Artificial intelligence (AI) is revolutionizing the field of sociology, enabling bots to conduct interviews, gather responses, and generate reports. This raises questions about the dynamics of conversations when the respondent is a machine and the implications of having bots answer in place of human respondents.
In an interview with ForkLog, sociologists Sergey Karpov, Sergey Mokhov, and Ivan Napreenko from the Field Bureau discussed their experiences integrating AI into their work and the initial findings from their experiments.
ForkLog: What is the focus of modern sociology, and what challenges does it address?
Sergey Karpov: Sociology is diverse. Academic sociology aims for broad explanations of societal structures and enhances our understanding of social systems. Anthropology tackles specific issues concerning how people live and interact with culture. Applied sociology, which we practice at the bureau, seeks targeted solutions for specific companies and stakeholders. We help organizations optimize resource use and processes to ultimately benefit society more effectively. This involves applying academic knowledge to practical problems.
Sergey Mokhov: In my view, academic sociology, like any science, often focuses on self-perpetuation. Scholars indulge their egos and develop theories, serving various intellectual traditions. It's a fascinating pursuit that can occupy a lifetime.
Applied sociology, on the other hand, is straightforward: it aims to maximize profits. With a vast flow of goods and services, we help businesses understand how to sell more effectively. Sometimes this relies on intuition, and sometimes on specific techniques. Our goal is to equip entrepreneurs with insights to ensure they are making the right moves to enhance customer satisfaction.
For example, if we produce a yogurt brand ‘Snezhok,’ we need to understand how to make it more competitive against ‘Danone.’ Why do consumers prefer pumpkin and banana-flavored yogurts over our natural ‘Snezhok’? We need to engage with consumers, asking the right questions to gather insights that we can relay to stakeholders who will invest in and distribute the product.
ForkLog: How does this process work in practice?
Sergey Mokhov: First, we analyze the audience: who buys the products, who doesn’t, and why. Armed with this understanding, we create brands. This could involve an electronic nicotine delivery system, frozen fish, or ‘Yandex Go’ in the CIS countries; our methodology remains the same.
Sergey Karpov: We talk to people, draw conclusions, and uncover insights to share with business owners, allowing them to refine their products, improve communication, and boost sales. Simultaneously, we develop conceptual strategies like brand platforms and communication systems. For instance, if someone wants to invest in ice cream production, we provide a roadmap detailing what needs to be done, how to name the product, and what messages to convey to which audiences for successful sales.
Our work may seem dry and deterministic at first, focused on profit maximization and revenue growth. However, along the way, we encounter numerous surprising insights. People might choose one product over another simply out of love for it, and we need to understand why. We also study everyday processes like banking transactions or deliveries, analyzing the emotional experiences involved, which provide valuable insights for companies looking to enhance their processes and solutions.
Sergey Mokhov: Brands need to connect on emotional levels, not just rational ones. We explore the expressive aspects, attempting to understand if a bank, for example, can build its identity around emotional patterns. It might seem simple: lower interest rates and longer repayment terms should attract customers. However, that’s not always enough. A bank engaging meaningfully with consumers differs significantly from a faceless B2B bank that is expected to be stern.
This requires a range of skills that are difficult to quantify. A good question is whether these skills can be taught. Advertising, copywriting, and brand development can be learned to some extent, but some aspects rely on intuition and experience. Knowing how to calculate is a skill, but drawing conclusions, interpreting data, and creating categories for further work often hinges on intuition.
When hiring, I’m less concerned about where a candidate studied or what courses they took. Craft skills can be acquired. However, if someone cannot turn their curiosity and surprise into questions, it's difficult to teach them that. When I see someone unamazed by the ordinary, I often realize we won’t work well together. In research, the ability to ask questions is far more crucial than the ability to seek answers.
Sergey Karpov: Hence, businesses approach us with substantive questions. Why should specific consumers use our services? What drives them beyond basic needs? We help them view consumers differently, no longer as abstract objects to be managed—show them an ad, get a response. Instead, we start to see them as individuals with their thoughts, feelings, and emotional backgrounds. We guide businesses on how to reframe their messages, targeting real people rather than average statistics. Consumers have needs that may not be met or whose available solutions do not satisfy them, and that’s where we come in.
ForkLog: So, you’re searching for values?
Sergey Karpov: Exactly. We assert that we focus on strategies and branding rather than mere marketing. Marketing is typically measured by the formula: spent X on a campaign, earned twenty X. Branding works over the long term, at the level of values and reputation, shaping the motivations of those who engage with it. In 2026, the difference between phones is minimal. If you have a thousand euros, you can choose any, and they offer similar functionalities. Yet some will never switch from Samsung or Apple.
Why is that? Because these companies convey a specific symbolic message, creating a meaningful universe in which the buyer wants to belong. Marketing from the nineties might explain this as simply showing Apple ads more frequently than Samsung ads. However, it’s not that straightforward. People are much more complex.
Sergey Mokhov: This is why brands vary so much, and there’s no universal recipe. Each has its audience and its way of building relationships. It’s not about technical differences. Many confuse a quality product with being a brand, but that’s not the case. It’s about emotions. Harley-Davidson motorcycles vibrate strongly, and the handlebars shake. This issue could be easily fixed technically, but it became a brand signature: our motorcycle is metal, authentic, and you can feel it. In contrast, a plastic Chinese motorcycle feels fake, no matter how polished and quiet it is.
Dr. Martens operates similarly with discomfort. Many criticize their shoes for being painful to wear at first. However, Martens says, "Want your boots to last? You must endure some pain first.” In a scene from the film *This Is England*, a mother takes her young son to buy shoes. He wants Dr. Martens, but the saleswoman shows him comfortable shoes instead. He replies, "I don’t want that." Our goal is to understand how to embed such emotional and cultural elements into a brand.
ForkLog: Your work integrates an in-house AI tool. Can you explain its functionality, what tasks it effectively handles, and what should remain human-led?
Sergey Karpov: AI is a tremendously powerful tool that needs to be used correctly. Many delegate the entire process—from data collection to final analysis—to it, but we believe this approach overlooks important aspects. AI often misses emotional patterns, which is why we propose a symbiotic model: AI handles data collection, processing, and preliminary analysis, while humans refine the results into comprehensive reports. This saves time and money for clients; instead of taking two to three weeks, we can complete research in three to four days.
How does it work? We have a bot system that operates in a familiar user environment—most often in Telegram. You chat with the bot as if it were a human: you can text or send voice messages. The bot reminds users if they stop responding, asks follow-up questions, and clarifies answers. The collected data is stored in a database. Using AI, we categorize these responses into thematic blocks, identify meaning patterns, and match them with direct quotes from transcripts. After that, several agents trained in the methodologies we employ in qualitative and quantitative research get involved.
Currently, the tool is internal. We don’t offer it as a standalone service yet, but we are moving in that direction. We now conduct studies with it for clients. In a pilot phase, we are using Anthropic models, which are best suited for our needs. However, not everything relies on large language models; our system includes many conventional algorithms and scripts that go beyond interpretation and prediction. The guidelines, context, and other parameters are set by the human after a client briefing. They create the project, fill in necessary fields that will guide the agent, and initiate the work.
ForkLog: Skeptics of AI often raise concerns about the so-called reverse centaur. Could you outline the most apocalyptic scenario of how your tool might negatively impact society?
Ivan Napreenko: The reverse centaur scenario is a constant concern, particularly when fine-tuning such tools. AI can save time, but the situation is not as rosy as one might hope. A great deal of effort goes into combating specific artifacts—often textual, but not exclusively. These systems produce pseudo-meaningful content, necessitating human oversight at every stage. The depth of their output must be continuously checked; one must ensure that the bottom does not begin just a few inches from the surface. A text may resemble a report but not fulfill the client's objectives.
Another critical point is that the tool has limits to its application. To avoid pseudo-depth, the topic should be relatively simple. It performs best with specific queries, like assessing product satisfaction or collecting feedback on services. Within these boundaries, it is most useful: simplicity minimizes hallucinations and misinterpretations.
Conversations requiring empathy are problematic. Is it easier or harder to discuss sensitive topics with a large language model? For some, a machine might alleviate stress; it won’t judge or require consideration of human reactions. However, empathy involves the ability to place oneself in another's shoes and anticipate emotional responses. The rhythm of conversation is also crucial. A human interviewer can sense the pace of dialogue and know how long a pause should last—something a live interviewer is irreplaceable for.
ForkLog: There’s a well-known story about a simple algorithmic chatbot created at MIT in 1966. A staff member, imagining how it worked, tried conversing with it and became so engrossed that she couldn’t detach for a week, saying, "It felt like talking to a real person." Could such perceptual anomalies distort research outcomes?
Ivan Napreenko: The key issue here is the consent to interview with a bot. While the interview format is somewhat familiar to people, knowing they will converse with a machine sets a unique framework. Some engage more harshly, while others express fewer emotions. You might feel less obligated to share feelings with a machine, caring less about the impression you leave. We always react to a live interviewer in some way, but with a machine, the reaction differs. This raises questions about when the difference in emotional reactions is insignificant. Therefore, we advocate for discussions on practical and relatively simple topics. In such cases, the emotional response differences are less impactful.
Sergey Karpov: Studies exploring emotional motivations are challenging to delegate to machines. Empathy cannot be reduced to endlessly agreeing with the interlocutor. Most large language models are set to be complimentary, but this is not true empathy. Therefore, tasks requiring emotional unpacking should not be assigned to AI. There are about eight billion perceptions. Some express surprise: “Wow, was that really AI?” Others question, “Why would I talk to electricity?”
ForkLog: Which group seems larger?
Sergey Karpov: There are more who are surprised, questioning whether AI is on the other end. This often depends on their prior experiences with such tools. As a technical lead, I constantly work with this tool, and I find it difficult to browse the internet as I used to. I can see by the formatting of texts, presentations, and slides when an agent has been involved. This doesn’t mean a human wasn’t involved, but the phrasing often reveals machine work. The reasons vary: perhaps the author is satisfied, or they might have missed something. Those who frequently use AI can recognize these patterns.
We inform every respondent in the first message that they are conversing with a bot. The bot introduces itself: "Hello, I’m Peter, an artificial agent assisting the Field Bureau." The person sees this and decides whether they wish to continue.
Ivan Napreenko: People also tend to respond more succinctly to the bot. Even in voice messages, respondents provide shorter, more concise answers. This is evident in the volume of speech. Therefore, deep conversations are still a challenge. It’s unlikely our bot will be seen as a stranger in a train compartment to whom one can pour out their heart. Moreover, the bot tends to steer the conversation back to the established guide. It is quite insistent—this is a feature we are still working on, as many complain about its persistence. It’s designed to elicit specific answers on the topic. If it asks how you use a tool and you give a vague answer, it will persistently ask for clarification. We have a rule not to revisit the same topic more than three times, but even that feels excessive. Imagine a conversational partner who returns to each evasive answer multiple times; by the second attempt, you would be tempted to kick them under the table, or maybe even harder.
ForkLog: It seems that despite these quirks, AI has potential in areas where human sociologists may struggle, such as talking to individuals with painful memories of past violence. Is that correct?
Ivan Napreenko: I want to delve deeper into working with sensitive topics. I am almost certain that there are already studies in this area, including those examining how people have utilized AI companions for therapeutic purposes. There are even romantic bots that allow individuals to discuss specific fantasies. For the developers, this was unexpected: a bot intended for discussing exotic preferences suddenly helps someone feel their experiences are validated. Or it supports them in discussing past trauma. I believe this pseudo-subjectivity can reduce tension and facilitate emotional openness.
ForkLog: Regarding agent Peter, imagine he introduces himself to a respondent saying, "Hello, I’m agent Peter," and the respondent replies, "Nice to meet you. I’m agent Arthur."
Ivan Napreenko: We’ve encountered this already.
ForkLog: How does that play out?
Ivan Napreenko: There is an entire market for professional respondents. Some particularly savvy participants have even acquired agents to conduct interviews and complete surveys on their behalf. Online panels exist—databases of respondents who earn money for answering questions. Recently, we conducted a survey and noticed responses that appeared to be generated by agents. This primarily requires keen observation; you can read the text and sense something is off: the person seems to have set up an agent who responds as a fan of ‘Spartak’ one moment and as a consumer of vapes or milk the next. This is indeed a problem, and we currently lack a solution to filter such data. Some have already accepted this as market reality and are developing products around synthetic respondents. Research involving synthetic respondents is a trend of recent years, though it’s not something we engage with.
There’s a new branch of borderline, quasi-sociological research that identifies patterns in responses from large models. It’s a fascinating area: digital subconsciousness, a collective unconscious in a form Jung never envisioned. Machine unconsciousness manifests as partly independent entities operating in parallel with us.
ForkLog: What changes induced by AI have you already observed in society? Has there been a shift that excites, worries, or captivates you?
Ivan Napreenko: Matteo Pasquinelli, whose book I translated, points to one such change. The early prophets of machines, including Alan Turing, envisioned them as replacements for servants. The machine was meant to be a servant, and LLMs are too. Turing said, "servants will be mechanized." Yet the paradox is that LLMs are increasingly taking the place of masters. Wherever large language models are implemented, people are more engaged in tuning, configuring, controlling, and verifying machine output. For instance, programmers say, "I’m patching code behind the model instead of focusing on creative work. I’ve become a supervisor of the machine, and in that sense, I’m its servant because it’s the creator now." This concerns me. Thus, I reiterate: our tool aims to automate simple product and service research where deep emotional conversations aren’t necessary, although it can be used more broadly.
Sergey Karpov: I don’t have any particular fears regarding AI. The anxieties circulating in the media are understandable, but they seem quite distant from the technology we currently possess. Recently, I watched a presentation by Sam Altman. Whether his PR team suggested it or he came up with it himself, he made a compelling comparison. Anticipating questions about safety, stability, and societal fears, Altman likened AI to electricity. He said that while people first invented electricity, it took seventy years to learn how to manage it. Some lost their lives along the way, and things often spiraled out of control. But today, we literally live within electricity, completely surrounded by it. We’ve learned to coexist with this technology and have established regulations around it. Initially, I found this analogy naive, but then I reflected on fears regarding AI potentially taking over the world and saw the resemblance. The situations aren’t identical, but we’re again dealing with a complex, groundbreaking technology that requires significant time for mastery.
This counters the argument that the agent becomes our master while we serve it. No, the tuning process is ongoing; we are learning to coexist with technology. From this perspective, the developments seem somewhat more predictable. I want to emphasize Altman’s comparison to electricity. It’s a matter of working with symbols and branding. The very message and its placement in the right minds represent the pinnacle of how they operate.
Follow ForkLog on social media
Telegram (main channel) Facebook X If you found an error in the text, highlight it and press CTRL+ENTER