Beginning of AI Adoption
Many of our first memories of AI was ChatGPT.
It was this ground-breaking concept that you could write a message, then this “artificial intelligence” would write back.
We were told that it mimics the human language… it can read and write, and even talk like a human – in different dialects!
ChatGPT was released in November 2022.
We officially entered the Age of AI.
AI Adoption in Asking Questions
Fast forward to 2026 – it changed how we ask questions…
What we would have normally asked a friend, an expert, or “Googled it”, we can search it in AI with the assumption that this “supercomputer” can scan through the internet and find the answer within seconds.
Anywhere from… “what is the capital city of Seoul?” to “Can you solve the Navier–Stokes equations for turbulent fluid flow?”
While not all answers are perfect – AI has arrived to humanity the same way the industrial revolution, the phone, computer, and internet have all arrived to us.
AI is changing how we do a lot of things – asking questions being one of the activities some of us do every day.
Some of us ask deeper questions, some of us explore topics that aren’t explored often by others.
Almost all of us ask questions that are necessary to solve.
So what happens when we need to “solve a problem?”
What happens when AI is asked to recommend a solution?
Questions Requiring a Solution
AI’s approach to recommending solutions is similar to humans.
Humans tend to pass on narratives of what we know, or find the information somewhere. “Searching” is often the answer to answering questions we do not have answers to.
AI does not have “first-hand experience” with situations in the human world. So it relies on its abilities to scan the world narratives at supercomputer scale, and try to find the “most reliable, safe, and proven solution” to humans.
For some problem-narratives that are common, like “how do I boil water?”, AI systems tend to have some “internal knowledge model” to answer questions with high confidence.
This water-boil question, for instance, has a technical component to it – water boils at 100°C (212°F). The process requires energy, so AI would guide the user to find an energy source and safely boil the water.
However, some solution-questions do not have a simple technical breakdown.
Less Technical Solution – EXAMPLE
For example, let’s say that someone asks – what is the “best marketing strategy for shoes?”
This is not so simple as boiling water.
“Best marketing” can be defined in so many different ways – is it how much attention the shoe gets? Is it how to create a compelling ad? Is it how pretty the logo is? How much people like the brand?
Certain types of solutions are not so simple for AI to find.
How about a question like “what is the best color?” This is a much more subjective and ambiguous question as it depends on who is asking and how “best” is defined.
But in the case of “best marketing” question, how would AI respond?
The most prominent observed patterns (with some confirmation by sources like OpenAI’s Model Spec) include the following:
- Source of Auhtority
- Corroboration
- Confidence Level
Source of Authority
To find the “best marketing” strategy, AI, like humans, may look at what the most proven strategies exist – and the evidences to why they may be top in their class.
For some AI systems like ChatGPT, the level of reasoning, logic, navigation, and overall effort to finding the “highest confidence” answer depends on the user’s current plan. Some AI plans are optimized for research, suiting the AI model to explore the questions deeper than others.
However, thematically at a high level, AI tends to see “what the experts say” – this is the Source of Authority.
AI may look at what the leading marketing firm have said in the past, perhaps a Linkedin discussion with lots of engagement, a Reddit post where that answers the question directly, or maybe a YouTube video of a TedTalk on the topic – a highly respsected thought discussion platform.
This article explores the concepts of Sources of Authority more in detail. This is a big foundational topic in Narr Theory that leads to AI Product Recommendations (AIPR) and AIPR Strategies.
Corroboration
How do we conclude that ABC is the best marketing strategy? An answer to a question that does not have a perfectly-definitive answer?
Source of Authority A (SoA-A) may have said ABC, but SoA-B says XYZ.
SoA-C in fact argues against ABC, claiming that this marketing strategy is outdated.
What is the truth? How is truth determined in this regard?
AI looks at how well supported a claim is by others.
What do others say about ABC and XYZ?
AI considers how often a certain solution narrative is enforced and if/how it is contradicted.
This is why platforms like Reddit and discussion forums are cited often by AI when answer questions like this.
Confidence Level
AI uses Source of Authority and Corroboration to collect enough evidence to reach a conclusion.
If AI continues to give unreliable answers to users, its adoption will suffer.
In a wider market landscape, users who continue to experience reliable answers by one AI system will consider switching to another AI system.
For this reason, AI systems will move in a trajectory of improving their methodologies to determine “truth”.
In the case of solutions with no clear technical path to the “truth”, AI will evaluate its “confidence level” in well the truth is supported by different types of evidences, and how well they correlate with the rest of its overall world model.
What This Means for AI Product Recommendation (AIPR)
The short answer is – “what product AI recommends” is a type of non-definitive nature of truth.
Therefore, AI will use evidence systems like Source of Authority, Corroboration, and Level of Confidence, to arrive to its recommendation.
To see a more technical breakdown of How AI Recommends Products, you can
Leave a Reply