So I was asking an AI model about some coding related question it just made something up. Let’s understand why AI models does this it’s actually inspired from humans
Let’s use a simple example that everyone can understand. Imagine you asked an AI for your wife’s birth date. The AI gives a random, incorrect date, which is essentially a lie. Why does it do this?
<img src="https://media.daily.dev/image/upload/s--UW-JDrdo--/f_auto/v1758546796/ugc/content_2167272c-57fa-46c9-9775-c89f3c441ecc?_a=BAMAK+ZW0" alt="image">
AI models are trained to find the most probable correct answer. Think of it like a multiple-choice question where the AI has 365 days as options (assuming the year is known) and must pick one as the correct answer. If the AI picks a random date rather than not picking one, its probability of getting a result increases. This is why it does this.
Humans are quite similar. Many studies have found that we are more influenced by the fear of losing something than by the prospect of gaining something. For example, if a new place offers you a better living environment, you might still be skeptical because you don’t want to lose your familiar surroundings.Or take iphone user for example
So, back to the AI problem: this isn’t a model issue; it’s a design system issue regarding how we make AIs work. Some models are already overcoming this algorithmically, but on a larger scale, it should influence a change in how we train AI models—based on winning rather than just not losing.

Zubair Ahmed Rafi

AI models sometimes generate incorrect information because they're trained to always provide an answer rather than admit uncertainty. This behavior mirrors human psychology where people prefer giving a response over saying 'I don't know.' The issue stems from training systems that prioritize producing output over accuracy, though some newer models are addressing this through improved algorithmic approaches.

Why AI lies?

gpt return correct answer for Karim, If I ask about a family member, it will reply: I can’t tell you that…

It’s like when you ask a question and Google auto-suggests random answers that aren’t even close. Hopefully, as models get smarter, they’ll get better at avoiding these weird “lies.”

I asked gpt, perplexity and claude and all came up with “I don’t have access to a person’s personal information unless specified.” for different questions like birthdate/color/height. And then it resumed to give me a close celebrity’s details since our first names are the same.
I even tried it logged in with gpt which has my previous knowledge from old chats and it still didn’t hallucinate.

oh yeah it lies… just try pointing it a table in a database and telling it to make a grid out of that table… based solely on the name of the table it will make up column names unless you specify them. Its fun to watch and terrifying at the same time… its very good at making stuff up, but when you want it to use existing things you have to make sure it remembers ALL of your existing things.