“In short term, artificial intelligence will not be able to completely replace editors and journalists,” said new media specialist Rustam Gulov. “However, neural networks can already become an effective assistant for content creators.”
The development of artificial intelligence (AI) over the past few years is both impressive and frightening. While students at some universities in Tajikistan, who only yesterday were googling materials for their term papers and theses, are already switching to using ChatGPT, journalists are bypassing this topic and are reluctant to master the relatively new technology. Although they, like all representatives of creative professions, should be wary: will neural network-based services be able to take away their jobs? Or is this an overly optimistic view of neural networks, and do they still have room to grow?
To study the effectiveness of AI in practice, we decided to ask three neural networks to develop an outline for an analytical paper on “How generative neural networks will affect media in Tajikistan”. For this purpose, we chose two versions of ChatGPT – free (based on the GPT-3.5 model) and paid (ChatGPT Plus) (based on the GPT-4 model), as well as chat in the search engine Bing.com, which uses a similar BLM (Bing Language Model) to GPT (Generative Pre-Trained Transformer).
There is no point in asking to write the whole article, because all the services used have limitations in the length of responses. That said, Chat of the search engine Bing.com (hereafter Bing) reported that he has problems with writing the text of the article himself.
“This is too complicated and time-consuming a task for Chat mode. I can help you with rewriting, improving or optimizing your own text if you want. You can send me your text or part of it and I will try to make it better,” Bing clarified.
Article plan
All three neural networks showed different results when creating an outline for a future article. Bing created a quite dense outline with a detailed description of what should be covered in each part, but aesthetically the outline looks a bit confusing.
Bing’s response:
Hello, this is Bing. I can help you with writing an analytical paper on “how
generative neural networks will affect media in Tajikistan.” Here is a possible outline of such
article:
- Introduction: what generative neural networks are and how they work. Examples of their application in
different areas of media such as image, video, text and music creation. Pros and
pros and cons of using generative neural networks for media content.
- Main part: analysis of the current state and prospects of media development in Tajikistan.
What factors influence the decline in the audience of traditional media and people’s transition to the network. What needs and interests the audience of media in Tajikistan has and how generative neural networks can satisfy them. What risks and challenges accompany the use of generative neural networks for media, such as ethical, legal, technical and qualitative.
- Conclusion: summarizing and concluding the topic of the article. How generative neural networks can affect media in Tajikistan in the near and far future. What recommendations
can be given to media workers, consumers and regulators in connection with the emergence and spread of generative neural networks.
Bing also suggested that it would be necessary to conduct research and collect data in order to write the article.
ChatGPT and ChatGPT Plus have developed a similar structure for the article, which consists of a short introduction, an analysis of the current situation, a description of the possible impact of the use of neural networks on media in Tajikistan and the risks and threats that accompany the process of their implementation. The full text of the ChatGPT-generated article can be found here.
What is different about ChatGPT Plus is that it suggested considering real projects of local organizations in Tajikistan that could potentially use neural networks in their activities.
ChatGPT Plus Response:
- Introduction
- Topic Overview: Description of generative neural networks and their capabilities.
- Purpose of the article: Studying the influence of generative neural networks on media in Tajikistan.
- What are generative neural networks?
- Description of the basics of generative neural networks and their functionality.
- Examples of using generative neural networks in the world.
- Current state of media in Tajikistan
- Analysis of the current situation and trends in the media in Tajikistan.
- Main problems and challenges faced by media in Tajikistan.
- Possible impact of generative neural networks on media in Tajikistan
- How generative neural networks can help to overcome existing problems.
- Possible new opportunities and directions of development for media, which are opened by generative neural networks.
- Consideration of possible risks and problems associated with the use of generative neural networks in the media. (e.g., ethical issues, copyright issues, data security).
- Examples of application of generative neural networks in the media of Tajikistan
- Cases or planned projects related to the use of generative neural networks in the media of Tajikistan.
- Conclusion
- A summary of the main conclusions and possible future forecasts.
- Tips and recommendations for Tajikistan’s media on the use of generative neural networks.
- Bibliography/Sources
This is less about actual practice and more about reviewing the potential for neural networks to be used by journalists and other content producers in practice.
Overall, all three plans consisted of three main parts:
Introduction
The main part, which included analytics on the status quo
Conclusions and recommendations
What does AI suggest to start writing a piece?
All three introductions are, in general, very similar to each other, as they create a playing field for the reader to properly navigate further in the text. ChatGPT gave a very similar introduction text in both of its versions. The shortest was the text given by ChatGPT Plus – it consists of only 158 words, whereas ChatGPT (free version) generated almost twice as much text – 258 words. However, the text of the introduction from Bing differs from them not only quantitatively – it is objectively longer (439 words), but also qualitatively – it consists not only of abstract sentences, but also of concrete examples of spheres of neuronics use in media. For example, Bing suggests that generative neural networks can be used by media for such tasks “as synthesizing the faces of celebrities or non-existent characters, restoring old photos or videos, translating images from one style to another, generating texts on a given topic or style, creating musical compositions or sound effects.”
In terms of producing the most interesting and widely understandable content, this approach may be more effective because it creates relatively real images and emotional connections in the reader through concrete examples.
In addition, Bing preferred a scientific approach to writing an analytical article on a given topic. This also influenced the length of the introduction he wrote as he described the intended research methodology as well. Bing assumed that the author of the paper would want to conduct a full-fledged data collection through literature review, statistics, and interviewing the professional community.
The main part of the article is an overview of the situation and analytics
Both versions of ChatGPT started the main part of the article with the definition of the concept of “generative neural networks”. Moreover, they allocated separate chapters for this purpose, about a page long. For a reader of such articles this is a lot, because often they are already aware of these concepts and are unlikely to spend time on reading such texts. And the description of the technology includes relative details about their work, which will definitely not be interesting for the audience of content producers. They are interested in how to use neural networks in practice, not how exactly these neurons work.
From this point of view, Bing again produced a more appropriate result. Since he had already mentioned the meanings of the main terms in the introduction he generated, he started the main part of the article with a description of the situation in Tajikistan. Moreover, this description, in comparison to the one given by ChatGPT, is the most reasonable and relevant to the media situation in Tajikistan from the point of view of potential.
For example, ChatGPT (GPT-3.5), when describing the situation in Tajikistan, generated a very superficial text using rather abstract expressions. Here is one paragraph as an example:
However, recent years have seen the growth of online platforms and social networks in Tajikistan, which creates new opportunities for media. The Internet allows a wide range of users to obtain information, exchange opinions and participate in discussions. Many media organizations are developing their online presence and using social media to disseminate their content and interact with audiences.
The ChatGPT Plus text (GPT-4) appeared to be more oriented towards discussing the challenges faced by media in Tajikistan. At the beginning of the text, this model was also quite lengthy in describing the situation, but still identified 5 main problems of media in Tajikistan separately in the form of a numbered list:
Limited access to the Internet;
Lack of professional skills;
Limited resources;
Problems of press freedom;
Lack of original content.
When analyzing the text, what catches your eye is the lack of facts. For such cases, ChatGPT Plus, unlike its free version, has the option of linking to the Bing.com search service so that ChatGPT can collect and process the most relevant data available on the web (at this particular moment, due to malfunctions, this plugin is temporarily disabled, but may return again in the near future). In this case, it generated a completely different text describing the situation in the country. The new text analyzing the situation was almost entirely devoted to the criminal prosecution of journalists in Tajikistan in recent years.
Bing, in contrast to ChatGPT, described the media situation in Tajikistan in a slightly different way, shifting the emphasis more to describing the media landscape of the country. This text turned out to be the most factual, but not the most relevant, as the data found concerned the statistics of different types of media in the country for 2019-2020.
It is worth noting here that Bing can operate on data that is freely available online. Therefore, it may not always have up-to-date data for our country at its disposal. Manual input of data for analysis and inclusion in the text of the article can remedy the situation.
Risks and threats
Speaking about the problems of media, in particular media in Tajikistan, the resources used paid much attention to risks and threats. In particular, according to the “opinion” of ChatGPT there are 5 main types of risks associated with the use of neural networks in media:
- Ethical and legal issues;
- Content quality and credibility;
- Dependence on technology;
- Loss of jobs;
- Dependence on data and algorithms.
ChatGPT Plus said pretty much the same thing, but in a narrative form, without being as point-by-point as its free version.
In turn, Bing again gave a more “thoughtful” and detailed version. It didn’t just list all the risks and threats, but organized them into 4 groups:
- Ethical risks and challenges;
- Legal risks and challenges;
- Technical risks and challenges;
- Qualitative risks and challenges.
- In this respect, Bing’s results look the most professional, balanced and weighted.
Conclusions
Having received three relatively different texts on the same topic, we can draw some conclusions. We will conditionally divide them into “pluses” and “minuses”. Among the disadvantages we can note such as:
- ChatGPT, both in the paid and free version, has limitations in the data used and does not know what happened after September 2021. ChatGPT Plus can try to change this situation through linking to the search engine Bing, but practice shows that even in this case it does not change the result of its work.
- Bing, by virtue of the fact that it uses data from the web and generates texts based on it, can find and operate with more relevant data. In addition, it relies more on factual data and provides references to the sources used. But, if the actual data is not freely available on the web, Bing will not be able to use it either.
- Not all facts produced by neural networks are true. Neurons can “hallucinate” and sometimes quite seriously, so it is important to verify the data they produce. The same applies to the sources referenced by Bing – Russian propaganda resources have come up more than once. In such situations, it is important to clearly specify in your requests to ignore data from such resources.
Along with “minuses” there are a number of advantages:
- The speed of neural networks is many times faster than that of humans. If AI takes 30 seconds to search, process and reproduce a certain set of data, a human can spend 100 times more on similar work.
- Editing texts generated by neurons is relatively easier than the same texts written by most humans. This greatly simplifies the work of editors.
- The cost savings are so significant that the economic impact of using even “raw” neurons may exceed the social responsibility to retain workers at all costs so as not to deprive them of their jobs. For example, just a couple months ago, German tabloid Bild laid off 200 employees, saying that some positions could be replaced by AI.
- ChatGPT and Bing may argue with you if your request violates any ethical standards. Yes, every now and then news comes out that someone has gotten around these restrictions, but they often involve the full version of the GPT model (unrestricted by ethical standards – Ed. note) that researchers are experimenting with. For example, in one case, GPT-4 was able to pass Captcha by tricking a real person into saying they had vision problems and asking them to take the test instead. But it was the GPT-4 model itself that was at issue here, whereas the publicly available version is heavily constrained by ethical guidelines and will insist on adhering to those guidelines. Among other things, this fact can be used to analyze journalistic materials for compliance with the principles of journalism.
But despite these opportunities, journalists in Tajikistan have yet to make full use of it. The reason for this may be that:
- The relatively low level of technical awareness of many authors prevents them from understanding all the positive aspects of using neural networks on their part. The same factor prevents many media executives from paying attention to ChatGPT, Bing and other services that could seriously change the work of their newsrooms.
- The journalistic community’s fear of technology that could replace their labor. This is reminiscent of the Industrial Revolution, when machines replaced workers in factories, and today it seems to many that AI will replace their intellectual labor.
- Focusing only on the disadvantages of this new technology and ignoring its advantages. This approach is rather harmful to humans themselves, as it inhibits the exploration and utilization of new possibilities. And given the speed of development of neural networks, the amount of information to be studied becomes larger every day and the later a person starts studying them, the more time, effort and energy it will require in the future.
In general, despite the presence of some shortcomings, some of which may be considered as such for purely subjective reasons, the development process cannot be stopped and neural networks are doing so at an unprecedented speed.
Even their current potential is enough to speed up the work of content makers and editors, save money for the media owner, and help launch new media with minimal staffing costs. In all these scenarios, AI acts not so much as a human replacement, but as an advanced personal assistant ready to help create modern content.
And in this respect, the now-famous phrase “You will not be replaced at your workplace by artificial intelligence, but by another person who uses artificial intelligence in their work” is taking on a tangible form.
Recommendations
Do not think that the trend related to the development of neural networks and artificial intelligence is something temporary and soon the fashion will pass. Every day there are more and more services and programs that integrate AI into their products. The vast majority of these products are related specifically to the process of content processing and creation. The process includes such giants as Microsoft, Adobe and Meta, which are not just not thinking of winding down their projects, but on the contrary are planning to expand the line of products that implement AI. These companies and services are with us for a long time and will shape the industry for many years to come.
For this reason, the main recommendation is that we should embrace this new reality, learn new technologies, and implement them in the content creation process. As with most new technologies, incorporating neural networks into the journalistic content production process can be a competitive advantage for media outlets, making them more productive and efficient than competitors trying to stick to “traditional ways” of producing content.
Recommendations for universities – revise curricula in the near future, at least for media professionals – journalists, future communication specialists, etc. In fact, many students and even professors already use generative neural networks, but at the risk of being accused of plagiarism, they do not publicize it. This circumstance can be used for the benefit of the educational process by giving students the task of generating the necessary content using neural networks and then finding all the mistakes they have made. Students will be forced to know more neural networks in order to find the mistakes they make. The same is true for educators.
Recommendations for employers – to include in the list of requirements for candidates for vacant positions knowledge not only of office programs and skills of handling office equipment, which has already become the norm, but also of using generative neural networks in future work, expanding the assessment of digital literacy of applicants to the issues of proper use of these technologies. To ensure that increased productivity is not accompanied by the leaking of sensitive data, companies should also develop clear policies on the use of neural networks in their operations.