Intro

I don’t know what got more hype: ChatGPT itself or the potential to replace human jobs. Anyway, the goal of this article is to define what can’t be done by famous AI and whether it can replace humans (spoiler: not now).
I won’t start with the ChatGPT terms. Surely, the chatbot that now has 100M+ users and is mentioned in all top media several times is known. BTW, check AI trends for 2023 in this article. However, it’s necessary to puzzle out the way it works.

How does AI work?

GPT-3 was taught using a combination of Supervised and Reinforcement Learning via Human Feedback (RLHF). The model is trained on a big text dataset collected from the internet during the supervised learning stage (Wikipedia, Twitter, and Reddit). These models are trained on the labeled dataset and then used to forecast future occurrences. Consider the following scenario: a teacher (Supervised learning) instructs a student (ChatGPT) on how to write an essay. The purpose of the student is to learn how to write an essay by being familiar with the tone, terminology, and structure of hundreds of essays. Look at the scheme:

Source: Medium

But, among those hundreds of essays, there will be both good and awful. That’s why AI may occasionally write a bad essay. Here is when reinforcement learning comes into play.

The instructor would then give the students periodic essay-writing assignments and provide feedback on the essay-writing task, telling them what they did well and where they may improve. The critique is utilized to motivate following essay writing assignments, allowing the student to grow over time. This is equivalent to the GPT model training’s reinforcement learning stage.

The model can answer questions after being given a large volume of text gathered from the internet. Yet, its precision will be insufficient.

What does ChatGPT think?

I asked ChatGPT about things it can’t do. Here’s the answer:

Oookay, we’ll keep it in mind and add some new insights.

The famous AI can’t:

Resolve logic tasks

ChatGPT passed exams like Law School Exams and Stanford Medical School clinical reasoning final (more on that here). We’ll check.
How would you resolve such a logical task?

A girl meets a lion and unicorn in the forest. The lion lies every Monday, Tuesday, and Wednesday, and the other days he speaks the truth. The unicorn lies on Thursdays, Fridays, and Saturdays, and on the other days of the week, he speaks the truth. “Yesterday, I was lying,” the lion told the girl. “So was I,” said the unicorn. What day is it?

The answer is Thursday. Sunday is the only day on which they both speak the truth, but today cannot be Sunday because the lion also tells the truth on Saturday (yesterday). Judging by the calendar, Thursday is the only day where one of them is lying, and the other is stating the truth with those two assertions.

That’s how ChatGPT resolves it:

However, he tries to figure out the flow. The answer is not right.

Maybe he was cheating on exams.

Discover more Tik-toks checking ChatGPT’s logic here and here.

Generate ideas

Remember our “How it works” intro? The ChatGPT was trained on existing ideas. It can’t produce something new.

Brain scientist Henning Beck says in a TED Talk that fresh ideas and thoughts are unique to the human brain. Machines work harder, more quicker, and more efficiently. The fundamental components of creativity and invention include trying and failing, taking a step back and taking a break, and learning from fresh and different perspectives. Because we are not computers, humans will always be creative.

Use ChatGPT for inspiration, but create your own new concepts utilizing such a human privilege.

In the case of ideas, there are many copyright questions. To be protected by copyright, it must be an original invention of the human intellect, according to copyright law. This implies that it must be the outcome of human invention, talent, or effort. Because AI systems lack their own thoughts, creativity, or consciousness, the output they produce is not considered an original invention of the human mind. As a result, it is not covered by copyright legislation.

It doesn’t mean that AI-generated content may be exploited freely without concern for others’ rights. To prevent infringing on the intellectual property rights of others, it is critical to seek permission or a license to use the content in a specified way in certain instances.

Write quality code

Yes, ChatGPT is able to write code and sometimes resolve really complex tasks. For example, it can assist in software development. The first answer may save you many hours of research. The next — use the syntax of another programming language.

Here’s the proof. I generated such a prompt.

ChatGPT used the wrong syntax.

Hopefully, your web development services company checks the ChatGPT code before implementing it. Anyway, you can always use our professional team. Check our web development approach here.

Filter and check facts

The AI can’t independently verify the accuracy of the information. It is limited to the information that has been trained into the model, which may not always be the most up-to-date or accurate information available on the web.

ChatGPT would be a bad journalist.

Adapt on the fly

The usual AI development approach now consists of two different phases: training and deployment.

During training, an AI model consumes a pre-existing static dataset to learn how to do a certain task. The parameters of a model are fixed once the training phase is completed. After that, the model is deployed, where it provides insights about new data based on what it learned from the training data.

Suppose we wish to update the model due to new data or changing conditions. In that case, we must retrain it offline with the updated dataset (a computationally and time-intensive procedure in general) and then redeploy it.

In real-world contexts, there is a constant stream of incoming data. Fresh knowledge is gradually made accessible; conditions shift over time, sometimes dramatically. Humans are capable of incorporating this continual data from their surroundings in a dynamic and fluid manner, modifying their behavior as they go.

Physical work

AI and robotics are incapable of doing sophisticated physical tasks that require dexterity or perfect hand-eye coordination. AI is incapable of dealing with unfamiliar and unstructured areas, particularly those that have not been observed.

All in all

While AI has the ability to automate many jobs now performed by humans, it is doubtful that it will totally replace human labor in the manner it currently operates. It can help, for example, assist developers in developing software, but can’t provide custom software development by itself.

While AI algorithms may be trained to do certain jobs with great accuracy, they are incapable of the creative thinking and problem-solving skills that humans excel at. This means that human labor will always be needed in industries requiring these abilities, including research, design, and strategic planning.

Additionally, AI technology research and implementation do not occur in a vacuum. As AI advances, it is conceivable that new sectors and career possibilities will emerge. For example, the AI industry’s expansion has already resulted in the development of new employment in sectors such as data science and machine learning.

Now AI can't replace humans, but it can significantly boost business. If you want to catch the wave, choose a reliable partner for integrating AI into your business.

Generative AI in Business: Practical Applications for Growth