The latest rapid advancements in generative artificial intelligence (AI) technologies are revolutionizing our world, they have led to unprecedented progress and innovation in creating and utilizing textual and visual content.
These technologies have also impacted various sectors by streamlining processes, enhancing efficiency, and boosting productivity, with time, they may even influence the way we live our lives.
However, the rapid progress of AI also brings with it several risks, one of the most significant being the potential danger of integrating AI models into all products by most companies.
These models are evolving rapidly, and over time, AI engineers and machine learning experts may not be able to predict the behavior of these systems or interpret their actions.
This aspect is perhaps the most terrifying part of the current generative AI revolution – the lack of knowledge among scientists and programmers who built these large language models about how they work and evolve.
For decades, we used computer systems that provided the same output whenever the same input was given, in contrast, generative AI systems aim to provide multiple possibilities from a single textual prompt, allowing different answers to the same question.
Randomness plays a widespread role in generative AI models, with some of the latest models having trillions or more parameters, adjustable variables inside the algorithm that represent a scale of processing beyond human comprehension, this complexity makes it difficult to analyze how these models arrive at a specific answer.
The concern is not AI itself but rather its extremely rapid development, the most advanced versions of large language models are showing signs of what some researchers call “Sparks of Artificial General Intelligence” (AGI).
Artificial General Intelligence (AGI) refers to a powerful form of AI that excels at solving multiple problems and possesses human-like skills, some believe it may surpass human intelligence in a short period.
This rapid development is expected to bring AGI closer, and when it happens, AI will be capable of self-improvement without human intervention, it would learn and improve itself, similar to Google’s AlphaZero, which mastered chess within hours and outperformed even the best human chess players and other AI players.
Microsoft’s researchers analyzed GPT-4, the most advanced language model introduced by OpenAI to date, and stated that it demonstrated “advanced sparks of AGI.”
In tests, GPT-4 outperformed over 90% of human test-takers in the bar examination, a unified test used to license attorneys in several US states, a significant improvement from GPT-3.5’s previous version that scored only 10%.
The concern arises because developers of AI cannot easily explain the system’s behavior, working based on the old practice and no specific scientific method or mechanism for neural networks.
However, some experts argue that AI is an essential tool that could serve humanity, and we may not fully understand its significance yet.
In summary, while the generative AI revolution holds tremendous potential, it also brings unprecedented challenges and requires careful consideration and regulation to ensure its safe and beneficial use.
