thoughts about the inevitable – how neural networks will change our work and who will have to look for a new one

For the past 12 months, the world media has been full of messages about the possibilities of Chat GPT. The neural network successfully passes tests, exams and technical interviews for a number of professions, copes with typical tasks for SEO copywriters, technical writers, and analysts. In particular, the neural network recently successfully completed a technical interview for Google applicants, and also passed a medical exam. Representatives of many professions, including in IT, had bad thoughts about the need to change their profession in the future and that Open AI did the same as the young lawyer from the button accordion joke. I will try to express my thoughts on this matter, as a person who often uses Chat GPT and as the head of the technological solutions practice of the EAE-Consult system integrator.

Image generated by Openjourney using the Stable Diffusion model
Image generated by Openjourney using the Stable Diffusion model

If you look at the situation soberly

So far, at least in the next 7-10 years, Chat GPT will not be able to fully replace programmers. Yes – the neural network can write code. She can set certain tasks and she forms a solution in the form of a code. Moreover, he can do this in Python, Java, Javascript, assembler, and even in various exotic languages ​​\u200b\u200blike Brainfuck. And already now, a neural network operating in test mode can be instructed to write basic things.

For example, you can delegate small games to the bot if you set the task correctly, i.e. explain what should be in the input and what should happen. Instruct some little logic. And the system will write this code, it will even somehow work. Perhaps some decisions will be quite a challenge for a person. But it is important to understand that at the level of solving a business problem, this is just a piece of code that can help a programmer or business analyst speed up the solution of a problem, but not do everything for him.

Often the system does this in different ways and not always fully, but the correctness of the answers is a matter of a short time. So far, ChatGPT can be called rather an assistant that speeds up work. She can help find errors in the code that she was “fed”, she can suggest a non-trivial solution, she can suggest how to write something. At the same time, it is not a fact that the hint will be correct and that there will be every single bug. Also, the system can create detailed documentation for the code with a description in human language about which part is used for what.

Thus, ChatGPT can now replace some simple autotests and speed up the work of developers by suggesting solutions that would take days or even weeks of work to find without it. To exaggerate, ChatGPT is already able to deprive a coder-june of his job, but the system is still far from being a middle, and even more so, a wise architect.

Platform knowledge problem

When creating a commercial product, it is important to know not only the language, but also the platform. ChatGPT knows far from all platforms, for example, if you take the Russian nocode and locode platform elma 365, then a request to the neural network to write an autotest system for such a platform will not work. Because the system doesn’t know how to code for that platform. Long training is required, and due to the fact that this platform is hardly a priority for Open AI, it is unlikely that the neural network will master this ability in the near future.

In commercial solutions and system integration
When we talk about large enterprises, especially in Russia, as a rule, applications are solved within some platform, such as SAP or 1. In order to write code for them, you need to know how they work. In such cases, ChatGPT, in principle, will not be able to replace the programmer. Rather be able to write basic code, but not make it a production application. Those. again, we are talking about the fact that the neural network will be able to facilitate the work of the programmer, but not replace him.

Meanwhile, in our field there are specialists who can already be called representatives of gradually dying out professions. For example, technical writers. So, ChatGPT is already coping with writing technical specifications in accordance with GOST and, in general, is able to generate fairly decent project documentation. In addition, the neural network will take away the work of a significant number of business and system analysts, in particular, from specialists who are mainly involved in formalizing requirements according to ready-made templates of analysis artifacts. And those analysts who will remain in the profession, i.e. capable of identifying and generating requirements, on the contrary, can significantly increase productivity by delegating the routine tasks of formalizing ChatGPT. A good example of using a neural network to generate technical specifications is the video of Andrey Kupriyanov, an analyst:

Obviously, specialists of call centers and customer support services working on template scripts can gradually look for a new job. My observations while working with ChatGPT show that the possibilities of targeted communication on typical customer issues are solved by the neural network more efficiently than by low-skilled girls in an outsourcing call center.

In the distant future, 10 years or more, ChatGPT or similar systems will most likely become a kind of neural network orchestrators that will be able to combine automated development functions of at least simple products into a single process. And only after this point, IT professionals in most specialties should worry about changing jobs. For example, if you need a website, Mid Journey will take over the visual part of the product, some other AI will create a front based on it (while the Open AI product does not cope with this very successfully), and ChatGPT will write the basic code, texts and in general generate data for process control.

Futuristic forecasts

If you look a little further, at the prospect of 10 – 20 years, you can see how rapidly ChatGPT is developing. She will be constantly “fed” with more and more new platforms, examples of good solutions and well-written code, examples for specific business problems. As a result, this will bring the system to a state where it will be able to cope with increasingly non-trivial tasks. At some point in time, the neural network will be able to generate the base code directly according to the terms of reference, without a lot of other additional conditions that are needed now. I assume that such a task will have to contain inputs, outputs, expected results and data types.

Obviously, in the future, with such a technical specification, it will be possible to obtain ready-made pieces of code, full-fledged modules that can be combined into a full-fledged commercial product. But the process of such a combination will still require the participation of a programmer. I believe that this is a fairly soon future, because. purely technical tasks, where empathy is not required from the developer, taking into account ethical criteria, will be perfectly solved by AI in 2-3 years. But, of course, a neural network capable of this cannot replace human intelligence in product development.

In the near future, neither ChatGPT nor any other neural network will be able to fully understand the task that needs to be solved, and, accordingly, create code based on even detailed abstract tasks. In other words, the creation of a product will be at least tied to the work of architects, programmers and analysts. At the same time, in the longer term, say, the next 5-7 years, depending on the intensity and quality of training, the neural network may, with the right request, be able to create simple turnkey commercial products on its own according to user template requests. Probably, its capabilities in generating such products will, however, be more modest than those of living developers.

New professions

ChatGPT will not only take jobs away from people, but will create a new market for specialists. Once upon a time, lowcode and nocode appeared, and today they have already begun to grow into entire teams that work with platforms, mastering them much faster than classical programming languages. A new formation of IT professions in nosode will be ChatGPT query specialists and neural network trainers. Obviously, many of the really valuable professionals who will “kill” neural networks will become their trainers. To do this, it will only be necessary to master the methodologies for transferring experience.

Doctors have nothing to worry about

The sensational passing of a medical exam by a neural network is unlikely to threaten the profession in any foreseeable future. As in the case of programmers, the neural network, on the contrary, still claims only the role of an assistant helping the practitioner make the right decision and preventing possible medical errors.

The degree of risk, legal restrictions and issues of personal responsibility will keep medicine conservative enough for a long time to prevent ChatGPT from making independent decisions regarding the health of living people.

This is exactly the case when the risks associated with the Lem barrier become incomparable with the possible benefits. Meanwhile, it is already clear that in the hands of a good doctor, ChatGPT will become a tool for minimizing the human factor in the diagnosis, choice of treatment tactics and assessment of changes in the state. It is important, however, that those who believe in the power of AI do not self-medicate and ignore traditional doctors, so the medical background of AI requires ethical regulation.

Humanization cost

When it comes to such bold assumptions about the future of AI, there are always questions about the cost of modeling a humanoid approach when training a neural network. Recreating such products of the human psyche as empathy, a kind of imagination and humor, as well as everything that we refer to as soft skills, requires a truly cosmic amount of data. For individual, even large companies, the cost of training can be incomparable with the profit from replacing human functions. Available resources exist only in developed countries, transnational corporations, as well as large companies with huge amounts of data. At the same time, it can hardly be considered that such an expensive solution can exist only as a product for internal use. Such a development will be profitable only if it is offered on the global market.

Open AI is counting on large-scale development. Now they are raking data from all available sources and feeding it to ChatGPT, while hundreds if not thousands of sad-faced employees from a poor African country sit and turn the chatbot into a “neural network with a human face”, preventing the appearance of what humanity fears with the appearance of the Terminator film franchise. And a commercial solution that uses these solutions, apparently, will be offered by Microsoft. If we talk about Russia, such things can be done, for example, by Sberbank, because they have an incredible amount of data, Yandex, a large federal telecom operator and a large provider. Other companies will obviously not be able to afford emulation of humanoid functions and soft skills in any foreseeable future.

A little about Copyright

ChatGPT – a model trained to predict the next word, many terabytes of code from Reddit, GitHub, Quora were used in training. The question arises – can AI-generated code be considered copyright-free? After all, to one degree or another, he will repeat someone else’s code. The corporation that owns the rights to this code, within the framework of the laws of its jurisdiction, can begin to protect its right to this code. Request changes, compensation for lost profits or punishment in the form of fines.

The question arises: how can the situation in the legal field be resolved? After all, a programmer who will use parts of the code generated by ChatGPT can thus substitute his company under the sword of Themis, punishing for copyright infringement. It will be difficult for a company in such a situation to prove that the code was generated by ChatGPT, that the copyright was not intentionally infringed. To many, such a scenario may seem absurd, but in the reality of corporate disputes, claims of authorship of the code are not uncommon, suffice it to recall the Niantic litigation and the Wargaming conflict with the Ex-development team, the Press Fire Games studio (better known as BlitzTeam).

As a conclusion

Summarizing what has been written, we can conclude that the programmer has nothing to fear so far, since all long-term forecasts involving the replacement of middle-level specialists with a neural network do not have a clear time horizon and are expected no earlier than in 10 years. So far, technical writers, low-skilled juniors, novice analysts and first-line customer support fighters are at risk. All owners of professions that require systems thinking, empathy, imagination and creativity, i.e. where we are talking about a creative approach and properties characteristic of a living person, it is too early to worry.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *