The ever-accelerating nature of development in machine learning has brought us to the brink of general artificial intelligence, it might be as near as 2025, or perhaps even closer. This explosive development is rapidly changing the way we produce, consume, enjoy, create and, in general, how we live.
Since the times when we were contemplating the opportunities the 2010 revival of AI would bring the term singularity started to pop-up in the media and ever increasingly in day-to.day conversations. What appeared to be science-fiction for some, or perhaps inflated through media hype, turned out to fall short in their expectations.
The term “singularity” was coined by mathematician John von Neumann, and popularized by science fiction writer Vernor Vinge. The idea is that at some point in the future, artificial intelligence will surpass human intelligence, leading to a rapid acceleration in technological progress and a fundamental change in the nature of human civilization.
When asked about it some scientists pushed its arrival as far as 2100, others to 2050, the most daring science fiction writers to 2030. Some computer scientists, especially those working with deep neural networks, dismissed it or labeled it as impossible, considering AI would excel in individual or “narrow” tasks to superhuman levels, but would never generalize to all cognitive tasks.
In 2022 we got GATO, GPT, DALL-E, and Stable Diffusion, which are single-purposed AI models that have made progress towards achieving general AI. They are specialized in specific tasks and can outperform humans in their areas of expertise, but they also have the ability to perform multiple tasks and adapt to new situations.
These models have pushed the envelope of single-purposed AI closer to general AI, and, thus, to singularity. While they are not fully general-purpose systems, they have shown that it is possible for AI to perform multiple tasks and adapt to new environments, which is an important step towards developing more general-purpose AI systems. And they’re doing so at an amazing speed.
In the case of Stable Diffusion it is noteworthy that it has gone the open-source way and this is making it grow impressively, without needing the ample single-source monetary resources that propel DALL-E and GPT.
It is now quite possible that these (and other) models will achieve such a level of generalization in their next two iterations that they will be used in every aspect of our lives. And they will be integrated into the craftsmanship of many creative and managerial jobs, making the current study programs and human resource profiles obsolete.
It is time to appropriate these new techniques deeply and rapidly. We need to understand and incorporate them into the dynamics of governments, the companies, institutions and our daily life. Democratization of these technologies is essential, and perhaps, the fact they are becoming generally intelligent is the factor that will enable every person to interact with them. This way technology can be again a means for inclusion. This way humans can stay relevant and centric in a new AI world.