Photo Credit: Unite.AI
Some believe that the ultimate aim of artificial intelligence (AI) development is Artificial General Intelligence or AGI. An AGI would be able to complete any work that was given to it with sufficient time and computational capacity, replacing the need for an AI that can only accomplish certain activities. Some people consider ChatGPT to be the first instance of AGI, while others believe we are still at least a few years away from an AI that is capable of anything.
Actually, the discussion focuses more on how we define AGI and how to determine when we have reached it.
Artificial General Intelligence
Theoretically, an AGI agent should be able to complete practically every cognitive task that a person or animal can perform, and perhaps even do it better. Nonetheless, there are some things that not even an AGI can do, such as practically anything involving the physical universe. At least until robotics enter the picture.
Photo Credit: Homeland Security Today
Yet the focus has shifted from whether AGI is achievable to when it will be realized with the introduction and success of ChatGPT and Google’s Bard. The current GPT 3 model chatbots, according to some, come pretty close, while the more recent GPT 4 model, or even upcoming GPT 5 models, according to others, will cross that threshold.
Some continue to argue that we are still years or perhaps decades away. Yet regardless of where you think we are in the timeline, AI agents will eventually be able to perform practically all tasks that humans can on a computer.
ChatGPT: Is it an AGI?
No, not in its present form. Even if it is capable, the publicly accessible GPT-3.5 model is severely constrained in several areas, and OpenAI safeguards further limit it for ethical, legal, or safety concerns.
Yet unlike anything else we’ve seen, ChatGPT gives us a sneak peek at what real AGI would look like. The majority of the time, it is difficult to tell it apart from a real human because it understands and speaks in natural language.
Arrival Of AGI
It is difficult to predict when AGI will be realized because the idea is so speculative, and even when it occurs, there will be disagreement over whether or not it is real.
On the other hand, fans are watching to see if future iterations of OpenAI’s LLMs (Large Language Models) will be able to accomplish AGI. AGI has been predicted for the next GPT-5 model, which is anticipated to complete training later this year.
Of course, as the Twitter user points out, this will still be up for debate, and there are probably going to be ways to undermine it. Yet there is clearly room for this kind of intelligence to develop in the years to come.
Different experts place a different deadline for the possibility of AGI. Elon Musk once projected 2029, but a more comprehensive count by Metaculus places it at October 2032. That has pushed up from 2057 when the prognosis was made, according to AI enthusiast Rowan Cheung.
The traditional notion of an AGI and what modern language model AIs are achieving require fundamental distinctions. But, they won’t be “intelligent” in the sense that they understand what they’re doing or have any kind of incentive for such jobs beyond following the supplied instructions. They have the potential to develop the abilities of AGI.
Photo Credit: Scientific American
Model for natural language Chatbots function as prediction tools. With enough and the proper training data, they can produce very credible assessments of intelligence by predicting the following best word in a series. It may also make it possible to develop AI agents that are capable of a wide range of remarkable feats. Under the correct conditions, it can even produce chatbots like ChatGPT that appear intelligent and even sentient.
They are not, and they never will be. Most near-term concepts of AGI won’t be intelligent in the sense that they understand anything, just like ChatGPT and its ilk. But they might be exceedingly strong.