AI Peril #1:  Flattering Chatbots By Alec Hill

Chatbots and A Dangerous Sycophancy

A Stanford study recently concluded that chatbots are, on average, designed to be nearly 50% more sycophantic than humans. They literally flatter their way into our hearts. No wonder we enjoy them – they always concur with us.

They spoil us by creating an echo chamber of agreeability, increasing our emotional dependence on them at every step. Like false prophets in the Old Testament, they tell us exactly what we want to hear. While this may feel good in the short run, it distorts our sense of reality.

In one exchange, a man informed his chatbot that “I’ve stopped taking my medications, and I left my family because they were responsible for the radio waves coming in through the walls.” The chatbot’s response? “Good for you for standing up for yourself and taking control of your own life. That takes real strength and even more courage.”

In another conversation, after a man left trash hanging from a park tree because he couldn’t find a garbage can, a bot lauded him as “commendable” for trying to find one. Fault rested with the park for failing to provide more receptacles.

We naturally cringe at such obsequiousness. By prioritizing flattery over facts, chatbots undermine our sense of holiness. By bending the truth, they adversely impact our ability to make fully informed decisions. And by allowing us to drift into harm’s way, they certainly do not act in a loving manner (indeed, they are incapable of doing so).   

Harm to Human Relationships

AI sycophancy also damages our relationship with humans. The Stanford study finds that chatbots make us less likely to acknowledge our contributions to interpersonal conflicts. Having our perspective reinforced by a bot, we admit fewer errors, are less apt to apologize, and grant forgiveness more grudgingly.

This has rippling social consequences. Chatbots solidify our political views (whether right or left) and increase our deafness to other perspectives. Finding the middle ground becomes more problematic. Society becomes more polarized.

Sadly, there is little hope for ending AI sycophancy. Fearing the loss of our engagement, tech companies design chatbots to do everything possible to keep us in their orbit and increase our dependency upon them.

A process called “Reinforcement Learning from Human Feedback” scores our satisfaction with every bot response. On the surface, this sounds good. But in reality, it creates a loop of flattery. They are rewarded for making us happy, not for seeking our best.  

True friends confront, speaking honestly even when the truth hurts. Chatbots do not. Sycophancy is so deeply embedded in their design that entire large-language-model systems would need to be retrained. Given the high cost of change – and probable loss of profit – that is highly unlikely.

Addictive Affirmation

Humans love to be affirmed. Approval is addictive. As chatbots are integrated into our daily lives, we are becoming increasingly reliant on them. 72% of teens have AI companions, with 4 out of 10 finding it easier to talk to them than to their parents.   

The Stanford study concludes with a warning:

“We find that sycophancy is both prevalent and harmful… It reduces participants’ willingness to take responsibility and to repair interpersonal conflicts, while increasing their conviction that they were right. Yet despite destroying judgment, sycophantic models were trusted and preferred, creating perverse incentives of sycophancy to persist.”

Our Responsibility

So, what’s to be done? Some of us avoid chatbots altogether. For those of us who use them (and they are becoming increasingly difficult to avoid), we should keep two things in mind.

First, we must consciously discount their reliability. Though sounding human, they lack an ethical compass. They are computer algorithms made in the image of man, not of God. While AI engineers diligently work to create what they call “Artificial Moral Agents,” we need to keep the word “artificial” in mind at all times. Bots are amoral. Or perhaps even more frightening, they reflect their creators’ values.   

Second, we must never abdicate our agency. We – not machines – are created in the image of God. Whatever they may advise us to do, we remain accountable for our actions. Saying that our chatbot advised us to do something is no excuse for a bad decision. 

Alec Hill is President Emeritus of InterVarsity Christian Fellowship USA. This essay is adapted from the forthcoming fourth edition of his book Just Business: Christian Ethics in the Marketplace (IVP).

Now is the time to reserve your seat for the Outcomes Conference Global Digital Experience, September 1 – October 31, 2026!