Reflections on Artificial Intelligence, as friend or foe

Reflections on Artificial Intelligence, as friend or foe

Subject : Science and technology

Section: Awareness in IT

Concept :

     Uses and Limitations of Artificial General Intelligence (AGI):

• Current AI systems excel in narrow tasks, displaying superhuman performance and becoming valuable aids to humans.
• However, their performance declines when confronted with more general or ill-defined tasks that require integrating inferences based on common sense.
• Artificial General Intelligence (AGI) refers to intelligence that is not restricted or specialised. It encompasses the human-like “common sense” that is absent in current AI systems.
• There is a range of expert opinions regarding the feasibility of building AGI, with some asserting that machines may never achieve it, while others speculate that it could potentially emerge in the distant future.
• A significant development in the AI field occurred with the release of ChatGPT which utilises a Large Language Model (LLM) to generate text.
• LLMs are massive artificial neural networks trained on vast amounts of text data to establish a statistical “model.”
• ChatGPT’s remarkable success in generating coherent paragraphs garnered widespread attention, leading to speculation about the emergence of AGI in future iterations like GPT-4.
• However, critics argue against such claims, emphasising that LLMs operate at a fundamental level by predicting the most probable or relevant words based on their statistical models.
• They are often referred to as “stochastic parrots” devoid of any understanding of meaning.
• LLMs are also notorious for generating false information and fabricating citations for non-existent academic papers.
• The advent of true AGI will have profound implications. AGI could potentially surpass humans in various intellectual or mental tasks.

Challenges with Superhuman AI:

• AGI systems could represent a superior species created by humans, bypassing natural evolution.
• Consequently, the arrival of AGI warrants serious consideration and preparation.
• Building AI tools has become relatively accessible, and even narrow AI tools can cause significant harm when used with malicious intent.
• Large Language Models (LLMs) can generate convincing fake news, leading to mental distress and even self-harm. They can also manipulate public opinion, impacting democratic elections on a global scale.
• The absence of effective defence against such malicious behaviour raises concerns, particularly regarding AI-powered weapons and the lack of enforceable bans.
• As AI systems improve and assist humans, unintended harm may occur, disproportionately affecting certain groups due to inherent biases present in the data used for training.
• For example, facial recognition systems have shown greater accuracy on fair-skinned men than on dark-skinned women.
• Such biases can have catastrophic consequences in autonomous vehicles and medical diagnoses. Privacy is also at risk, as algorithmic systems constantly monitor individuals, violating the right to privacy.
• Recent advancements in AI have primarily taken place in companies with substantial computational, data, and human resources but lack effective public oversight.
• Ensuring ethical practices and the alignment of AI systems with public interests becomes crucial.