What's actually wrong with the concept of AI_post_2

Company

Company

What's actually wrong
with the concept of AI

Computer science has developed to a stage where delivering solutions/products doesn’t require direct programming. Many consider it a real breakthrough. The phenomenon is called artificial intelligence or, more rarely, machine learning. Even as we write this, a revision of AI achievements and efficiency is gradually taking root.

01

02

In its current state, the power of AI is greatly exaggerated. For now, it’s just a brilliant engineering solution. To qualify as intelligence — not to mention AGI — at least in some way, machine learning needs elements of human thinking.

The term artificial intelligence was coined in 1956 by John McCarthy, one of the founding fathers of the discipline. He defined AI as the "science and technology of creating intelligent machines."

04

AI does not, and cannot, possess any of the above. The one thing that has really been done well is ML methods for algorithmization of endless databases. Modern AI, or rather machine learning to be more accurate, can be referred to as database algorithmization (finding connections without understanding the "meaning"), a kind of infinite memory.

07

As such, intelligence (from Latin intellectus) means "understanding", “reasoning." It is derived from a combination of the words intelligere, i.e., "to perceive", "to cognize", "to think", and legere, i.e., "to collect." Consequently, to be called intelligence, artificial intelligence must be able to reason, understand, perceive, and think.

06

Scientists, developers, and engineers seem to have long ago "lost" the "super idea" of "artificial intelligence", which was about creating an artificial mind, an emulation of how the human brain functions. The level of AI development currently compared to the human brain is as infinitely far from it as a calculator.

03

But there is still no single answer to the question of what intelligence is. Philosophy has not yet resolved the question of the status of the human intellect. No criterion has yet been set as to when computers can be said to have achieved “sapience." Will we be able to come up with any such test at all before we find a universal (accepted by all) philosophical, metaphysical, and scientific answer to the question of what intelligence, let alone consciousness, is?

05

We know robots whose mechanics were "taken" from the mechanics of the human body. Airplane wings copy the way bird wings work. In "artificial intelligence", biology-related ideas are virtually non-existent, and at least for this reason, no comparison could be made to natural intelligence. It has NOTHING to do with thinking.

14

So what exactly is the problem? Let's try and look at it from another angle: artificial intelligence has NOTHING related to biology except for the word neuron, and an artificial neuron has nothing to do with a natural one.

13

As we can see, ChatGPT and its analogs are not capable of "human-like" learning and could only write a freshman term paper if consuming gigantic amounts of energy.

12

After graduating from school, institute, etc., people aren't supposed to refresh what they already learned. The purpose of learning is to transform the acquired knowledge into skills to be further used and improved. Otherwise, particularly talkative parrots could also be defined as multitaskers.

11

Today, there's ChatGPT, alongside Gemini and the like, and they seem to be solving a variety of problems. But these AI models are still a world away from human intelligence.

10

But this has not been done — and is unlikely to be. At most, this could qualify as some, albeit huge, set of facts/data/information, but certainly not a world model. It's like referring to an encyclopedia as a world model.

09

Modern multi-tasking AI systems were made to focus on games and tasks formulated in natural language. Thus, AI people are convinced that the self-supervised learning approach, for example, will allow creating a certain world model (e.g., for a game or a language) and then reusing that model to solve various end tasks rather than just the one for which that solution was created and customized.

08