Can AI replace IT experts?

Neural networks and artificial intelligence are becoming more advanced and receive new functions. The very first versions could identify a song name and write a response to a user’s simple question. Now, ChatGPT and similar systems can write code, and this caused panic in the IT community. Participants in professional forums discuss whether AI can replace programmers. We will try to answer this question in this article.

What can neural networks do

In programming, there are standardized rules that govern a programming language, and artificial intelligence can easily learn them. At the same time, there is much documentation and examples of code to learn and improve one’s skills. Due to the typical writing and big amount of data, neural networks can successfully deal with the following tasks:

Code generation. User describes a function, program or website and the neural network creates working software based on a prompt.
Code completion. AI completes code in accordance with a text description and with regard to the context.
Code refactoring. AI analyzes code and improves its structure.
Optimization. The neural network checks the speed and performance of the program and improves it.
Debugging. AI launches a program, finds bugs and suggests ways to fix them.

What are the disadvantages of neural networks?

As regards programming, the main disadvantage of artificial intelligence is inaccuracy. The current neural networks give more accurate answers than they did two or three years ago. However, these systems can still make mistakes, and there are not enough tools to verify the information yet, as well as it is impossible to warn the user that the result might be wrong.

Here’s an example of a case when the neural network’s mistake cost a business dearly: at the presentation of the Bard neural network, the chatbot said that James Webb Space Telescope was first to take pictures of planets outside the solar system, which is incorrect. The very first image of an exoplanet was taken by a group of researchers who did not use James Webb.

The mistake was noticed both by users and investors. Google share prices dropped nearly 10%, losing about $100 bln in market value.

In programming, small details and nuances have a great importance. A small mistake can lead to critical failures and temporary shutdown of an online service or program sale. Even though neural networks attract the attention of business, companies are not ready to take such risks yet.

The second issue with AI is reproducibility. Programming is not just about writing code and assembling programs from readymade fragments. Developers have to work out the logic of software performance, which current neural networks cannot do.

How neural networks are used in IT

Artificial intelligence has been already adopted by the industry, but it is too soon to speak about mass layoffs of programmers. Neural networks help employees by automating routine actions and make major tasks simpler. Here are several examples:

Big data analytics. Unlike people, computer can perform millions of small actions per second. That is why recommendation engines are created using neural networks. They analyze information, detect complex dependencies and summarize results.
Generating typical programs and functions. IT engineers use them to create fragments of code. They do not need to waste time on routine actions or delegate them to colleagues because neural networks is perfectly capable of dealing with simple tasks.
Monitoring and diagnostics. A system administrator can use a neural network to maintain an infrastructure designed for 1000+ workstations, with AI checking logs, monitoring employee computer activity, and performing other tasks.

In the future, AI applications will expand. It is believed that programming may change beyond recognition soon.

With the advent of computers, developers used to work in an assembly language, later switching to high-level programming languages such as FORTRAN, LISP and COBOL. Currently, the user-friendly Python has become the most common one. Experts suggest that neural networks will be able to create an even more ‘humane’ language. Just like the fictional character J.A.R.V.I.S., AI will translate simple words into a computer code.

In this case, programmers will not have to spend too much time memorizing syntax and code, with understanding program logic and the ability to write correct prompts for neural networks to become major tasks. Web reports are emerging that mention a promising profession – a prompt engineer who can communicate with a neural network and receive answers to any questions.

Types of neural networks used in IT

Here are the tools programmers use to automate and speed up the work:

A popular bot for code analysis and error correction.
A tool that uses the context to predict further lines and supports all popular programming languages, including Python, Java and C++.
A multifunctional assistant that can generate a code using a text description, perform program reviews, and correct errors.
A tool that simplifies back-end development.
A tool that generates a code using text queries and performs code refactoring.
Buildt AI, a search engine for VS Code that can search for ready-made code snippets in open databases.

Most high-quality tools are paid, but there are also free alternatives which have limited functions, lack support, and provide lower quality answers.

What does the future of IT hold?

OpenAI has openly claimed during news conferences that it seeks to make ChatGPT a strong competitor to developers. This has led to some forecasts saying the demand for professional programmers will decrease and companies will quit competing for specialists, which will affect labor conditions and wages.

Technically, developers can be replaced – but only partially. Unless AI learns to think like humans or outsmarts them, it will be unable to independently manage projects.

Iain Smith, director of Diaz Research, is skeptical about this stance. The expert believes that the developer job will drastically change, claiming that AI tools will make work process easier, reduce the amount of routine tasks, and allow for the development of new programming languages.

Here are two more predictions about the future of neural networks:

IT will reach a new level of automation, with one professional to perform a task that previously required five employees. Developers will continue to be in demand – however, the number of projects carried out by one worker may increase exponentially. We will also see new jobs appear; specialists who will learn working with AI will remain relevant.
Programmers will be literally the last profession to be replaced by AI. When AI becomes autonomous, the employment issue will be the last thing for us to worry about. Experts say that so far the only self-sufficient superintelligence system capable of developing and learning independently is the notorious Skynet of the Terminator franchise – and AI will definitely not be developed to reach this level in the years to come.