![image.png](https://images.hive.blog/DQmcnVZieB98DEy2MB9ddCxRwk2rAtGj6Pr8kgoxMDPcvZo/image.png)

Artificial intelligence-based applications such as ChatGPT, Dall-e, and Midjourney have become popular recently. The rapid development we witness makes us wonder what artificial intelligence can do in the future. To interpret the potential of artificial intelligence correctly, we should familiarize ourselves with the basic concepts used in this field.

### **Intelligence**

I want to define intelligence first. Let's start with the dictionary definition: The ability to acquire and apply knowledge and skills. In other words, intelligence is the ability to learn and put what we have learned into practice. For those looking for a shorter definition: Intelligence is the ability to predict. We need intelligence even to drink a glass of water.

Intelligence is traditionally associated with the ability to think abstractly. Robotics studies reveal that moving the body and interacting with surrounding objects require considerable cognitive skills. We also need social intelligence to communicate with other intelligent beings. The weakest aspect of artificial intelligence is its social intelligence. However, we will undoubtedly witness progress in this area over time.

### **Artificial Intelligence**

IBM defines artificial intelligence as machines mimicking human intelligence. How can we judge that an AI agent is intelligent? It would be right to accept that the agents who create the impression of being clever are intelligent. But does an intelligent agent have to be conscious to be considered intelligent? I believe what matters is the social role. Since I want to keep the content simple, I will refrain from entering the discussions about consciousness in this article.

So how does artificial intelligence work? Systems that we call AI today can accomplish complex tasks with pre-written codes. Programmers can write the code of an AI agent by hand or use machine learning methods to generate the code. Navigation applications are an example of AI in which codes are written by hand. Translation programs, speech-to-text, and text-to-speech systems are created using machine learning. Likewise, chat programs and programs that produce images.

### **Types of Artificial Intelligence**

There are three types of artificial intelligence: narrow, general, and super. Systems specialized in a particular field are called narrow artificial intelligence. Systems that can function in multiple ways similar to humans are called general artificial intelligence. Artificial intelligence has yet to reach this maturity stage. Futurist Ray Kurzweil predicted twenty years ago that by 2028, artificial intelligence would reach a cognitive level equivalent to a human. Kurzweil said last year that his prediction is still valid. 

For now, we only encounter super artificial intelligence in science fiction movies. For some reason, they are constantly behaving like a devil.

### **Machine Learning**

Problems such as translation, handwriting reading, and object recognition were solved with machine learning methods in the 2010s. Previously, these problems were tried to be solved using expert systems, but the desired success could not be achieved. Programmers and experts worked together to create expert systems. For example, to solve the translation problem, software developers tried to manually write an algorithm that would translate from one language to another with the support of linguists.

Machine learning models are programs that write programs. They use relationships between data to write programs. These programs examine the data and detect regularities in it. Machine learning models correspond to universals in philosophy. However, these universals are not ideal forms as Plato imagined. Aristotle's definition of the universal starting from concrete entities is more appropriate to the context. We learn the characteristics of a cat as children. On the other hand, machine learning models create the cat's universal features by comparing millions of images.

So how does learning take place? Machine learning algorithms find parameters that describe regularities in data by trial and error. After each test, the machine learning algorithm calculates the error amount. As the number of trials increases, the error decreases, and the resulting model explains the input data better.

In my post titled [Articial Intelligence - A Brief Introduction](https://peakd.com/hive-175254/@muratkbesiroglu/artificial-intelligence-a-brief-introduction), I explained with an example of how machine learning models learn. If the subject interests you, I suggest you also read that article.

### **Deep Learning**

Artificial neural networks have a history dating back to the 1940s. Artificial neural networks, one of the machine learning algorithms, created great excitement in people from their early stages. Because although we do not know precisely how the human brain works, we do know that the networks formed by neurons are effective in learning. On the other hand, those who deal with machine learning have experienced that neural networks require a lot of processing power. It took the 2010s for artificial neural networks to produce effective results.

The layered application of artificial neural networks is called deep learning. Deep learning has the merit of discovering hierarchies within data. For example, when we look at a book page, we do not see randomly scattered letters. Thanks to the education we have received, we are aware of the letter-word-sentence-paragraph hierarchy. Deep neural networks also learn these hierarchies by examining data. Object recognition models use millions of images, each labeled with words. Thus, for example, it is possible to know that a forest has a landscape consisting of soil, trees, and animals.

A considerable amount of data is required to train deep-learning models. The use of video cards for learning is also one factor that makes deep learning possible. The most popular applications of artificial intelligence are made possible by deep learning.

### **Components of Machine Learning**

The systems we call artificial intelligence today consist of a combination of handwritten codes and codes created through machine learning. As the scenarios that the AI agent encounters become more complex, the effectiveness of handwritten code decreases. Returning to the translation example, it can take years to write a program to translate a text from one language into another. However, we can produce a deep-learning model in a few weeks.

We need processors, data, and algorithms to operate machine learning models. As I mentioned above, discovering the regularities in the data is done by trial and error. So we must have powerful processors accompanied by a large amount of memory. On the other hand, the data we need grows in parallel with the complexity of the model we are trying to produce. Finally, the algorithm affects the efficiency and quality of the learning process.

### **The Future of Artificial Intelligence**

So to produce more proficient AI agents, we need more powerful computers, sophisticated algorithms, and vast chunks of data. In addition, there is a need for qualified human resources who can use this material efficiently.

In the future, we will ask AI agents to take more initiative. Because the less they are subject to human manipulation, the more value they will produce. That means they will need to interact with other AI agents and humans. How will they know how dominantly they will behave during this interaction? Their social intelligence will need to develop to make such decisions. They may learn by trial and error the limits that we shouldn't cross as we did when we were kids.

As computing costs decrease over the years, data in the digital environment will continue to grow. As we saw in the ChatGPT example, artificial intelligence agents already have a general knowledge of the world. The information in question will inevitably become more detailed over the years.

### **The Relation of Artificial Intelligence to Finance and Crypto**

Artificial intelligence has been actively used in finance since the 2000s. For example, whether a retail customer can pay the loan to be allocated is determined through machine learning models. The content of the marketing communication to be made with the customers is also determined through machine learning. Which financial transactions are suspicious are determined through models. On the other hand, machine learning is intensively used in portfolio management and trading.

The use of artificial intelligence on crypto networks is still in its infancy. In crypto, too, it is possible to predict which transactions are suspicious through machine learning. No blockchain is taking advantage of this possibility yet. There are projects to use artificial intelligence for digital trading assets.

The popularity of applications such as ChatGPT and Dall-e has enabled crypto projects operating in artificial intelligence to gain value. In the last 30 days, Singularity.net has gained 520%, Fetch.ai has gained 182%, and The Graph has gained 147%.

### **Artificial Intelligence Agenda for 2023**

ChatGPT4 is expected to launch in the coming months. It's possible that Chatgpt4, trained with a larger dataset, is creating a new wave of hype.

Three days ago, Sundar Pichai, CEO of Google and Alphabet, blog titled [An Important Next Step On Our AI Journey](https://blog.google/technology/ai/bard-google-ai-search-updates/) published the post. The blog post was about Google's new conversational chatbot. There was also this sentence in the blog post that @taskmaster4450 might like: **Today, the scale of the largest AI computations doubles every six months, far outpacing Moore's Law.** One of the countless examples of the exponential evolution of technology.

Home robots will undoubtedly be one of the most important practical AI applications. Last October, Elon Musk introduced his home robot project Optimus. There will likely be new developments about Optimus this year. Many other companies, such as Atlas, produce humanoid robots too.

Self-driving cars, which have been on the agenda for many years, can be used more in the USA this year. Do you remember San Francisco Streets TV series? We used to watch the chase scenes on the bumpy roads. Self-driving cars are expected to appear on those roads this year. In addition, self-driving vehicles may be approved in states other than California.

### **Challenges With AI**

AI also brings problems to humanity in many areas. For example, AI systems can make biased decisions because of the data they feed. Maybe you remember the news about Facebook's artificial intelligence learning about racism and discrimination from users. Just as it is essential who our children make friends with, it is also vital that artificial intelligence agents are fed with healthy information.

Artificial intelligence systems have difficulty explaining the reasons for their decisions because they involve too many parameters and result from a complex calculation process. In addition, new-generation applications such as ChatGPT can pretend to know about things they do not know. Moreover, the fabricated information they produce is persuasive.

### **Limits of Artificial Intelligence**

Electronic components such as processors and memory are getting smaller over the years. Thus, faster and higher-capacity computers can be produced. However, uncertainty arises when it goes below a specific size. On the other hand, while the unit transaction cost decreases exponentially, energy consumption does not fall at the same rate. Considering global warming, this is a significant problem. Moving servers into space or manufacturing them there could be a long-term solution.

The volume of quality data to feed artificial intelligence is still largely exploited. For the further development of artificial intelligence systems, it will be necessary to make better use of existing data or to add some creativity to work.

Jumping from general AI to super AI may be more challenging than in the movies. Because the more sophisticated something becomes, the harder it is to operate it. In addition, it may be tough to imitate or surpass the human mind when considering its emotional and social dimensions.

Digitization is a trend that fuels artificial intelligence. On the other hand, it is easy to predict that there will be political problems regarding adopting artificial intelligence. AI agents will need to be autonomous to be more efficient. It will even come to the fore that they have certain political rights. It seems inevitable that this situation will create tremendous tension.

### **Conclusion**

Artificial intelligence may be humanity's most important invention. Likewise, for the first time, technology is a candidate to replace human beings. I just requested Chatgpt: "I want to have a basic understanding of AI in 2000 words". I was worried that the app would have done better than me because I had been writing this blog post for three days. I also have experience in this field. Fortunately, the text Chatgpt created was not comparable to mine. On the other hand, ChatGPT's text inspired me to add the title "Challenges With AI." Nevertheless, the age of artificial intelligence has not started yet. We are in the human-machine cooperation era for now.

Thank you for reading.

**Image Source:** Midjourney App, Prompt: AI, cute, detailed




Posted Using [LeoFinance <sup>Beta</sup>](https://leofinance.io/@muratkbesiroglu/let-s-have-a-basic-understanding-of-ai-in-10-minutes)