Could humans get pushed out of the market research industry?

?? Quick reminder! We are hosting a webinar on AI-powered research. Come for the tech predictions, applications and a reveal of a new tool. Register now!

Article by Asya Zorya , head of AI lab at Fastuna

By now, ChatGPT managed to make a lot of noise. OpenAI has lowered the barrier to entry into AI technologies to the level of "if you know how to use the chat - go ahead, have fun."

Users were simultaneously impressed and frightened by the illusion of a deep understanding of whatever the chat was fed, the high speed of processing material of any complexity, and its variability. People instantly realised that AI could easily do the jobs that many of them were paid to do.

Programmers' biggest fear is being replaced by a program. Strong AI has brought this prospect closer: some large language models (LLMs) already write good enough code. Let's use an analogy to understand that researchers are not very different from programmers.

Don't panic! We put together a guide to help insights managers secure their roles. Read all about it in our blog →?

Here is a simple problem that even a junior student can solve: "Sort the numbers in descending order: 2, 8, 1, 4, 6, 3, 5, 9, 7."

In order for a machine to solve this task, a specially trained person is needed - a programmer who a) reads the problem, b) comes up with an algorithm, c) writes code in the language of the machine. For example, in Fortran:

Now let's consider a simple research problem:

"Brand XXX is entering a new market. The business is localising and must change its name. It is important to ensure continuity and not lose current brand users. The task is to choose the best of five options for a new name."

The client cannot come up with an algorithm for solving the problem, but they can explain what they need. At this point, you will need another specially trained person - a researcher who will:

1. read the client's problem,
2. come up with a solution algorithm using research methods and tools,
3. provide the client with information for making a business decision in the form of a report.

The researcher is a kind of "interface" for the client's access to research tools. The "interface" accepts a task from the client in the form of text and gives them an answer – again, in text. It turns out that text, language, is the most powerful of researchers' tools. Researchers use language to perform each and every task.

Working in ChatGPT is similar to programming in simple human language. Instead of code, we use prompts - a set of arbitrary instructions about how AI should behave and what it should do. A prompt cannot be specified to just any length, it has to adhere to the limitations of the LLM model. In our experiments (which we will describe below) we used gpt-3.5-turbo and a "smarter" version of gpt-4, as well as free open-source models (gpt4all-j 1.3, BLOOMChat-176B, MPT-7B). The listed LLM models understand complex texts well and provide (seemingly meaningful) answers.

What then is the fundamental indispensability of the human researcher?

Let's try to take everything "human" out of the standard research process and take a closer look at where we end up!

??We are hosting a webinar with the takeaways from our year of vigorously testing AI for research purposes. Last chance to register!

?? Familiarise yourself with Fastuna AI in advance - https://fastuna.com/ai

Get our map to learn when to use AI research solutions during NPD →?

The process of collecting insights is changing dramatically. Here is how →?

Follow our page →?not to miss new tips about market research.

Visit our website →?test product and marketing ideas at any stage within 24 hours across 50+ markets. Book a free 30 minutes session to learn more.