As natural language processing technology advances, the potential of chatbots and conversational VI systems is generating increasing hype.* One such system, Chat GPT, claims to be able to engage in natural conversation with people and even provide useful information and advice. However, there are legitimate concerns about the limitations of this and other conversational VI systems and their ability to faithfully reproduce human intelligence and interaction.
This paragraph is not mine. He wrote it Chat GPT, a conversational artificial intelligence (AI) software program, after I asked it to create an “introductory paragraph for a text in the manner of Kenan Malik, about doubting the ability Chat GPT-a“. I might have some objections to the style, but the attempt is certainly impressive. It's not hard to see why the latest version of the chatbot has caused such excitement, even fanfare, since it was released the week before last.
Fed by vast amounts of text from various people, Chat GPT it looks for statistical regularities in that data, learns how words and phrases are interconnected, and thus is able to predict the order of words in any sentence, as well as the mutual relationships of entire sentences. The result is a machine that convincingly imitates human language.
This mimicry ability allows the chatbot to write essays and poetry, make jokes, formulate programming code, and answer questions posed by both children and experts. Moreover, it works so well that there was both celebration and panic last week. "Essay writing is dead," wrote cognitive scientist Tim Kitzman, and he wasn't alone. Others argue that it will end Google as a search engine. The program itself thinks it could replace people in jobs like insurance agents or court stenographers.
However, a chatbot that writes essays for top marks will also tell you that if one woman can give birth to one baby in 9 months, 9 women can give birth to one baby in one month; that one kilogram of beef is heavier than a kilogram of compressed air; that crushed glass is a useful dietary supplement. He can make up facts and reproduce the prejudices of the human world in which he has been trained.
This chatbot's mistakes are so convincing that it is Stack Overflow, a discussion platform for computer programmers, banned users from posting his answers. "The main problem," explained the moderators, "is that, although the answers which Chat GPT the products have a high percentage of error, they seem to be correct". Or, as one critic put it, that chatbot is a masterful fool.
Some of these problems will be solved over time. Every conversation he participates in Chat GPT becomes part of the database used to improve the program. The next iteration, GPT-4, is expected next year; they will be more convincing and make fewer mistakes.
Behind the gradual improvements lies the fundamental problem of any form of artificial intelligence. The computer operates with symbols. Its program determines a set of rules by which one sequence of symbols is transformed into another or statistical patterns are recognized. But it does not determine what those symbols or patterns mean. For a computer, meaning is irrelevant. Chat GPT "knows" (mostly) what seems meaningful to people, but not what makes sense in itself. That program is, in the words of cognitive scientist Gary Marcus, "a mime that doesn't know what it's talking about."
While thinking, speaking, reading and writing, people also operate with symbols. Unlike computers, however, meaning is everything to humans.
When we communicate, we communicate meaning. It is not only the sequence of symbols that is important to us, but also their meaning; not just syntax, but also semantics. Meaning is realized for people through our existence as social beings, embodied and embedded in the world. I have meaning for myself only to the extent that I live in a community of other beings who think, feel and speak.
Of course, people lie, manipulate, are attracted to conspiracy theories that can have devastating consequences. And it is an integral part of the existence of a social being. But we recognize people as imperfect and dishonest; we are able to tell when they are bluffing or manipulating.
When it comes to machines, however, we tend to treat them either as objective and impartial, or as potentially evil if they are "conscious". We often forget that machines can also be biased or simply make mistakes, because they are not grounded in the world like humans, but also because they are programmed and trained by humans, based on data collected by humans.
We live in an age when appearance is often more important than deeper meaning. An age in which politicians too often pursue a policy not because it is necessary or in principle right, but because it goes over well in focus groups. An age where we often ignore the social context of human actions or statements, blinded by literal nonsense. An age where, as author and educator John Warner puts it, students are "rewarded for digesting existing information" in a system that "privileges surface-level accuracy" rather than "encouraging writing and critical thinking skills." That seems to be Chat GPT that it is so easy to write top grade essays, he says, "reflects the poor level of our valuation of knowledge."
All this does not deny the extraordinary technical achievement of the latest chatbot, nor the amazing feeling of interacting with it. Undoubtedly, it will develop into a useful tool, which will help improve both human knowledge and creativity. But it is necessary to keep our distance. Chat GPT reveals not only the progress in the field of artificial intelligence, but also its limitations. It also sheds light on the nature of human cognition and the character of the modern world.
On the first ball, Chat GPT it raises the question of how to treat machines that are far better at fooling around and spreading misinformation than humans themselves. Given the difficulties in dealing with human misinformation, this issue should not be delayed. It's not good to be so mesmerized by the persuasiveness of chatbots that we forget the real problems these programs reflect.
(The Guardian; Peščanik.net; translation: M. Jovanović)
___________
* Chatbot (eng. chatbot, from: chat, chat + robot), a computer program, simulates and processes human conversation, written or spoken, in the interaction of people and digital services. It is based on the elements of the so-called artificial intelligence (AI) such as automated rules, natural language processing and machine learning; prim.trans.
Bonus video:
