According to predictions, one of the jobs most affected by AI is application (APP) developers. After all, AI models like ChatGPT are experts in language manipulation, so language-based careers — including computer languages — would suffer. But the application developer role may change differently than you might think.
Predictions about the impact of AI on the programming industry vary widely. On the one hand, some predict that developers will be completely eliminated, while others are convinced that AI will turn existing programmers into "superbrains."
If you are worried that AI will replace your programming job, it is likely to already be a reality. However, today's AI has a major limitation, which is that it will require human developers to be involved for a while (perhaps forever) because of creativity, which robots have not yet had.
"We haven't developed true AI as people usually think." Vaso Peras Likodrić, senior software developer at artificial intelligence automation company ABBYY, said. "We have predictive neural networks that simulate inference by output, but the essence of accurate (human) intelligence should encompass more."
The data suggests that we are still in the early stages of AI-assisted programming. ChatGPT has only been available for six months, while GitHub Copilot has only been out for about 18 months. Using these tools to improve efficiency requires more than just signing up for an account.
Another interesting finding is that these AI tools are more beneficial for junior or advanced programmers. The survey data is inconclusive: 13 percent say they are useful for advanced programmers, 11 percent say they are useful for junior programmers, and 11 percent say they are not useful to anyone and do not increase productivity at all.
The prevailing view is that Copilot is more advantageous for advanced programmers, who are using it to help design and build systems and even create production code. They are better able to describe the problem they need to solve in a complex set of prompts and quickly notice errors in the generated code. And junior developers are using it to learn and reduce the time to solving problems by finding solutions on Stack Overflow or searching online documentation.
Today's AI models are able to predict the next word or action based on large amounts of human-generated training data. Herein lies the problem: AI models can't know anything that hasn't happened yet. Therefore, it is unlikely that original ideas or ideas that were not previously thought of will appear in the results of AI. Simply put, there is a lot of room for human creativity, and these tools can't fully encapsulate that space.
Peras Likodrić also noted that while AI models "primarily produce satisfactory results," their tendency to produce confusion is another factor that may leave the human hand on the keyboard for a while.
Peras Likodrić said: "Even with state-of-the-art systems, there can be situations where AI is 'out of order' or the output is inaccurate. "That's why human verification remains critical for all the critical tasks that are integral to accuracy. For example, the validation of ABBYY's products emphasized the importance of human factors. While AI tools can speed up this process, in their current state, they can only be partially relied on.
So, where will AI go? Vladimir Khil, head of ABBYY's platform processing group, said current AI can stand out in many other application development activities.
"Imagine having a private version of an AI engine that analyzes our code, provides recommendations on coding styles and efficient algorithms, and even finds unreadable code without any privacy concerns associated with third-party technology," he said. This will be a game changer! ”
AI can also help with test scripting. "Writing tests can be boring for developers," Khil says. But with the power of artificial intelligence, we can describe interfaces and scenarios, and it can generate test code effortlessly. ”
A new report by O'Reilly on trends in low-code and no-code app development reveals the use of tools like ChatGPT and GitHub Copilot, which are growing in popularity among programmers.
Perhaps the biggest finding of the survey of 739 organizations was that only 33 percent reported using such tools, while 67 percent reported not using them. But O'Reilly's folks were skeptical of the results. Mike Loukides, of O'Reilly, wrote in the report: "We suspect underestimating the actual use of Copilot. ”
That number is likely to be much higher, just as the use of open source software is much higher than it was 20 years ago when O'Reilly surveyed CIOs. "They know very little!" Loukides wrote. "The actual usage of Copilot, ChatGPT and similar tools is probably much higher than 33%. We are sure that even if they are not using Copilot or ChatGPT in their work, many programmers are experimenting with these tools or using them in personal projects. ”
Interestingly, these tools are not as easy to use as expected. The report found that training was the biggest difficulty, with 34 percent saying training was their biggest difficulty. This was followed by ease of use, which 12% chosen.
Loukides wrote: "This was a surprise because many of these tools should be low-code, or no code." "We specifically consider tools like GitHub Copilot, Amazon CodeWhisperer, and other code generators, but almost all productivity tools claim to make life easier. At least at first, this is clearly not true. It's a learning curve, and it seems steeper than we thought. ”