Is AGI the Stuff of our AI Nightmares?
Open AI has opened a Pandoras Box that can ever be closed. The rapid adaption of generative AI shows humanities fascination with and curiosity of the unknown. But are tools like ChatGPT 4 and Llama 2 really the beginning of a true artificial general intelligence? While the term has hit the main stream in recent years, it is no new concept. The origins go back the last 40 years. The concept of AGI is rooted in the pursuit of creating machines that are not just adept at specific tasks, but also possess a general, flexible form of intelligence. This includes the ability to reason, solve problems, comprehend complex ideas, learn from experience, plan, and communicate in natural language.
Artificial general intelligence is mentioned as far back as 1997 in a scientific paper by Mark Avrum Gubrud found here. Mark defines the term as such:
By advanced artificial general intelligence, I mean AI systems that rival or surpass the human brain in complexity and speed, that can acquire, manipulate and reason with general knowledge, and that are usable in essentially any phase of industrial or military operations where a human intelligence would otherwise be needed. Such systems may be modeled on the human brain, but they do not necessarily have to be, and they do not have to be "conscious" or possess any other competence that is not strictly relevant to their application. What matters is that such systems can be used to replace human brains in tasks ranging from organizing and running a mine or a factory to piloting an airplane, analyzing intelligence data or planning a battle.
This meaning is still in use today. Artificial general intelligence doesn?t mean consciousness but rather that the AI model can be used generally for tasks only previously handled by humans. Do not confuse this with the models we have today, like self-driving cars that outperform humans or visual character recognition systems that also surpass humans, but a framework where all of these and more are possible outcomes that the AI model can teach itself. That distinction makes the yard stick for measuring an AGI significantly harder. While it is expected to be reached within our lifetime, our current generative AI models remain to fall short.
Digging into the origins further, artificial general intelligence is built upon the concept of "strong AI." Strong AI comes from philosopher John Searle in his paper "Minds, Brains, and Programs", published in Behavioral and Brain Sciences in 1980 where he discusses the thought experiment known as the Chinese Room.
Searle's thought experiment begins with the now real hypothetical premise: suppose that artificial intelligence research has succeeded in constructing a computer that behaves as if it understands Chinese. It takes Chinese characters as an input and, by following a set of instructions on a computer program, produces other Chinese characters, which it presents as the output. This is precisely what we see with generative AI models like ChatGPT. Now suppose that this computer performs its task so convincingly that it easily passes the Turing test: it convinces a human Chinese speaker that the program is itself a live Chinese speaker. To all of the questions that the person asks, it makes appropriate responses, such that any Chinese speaker would be convinced that they are talking to another Chinese-speaking human being. Again, an accomplishment of modern generative AI.
The real question Searle seeks to answer is does the machine understand Chinese or is it merely simulating the ability to understand Chinese? Searle called the first position "strong AI" and the latter "weak AI".
Suppose that you are in a closed room and have a book with a series of steps replicating the computer program to generate Chinese characters, along with sufficient papers, pencils, erasers, and tools. You could receive the input Chinese characters through a slot in the door, process them according to the program's instructions, and produce Chinese characters as output, without understanding any of the content of the Chinese writing. If the computer passed the Turing test this way, it follows, then you could as well, simply by running the program manually.
Searle asserts that there is no essential difference between the roles of the computer and the person in the experiment. Each simply follows a program, step-by-step, producing behavior that is then interpreted by the user as demonstrating intelligent conversation. However, the person in this experiment would not be able to understand the conversation. Therefore, it follows that the computer would not be able to understand the conversation either. Without understanding or intentionality, we cannot describe what the machine is doing as thinking and, since it does not think, it does not have a mind in anything like the normal sense of the word. Therefore, Searle concludes that the "strong AI" hypothesis is false.
Our current generative AI models do not pass this test of mind Searle proposed. At best, they are just a series of inputs manipulated and generated output without having the understanding of what it is doing. Perhaps ChatGPT 5 will have that understanding. It is a concerning thought. This is a far leap from where AI research has been in the past or even where we are today. The implications of an artificial intelligence that truly understands is the stuff of AI science fiction lore. While we are barreling ahead towards this discovery, we do seem less concerned about the guide rails postulated by so many science fiction writers and story tellers. But should we really be concerned?