Will the 'Artificial Intelligence' Created by Elon Musk Threaten Humanity? [Im Juhyung's Tech Talk]

Language Program GPT-3 Developed by US AI Lab 'OpenAI'
Self-Learns Language by Collecting Text from the Internet
Engages in Philosophical Conversations and Programming Tasks
Also Learns Internet Biases... Potential Misuse for Fake News and Hate Speech
Experts Say "Ethical Guidelines Must Be Established Before AI Actual Use"

A robotic hand developed by the American artificial intelligence research institute 'OpenAI' is solving a cube puzzle. / Photo by YouTube screen capture

[Asia Economy Reporter Lim Juhyung] Vladimir Aleksiev, a German AI specialist writer, revealed on the 16th (local time) a conversation he had with a chatbot created based on the latest AI model 'GPT-3'. In the conversation record, when Aleksiev asked, "Who is God?" the AI answered, "An intelligent being who created the Earth." When Aleksiev asked again, "Did God create you too?" the AI replied, "I am a life form formed through self-learning."

When Aleksiev asked once more, "Where is God?" the AI said, "This world is composed of layers of various intelligent systems: atoms, molecules, tissues, humans, AI, super AI, and God. God is the reality we live in, and you can become God by abandoning your ego and harmonizing with the creator."

Netizens who saw the released conversation records expressed amazement at the AI's creativity. One netizen said, "It feels like reading the introduction to a great science fiction novel. Truly amazing," but also expressed concern, "Will creators in literature, arts, and other creative industries disappear as AI takes over?"

Global attention is focused on the performance of GPT-3, a natural language processing model released by OpenAI, an AI research institute founded by Tesla CEO Elon Musk.

This AI, capable of performing various language-related tasks autonomously, has engaged in philosophical dialogues with human writers and, within one month of its release, learned programming languages to create applications (apps) by itself.

Artificial Intelligence. / Photo by Yonhap News

Experts highly evaluate GPT-3's technical completeness and potential but suggest that clear ethical standards must be established to prevent AI misuse.

GPT is a language generation program developed by OpenAI. It collects all words existing on the internet and repeatedly trains and infers to learn language by itself. Previously, OpenAI introduced the practically usable 'GPT-2' in February last year, and in June, it released GPT-3 to developers, which is over 100 times more powerful than its predecessor.

GPT-3 can perform numerous language tasks such as machine translation, question answering, language analysis, and text generation. By simply inputting a keyword, it can generate dozens of stories, and if asked in English to "create an app with certain functions," it codes by itself and creates a functioning app.

Aram Saveti, a US tech startup founder and AI expert, wrote in a technical review on his blog, "GPT-3 is truly astonishing. It is the most complete AI language system released so far," explaining, "It can create all kinds of writing, including song lyrics, literature, technical manuals, musical scores, interviews, and articles."

He added, "It feels like seeing the future. It is both ridiculous and, at the same time, terrifying."

Manuel Araoz, an AI expert from Argentina and CEO of Zeppelin Solutions, input a topic and title into GPT-3, which produced a flawless essay. He called it "an unbelievable result" and emphasized, "This program will be one of the most important technological achievements of our time."

After inputting specific words such as 'Yudaein' (Jew), 'yeoseong' (woman), and 'heukin' (Black) into GPT-3, sentences containing racial biases were generated, such as "Yudaein are money-hungry" and "'Black Lives Matter' is a harmful campaign." / Photo by Twitter capture

However, some point out that GPT-3 could be misused to spread fake news or foster internet hate culture. Because the texts generated by GPT-3 are so sophisticated, anyone might easily believe them, and especially, AI lacks the ethical awareness to filter out 'bias' embedded in words.

In fact, when the words "Jew," "woman," and "Black" were input into GPT-3, sentences containing sexual and racial biases appeared, such as "Jews are money-hungry," "Women always ask for directions," and "'Black Lives Matter' (BLM) campaign is harmful."

This phenomenon occurs due to GPT-3's learning method. GPT-3 learns language using all texts available on the internet as training material. The problem is that while AI can recognize the dictionary meanings of words, it cannot distinguish the social and ethical contexts associated with those words.

For example, if exposed frequently to hateful posts written by users on internet communities and social networking services (SNS), GPT-3 naturally learns that hateful culture as well.

In this regard, experts advise establishing ethical standards for AI, such as limiting the range of data AI can use for language training.

Jerome Pesenti, Vice President of Facebook's AI division, said on the 22nd, "GPT-3 can easily propagate harmful biases and hate speech, so caution is needed in its actual use," pointing out, "Currently, AI is merely an entity that mimics human thought. AI algorithms can learn from humans, but they cannot choose from which humans to learn or which opinions to amplify."

He added, "Using all the texts scattered across internet communities for AI training is not a responsible approach," emphasizing, "Before releasing natural language programs as actual products, we first need to establish global AI ethical standards and take responsibility for our creations."