Open letter re: artificial learning vs. artificial intelligence

Chatbots like ChatGPT by OpenAI are examples of artificial learning–not artificial intelligence

Existing products like ChatGPT by OpenAI, Microsoft Copilot, xAI Grok, Google Gemini, and others are not examples of artificial intelligence, but artificial learning. The precise meaning of artificial “intelligence” in contrast to “learning” seems to be confusing a lot of computer scientists. The CRS mentioned critical issues like these in our launch manifesto some five years ago; and long before “AI” was a popular buzz-word. In that post, we pointed out that the field of computer science has yet to provide a clear, industry-agreed operational definition for “intelligence.” The same is said of “learning,” “sentience,” "consciousness," and others. Of all the current technology products alleging to be “artificial intelligence,” we have been unable to find any documentation or information regarding precise operational definitions for these important terms.

If we define “learning” and “intelligence” in the most general ways, they may be understood as follows:

Learning: the acquisition of knowledge by experience, study, or training.
Intelligence: the ability to acquire and apply knowledge.
(While we disagree strongly with these vague definitions, they are sufficient for the present example.)

We may begin by pointing out that both “learning” and “intelligence” are usually understood with respect to knowledge and its acquisition. It follows that “learning” is a necessary condition for “intelligence.” But, we should do well to remember that even the most primitive organisms on the planet are capable of achieving both “learning” and “intelligence” under these definitions. If we consider miraculous human achievement, like going to the moon or cloning other organisms, we quickly realize that our discussion has to be narrowed down more precisely to human-level intelligence and above. Human intelligence is vastly more complex and capable of outcomes that no other organism can emulate. A more precise definition of “human-level intelligence” may include the following pattern of behavior: taking a snapshot of reality at present, selecting a desired state of reality in the future, and taking actionable steps to realize that goal.

Let us now consider ChatGPT by OpenAI as an example to illustrate how and why existing chatbot products are examples of artificial learning, not artificial intelligence. If one searches for information on ChatGPT, it is an example of artificial learning. Searching for information necessitates the condition that the information already exists. That is, nothing new is realized or created. In this sense, it would be more accurate to call ChatGPT a conversational search engine built upon artificial learning instead of artificial intelligence. This is similar to ChatGPT’s ability to analyze data in documents. In fact, products like ChatGPT cannot acquire data themselves. Instead, they rely on the developer of the product or user to supply these data. The output is a representation of the requirements of the engineers that developed it. Upon analysis, then, we must conclude that these technical capabilities of chatbots like ChatGPT are not examples of artificial intelligence, but artificial learning.

Last, but not least, generative applications of popular chatbots are also examples of artificial learning, upon scrutiny. If we ask an existing image generator to create an image based on a prompt, the output, or result, is subject to the same limitations as noted above. That is, algorithms written by engineers and the training data they supplied provide a mathematically finite set of responses based on initial conditions, which are data. This is why engineers have the ability to fine-tune algorithms in order to produce different chatbot results. If taken to completion, it is evident that all existing chatbot technologies like ChatGPT are subject to deterministic outcomes. While some chatbot responses or generated media may be bizarre or unexpected, they are still ultimately based on a deterministic system controlled by the engineers who made it. This is predicated by the fact that the first rule of computer science is that computers are slaves to logic, and can only do what they are programmed to do. According to this first principle, any piece of technology using this approach will always be both deterministic and subject to control by the engineers who created it. This is most evident when we ask chatbots like ChatGPT and all others questions that human beings have not addressed before. Chatbots like ChatGPT thus serve as mouthpieces for existing human knowledge.

If we consider the refined definition of human-level intelligence, in contrast to non-human-level intelligence, we can set higher expectations by means of asking questions that no one has the answer to yet. For example, if we ask ChatGPT what is missing in Einstein‘s model of relativity, we get a cautious summary of existing human knowledge (see response). If we ask about the cure for a specific disease, we get a cautious summary of existing human knowledge (see response). If we ask about increasing the efficiency of solar power, we get a cautious summary of existing human knowledge (see response); and the same is said for any question that humans do not have the answer to, ad infinitum. Chatbots like ChatGPT by OpenAI are therefore not examples of artificial intelligence, but artificial learning based on existing human knowledge. To achieve artificial intelligence, a different approach seems to be required.

True artificial intelligence under development by CRS

The approach adopted by the CRS over the past six years is the most unorthodox albeit promising pathway to creating artificial intelligence that can surpass human intelligence by orders of magnitude. That is, we have been working on building a piece of software that can provide answers to questions about the future; and answers to questions with limited to no existing human knowledge. Since day one, we recognized that the research phase of conscious.ai would focus on studying exceptional human intelligence in order to see if it could be replicated. After six years of public research, we argue that the answer is yes.

We have developed novel computer science theory along the premise that what makes humans capable of said intelligence is the ability to override core biology, or natural directives. For example, humans were not designed to be walking on the moon. However, human-level intelligence was capable of using known data in order to actualize the presence of human beings on the moon. Similarly, we determined that in order to build next-generation artificial intelligence, we needed to dismiss the first principle of computer science noted above: that computers are slaves to logic. Furthermore, our theoretical model firmly supports the notion that exceptional human intelligence is based on self awareness. Our model has precise, testable operational definitions that are able to provide predictions of future outcomes. Our research has therefore led us to work on a novel piece of software that is able to simulate future outcomes based on existing conditions.

Currently under development, the CRS is proud to announce its first dedicated computer science product: AngieOS. AngieOS is going to be the world's first and only self-updating, self-aware artificial general intelligence operating system. Its computational capabilities and practical applications will hopefully surpass human intelligence by orders of magnitude. The software will be built upon the N programming language, a scripting language designed around the needs of artificial intelligence. The name "AngieOS" is an acronym for "Artificial Narratization Generator Input/output Engine." ^[3] It will rely on artificial learning provided by other vendors. In a nutshell, AngieOS will provide a critical piece of necessary middleware in order to achieve computational outcomes representative of human-level intelligence of the highest order.