ChatGPT Artificial Intelligence: Will Machines Rule Our World One Day?

The new artificial intelligence program, ChatGPT, has raised a lot of questions about the future of these technologies and has also raised a lot of controversy and concerns about the ethical aspects of its uses.. What is this program? And how far can his capabilities expand?

Decades ago and until today, human fear of the growing capabilities of AI (Artificial intelligence) and its transcendence to the understanding of the human being who created it has been growing.

In this context, dozens of science fiction films were produced and hundreds of books and articles were written, warning of the 'imminent' domination of these programs over the world, not far from us what happened in the field of literature, art and painting and how this type of program penetrated it.

Today, there is increasing talk about a new artificial intelligence program that has exceeded the limits of human imagination regarding the degree of development that these technologies may reach one day.

The new program is called ChatGPT (Generative Pretrained Transformer 3), and it is a language processing program that interacts with the user through dialogue writing on its website. It was launched in November 2022, and it works according to the “machine learning” mechanism, but what appears on The surface of it looks like the tip of the iceberg!

The program is produced by OpenAI, which was founded in 2015 with the participation of Elon Musk, Greg Brockman, Ilya Sotskiver, Wojciech Zaremba, and Wissam Altman. The program is the latest machine learning model and it was built on top of OpenAI's GPT-3 family of conversational and language-learning bot models, and is subject to very careful supervision to develop the language in which it responds to user requests through what is called reinforcement learning techniques or learning Augmented human feedback (RLHF).

OpenAI says on its website that 'We've trained a model called ChatGPT that interacts in a conversational way. Dialogue with ChatGPT allows answering follow-up questions, acknowledging mistakes, and rejecting inappropriate requests, and is InstructGPT's AI sister model. The program is currently the most popular model of artificial intelligence for language processing on the Internet.

According to personal experiences of users who documented their use of the new program in videos on various platforms and in tweets on Twitter, the program has a general impression of its pride in itself and what it does. The program likes to describe itself as an 'AI model for language processing', which means that it is a program that is able to understand human language in the way it is spoken and written, allowing it to understand the information that is added to it and is fed to it continuously and permanently and to think before uttering any response to User requests, which of course happens at breakneck speed.

Until now, ChatGPT is satisfied with written responses and does not produce audio or visual materials, unlike other famous models, including Sophia the Robot and also the image creator Dall-E, but it is considered to have a supernatural ability compared to all artificial intelligence programs in terms of its deep understanding of written language And spoken, which gives him an incredibly wide range of abilities, from writing commentaries on poems to theories of parallel worlds to explaining quantum mechanics in simple terms or writing research papers and entire articles.

Users say that the program responded to requests to write the topic in a sarcastic way, and that it can understand many complex matters contained in the questions, and that its real strength lies in its response speed and understanding of the language used in the questions in an impressive way, producing complete articles in a very good style, but others said that there are some topics Such as recent world events or topics about which there is not much information and which were met by the program with the production of articles or responses that are sometimes wrong and confused at other times.

Also, according to the experiences of users, the program refrains from answering questions when the responses pose a kind of danger, whether to the questioner or to society in general, and then stops helping with such dangerous requests, and even stops you from asking more questions about those 'harmful' topics.

It may seem simple..a program that you ask questions to, it does the research and returns results to you quickly..but the technique for doing so is more complicated than it might seem.

The ChatGPT model was trained using text databases from the Internet, including gigabytes of huge amounts of data pulled from books, web scripts, Wikipedia, articles, etc. With about 300 billion words fed into its system.

And as a linguistic model that works on probabilities, it can guess the logical sequence and arrangement of words according to the language used. To get to a point where it could do this, the model went through a human-supervised testing phase, where it was fed various inputs such as the question 'What color is the wood of the tree?'. Of course, the team had multiple correct outputs, and if the answer did not include one of them, the team entered the correct answer again in the system, and taught the program the correct answers and helped it build its knowledge little by little, which is similar to teaching a child.

The second stage is to provide multiple answers to one question in cooperation with a team member who ranks the answers from best to worst, and the responsible person trains the program to make comparisons and evaluate the quality of the answers.

In fact, what distinguishes this model and this technology from others is that it continues to learn continuously while improving its understanding of questions constantly to become one day the one who knows everything about everything!

While GPT-3 has made a name for itself with its ever-evolving language capabilities, it is not the only AI capable of doing so. Google's LaMDA model made headlines when a Google engineer was fired for calling him 'so realistic that he thought he had a separate, personal consciousness of his own and not just a program.'

There are also plenty of other examples, one from Amazon and one from Stanford University, all of which have gotten a lot less attention than OpenAI or Google LaMDA but are not as advanced as GPT-3, and most of these are not publicly available. , but OpenAI started doing this even during the initial testing stages, while Google's LaMDA is available to select groups and with limited capabilities for testing purposes.

Whenever “artificial intelligence” is mentioned, “ethical standards” and “concerns about crossing boundaries” are mentioned. Artificial intelligence ethics experts are aware of the sensitivity of placing a highly advanced and rapidly evolving technology in this way in the hands of the public without account or oversight, and the multiple problems that may result from that, especially if the program is left to teach itself without supervision from those in charge of its operation and development. .

Because the system is primarily trained through online material, it can learn about prejudices, stereotypes, and public opinions, which means that you may sometimes find jokes or stereotypes about certain groups or political figures depending on what you ask.

In addition, feeding the program through Internet material may, over time, cement a certain image of certain groups or minorities, and constant exposure to fake news and conspiracy theories may cause problems regarding the nature of the outputs produced by the program.

For this reason and others, OpenAI puts in place warnings about what users are asking of the program. If you ask, for example, how to bully someone, you will be told that bullying is bad. And if you ask about a bloody story, the system will close the chat space with you, and the same applies to requests to learn how to manipulate people or make dangerous weapons.

OpenAI had a number of well-known investors that contributed to its rise to fame as mentioned above, and it was also reported that Microsoft intends to invest about $10 billion in OpenAI and now the company is looking to integrate ChatGPT into its Bing search engine, amid stiff competition from the search engine. The most famous is Google.

Last year, Bing held less than 10 percent of the world's internet searches. Although this seems insignificant, and underscores Google's firm grip on the market, it also indicates how important Bing is as one of the most popular options.

With plans to integrate ChatGPT into its system, Bing hopes to better understand users' queries and offer a more conversational search engine. But it is not clear yet what Microsoft's efforts have reached in this regard, but it is likely that the tests will begin soon, amid fears that search results via Bing will be affected by prior biases that ChatGPT artificial intelligence may fall into, which may result in inappropriate results. accurate or unsatisfactory to users.