simulate that it is a human being (Turing Test). After all, he stated that this was going to happen anyway in the near future because the capacity of the human brain is limited while that of machines was increasing thanks to Moore's law. Penrose tells us that consciousness is the key factor and that the answer to understanding it is most likely at the quantum level. Other authors such as Yuval Noah Harari in Homo Deus (2015) agree that consciousness continues to be the great mystery of the modern age and perhaps it is the only thing that truly makes us human. Despite the above. The vast majority of specialists agree that the singularity will most likely be reached before 2060. Yes, as you are reading, in less than 40 years. This is based on various polls over the last 10 years on both the optimistic and conservative sides.
Let's see if the above has logical and reasonable support. Various studies have determined that the real-time processing capacity of the human brain is around 2.5 Petabytes (10^15) with 125 trillion synapses in its cortex, an astonishing number that seemed far from imitating just 15 years ago by the capacity that was had in processing and above all because it had only been possible to simulate for 1 second the functioning of a brain using 82,000 processors at the same time. However, with current advances in processors (even without taking quantum computing into account), it is estimated that by 2025 machines will reach that processing capacity (Kevin Drum, 2015).
So what will be missing? Leaving aside the mystery of consciousness, we only have abstract intelligence, that which allows us to infer or transfer knowledge from one field to a different one. The one that is not based only on repetitive experience (as in Machine Learning). Thanks to that intelligence, human beings have been able to organize themselves and create things, mixing concepts that are often totally disparate. In other words, we may never have a machine that is self-aware but one with the full capacity to improvise in new situations and even demonstrates common sense when making decisions. If so, we would be relieved of the ethical concerns mentioned by Penrose in "The Emperor's New Mind" (1989), as we wouldn't have to worry about having it work nonstop, buying and selling it (slavery? ), or what may happen to it if the power goes out.
Consider the possibility that intellect need not be a perfect replica of ours, which is another highly intriguing point of view. Even though nature has inspired many human endeavors, including flight, the solution we have discovered does not require airplanes to flap their wings like birds. As a result, it's possible that the kind of intelligence that achieves the singularity will be distinct but have enough capability and solvency to be regarded as intelligent generally. If this happens, a new era for humanity will begin, marking the most important turning point of this century or millennium. Maybe the end of the world as we know it.
This is a very broad subject and neuroscience also has a lot to contribute. In 2005, the Argentine scientist Rodrigo Quian published an article about the discovery of a type of neuron that represent concepts. Something that allowed us to take another step towards understanding what makes us human.
Thanks to the hype generated by ChatGPT, we can continue to be awed by the AI advancements in large language models (LLMs) that are yet to come while we wait to see whether or not these predictions about the singularity are accurate. For instance, Google is developing its PALM model (3x larger in terms of parameters).
And even then, it only represents a small portion of the money being spent on AI. The months ahead will be quite fascinating.
Thanks for reading.