First, the first misconception revolves around AGI:
1. The AI systems currently in use are limited versions of AGI
Despite what many believe, the latest artificial intelligence is still far from human intelligence. Artificial General Intelligence has been the catalyst for all artificial intelligence scientists from the time of Alan Turing until today to reach an AGI that can mimic and even outperform human intelligence, resulting in many technologies and scientific discoveries.But on the ground and in the practical application of artificial intelligence, scientists do not commit themselves to pure models of human decision-making, learning and problem-solving, and instead work to build as practical a system as possible.
For example, the basis for significant advances in algorithms - resulting in deep learning systems - is a technique called back propagation, which does not function as the brain works.
This leads us to the following misconception:
2 - Artificial intelligence has a solution to all problems
Another misconception is the ability of artificial intelligence to solve all existing problems. I've heard some talk how moving from one problem to another makes the AI system smarter as if the AI system itself solves both problems at the same time.The reality is quite different: artificial intelligence systems sometimes need to plan deeply and specially trained models to apply them to a problem.
Although similar tasks such as voice recognition, image or video now have a library of reference templates, they need to be specifically tailored to meet usage requirements.
What's more, AI systems are not the only component of any intelligence-based solution. They often require many specially programmed components to fit together to combine one or more AI technologies.
Of course, there are a number of different AI technologies that are used alone or in combination so it is not correct to say:
3. Artificial intelligence is the same as deep learning
In the past we thought that the term neural networks was beautiful and there were many hopes that bet on its potential until its weakness in standardization and proportionality to the task.Now that these problems have been solved, the term has been replaced by the name Deep Learning. Deep refers to the number of hidden layers that we can place within the neural network.
While learning refers to the generation of models not in real time, but in the form of online and this needs time and processing is large and difficult to achieve in parallel.
Recently, deep learning models have been used in online learning applications where online learning is implemented using various AI technologies such as enhanced learning.
The only disadvantage of these systems is that deep learning models can only be used if the field used can be tried within the online learning period.
Once the model is generated, it stays the same and will not be flexible for changes in the field of application. A good example is ecommerce applications where seasonal changes or short sales periods on ecommerce sites require taking an inline deep learning model to retrain it on new products.
Deep learning systems are often supported by huge data sets, so a concept has emerged that new and useful models are generated from a unique and huge data set, supporting the misconception that…
4. Depends on big data
But this is not true because the greater reliance on good data. Large, inaccurate data in one area can cause system-building errors. Also, in many areas, old data quickly becomes insignificant. For example, in the New York Stock Exchange, recent data is more important and more valuable than, for example, data from 2001.Finally the last misconception: