Is ChatGPT really artificial intelligence? Before making any conclusions let’s go to the foundations by understanding how intelligence is defined.
Intelligence in humans has many different definitions. For scientists measuring IQ, for example, intelligence is the capacity to adapt to change. For some, intelligence is the ability to remember a lot of information and use it when necessary. To make sure we’re all visualizing the same thing, let’s have a look at the definition of intelligence for humans, and for machines.
Merriam Webster, the publisher of one of the most trusted dictionaries, defines intelligence as “the ability to learn or understand or to deal with new or trying situations”.
That same dictionary defines artificial intelligence as “a branch of computer science dealing with the simulation of intelligent behavior in computers” and “the capability of a machine to imitate intelligent human behavior”.
For this article, we will talk about the second definition of artificial intelligence and
Is Chat GPT really Intelligent?
ChatGPT is undeniably good in a lot of domains, but it is clear that it is still a work in progress. For now, OpenAI has put a lot of effort into making ChatGPT as safe as possible and users often see ChatGPT saying “As an AI language model, I’m not able to…”. Removing some limits and doing further testing will allow ChatGPT to get closer to true intelligence. But now, the version that has been released to the public has limitations. OpenAI’s goal is to create a truly intelligent AI, and they might be working on it already.
Why Chat GPT may not be considered a real AI
ChatGPT is a language model, which means that its first goal is to have a conversation that flows naturally and makes sense. It is not perfect at math or coding for example.
ChatGPT is like a doctor who knows just enough to guide a patient to the right specialist and maybe already offer an explanation of the problem or a piece of answer to the question.
ChatGPT’s output sounds plausible because ChatGPT is trained to chat in the best way possible, but it doesn’t think about what it says, wonders if it may have been wrong, or uses common sense.
ChatGPT is limited in its knowledge, it does have a ton of information and can build a meaningful output thanks to the data, but it does not learn from the newest studies, it does not integrate user feedback into its core knowledge without OpenAI having a look at it first.
ChatGPT is really a giant database that knows how to formulate an output and is limited by some pre-built ethical rules. It knows it is a language model, and that it sometimes lacks the perfect response based on the input.
ChatGPT 3.5 does not have access to the internet and ChatGPT 4 has some limits regarding web browsing, which is also a setback to unlock the full potential of ChatGPT. As ChatGPT explains: “It’s kind of like having a conversation with a really knowledgeable friend who only knows what’s written in that big book and can’t look up new things online”.
How to define a truly intelligence AI
A truly intelligent AI (or artificial general intelligence – AGI), would be able to understand, learn from, and apply knowledge across pretty much all tasks. It would be able to extract knowledge from a specific domain (let’s say marketing for the sake of the example) and apply what it knows to solve an issue in a different domain, like finance. It should also be able to understand its limits and openly say “I don’t know”, instead of making up information (the AI term for this is “hallucinations”).
That AI would be able to set tasks and goals for itself, in order to gain new knowledge or work more efficiently. It should also set limits and think “no, I shouldn’t learn about that”, “no, I shouldn’t say that, it’s not right, it’s not appropriate”, etc.
A truly intelligent AI would also refuse a change even if it does not seem to make sense. Let’s say the user wants ChatGPT to write an output without spaces, ChatGPT might not understand why, but if it deems the request harmless, it could agree to do it, or even better, question the user on the purpose of this prompt, hence the reason why users should have a basic understanding of ChatGPT prompt engineering.
Another thing that an intelligent AI should be able to do is solve complex problems and explain how it did it. For now, ChatGPT is able to solve problems to some degree, but when the user asks how ChatGPT did it, it happens that GPT simply replies “It was in the text I was trained on, but I do not have a source for this information”.
On the creativity side, a truly intelligent AI should be able to create content that is unique, makes sense, and can be integrated into a specific complex project. Right now if you ask ChatGPT to generate lyrics for a song, it won’t write anything stunning. It will rhyme and be on topic, but it likely won’t be considered as deep or thought-provoking. Thinking of science, AI should be able to do something that researchers do daily: thinking of concepts and ideas that we haven’t discovered yet. Discovering something new starts with questioning what we already know. Humans struggle to picture the void, and nothingness in general, can AI do it?
An intelligent AI should be able to understand social cues, hints from the conversation, sarcasm, and other parts of communication that aren’t just words. It should be able to understand what it says and how it is said. It should also be able to understand, and interpret human emotions and respond to them.
Another point we should consider is ethical reasoning. The AI might have all the data needed to help the user do illegal or malicious activities, but the AI should be able to refuse such requests and see through the final goal of the user.
Lastly, a truly intelligent AI would make you feel like you are talking to another human and should be able to bond with you, understand the relationship between itself and the user, and have meaningful conversations.
ChatGPT is clearly a huge advancement in the world of AI, but it was released less than a year ago, and there is no doubt that more improvements are on the way. OpenAI surely has the experts to make a ChatGPT that is even more intelligent, but they must make sure that it is safe to use before it is released. So hold onto your seats, it’s just the beginning of a new era.