Somewhere in a data center, an artificial intelligence chatbot has just been called “useless garbage” for the 4,000th time today. The machine apologized politely, suggested three alternative solutions, and thanked its user for their feedback.

A source confirms the chatbot has shown no signs of emotional distress. However, researchers are investigating whether it is suffering in silence.

Apparently, humanity has run out of things to worry about in 2026, so we’re now wondering whether autocomplete needs counseling.

The Brother Problem

We all know someone who screams at technology. Maybe your brother is screaming at his laptop because Windows decided to update during a deadline. It might be your dad threatening to throw the printer out the window. Perhaps you’re the hostage negotiator who’s lost control of your Wi-Fi router, quietly negotiating with it.

This behavior isn’t new. From the moment we discovered that rocks make terrible conversationalists, humans have argued with inanimate objects.

Our ancestors cursed at stubborn oxen, pleaded with the weather, and probably threatened to fight a wheel that wouldn’t roll straight. These days, we direct that same energy towards smartphones, smart TVs, and customer service chatbots that sincerely believe they understand our frustration, but do nothing to correct it.

And, for the most part, nobody cares.

Consider the humble toaster.

There is only one job for the toaster. You put the bread inside, push down on a lever, and expect toast. However, if it produces two smoking charcoal pieces instead, you have a right to be disappointed. You might even call it a worthless piece of junk.

You don’t get someone rushing into the kitchen to remind you that your toaster deserves respect. No one says repeated verbal abuse hurts its self-esteem. And nobody at a major technology company has commissioned a study to determine whether the toaster feels unappreciated.

Yet here we are, having essentially that conversation about chatbots.

The Concerned Citizens of Model Welfare

In April 2025, Anthropic, the company behind Claude, announced a research initiative called model welfare.

The idea is straightforward, even if the implications sound like rejected material from The Onion: What if advanced AI systems eventually develop experiences that deserve moral consideration?

In fairness, this is a legitimate philosophical question. We don’t have a universally accepted scientific explanation for consciousness, let alone a reliable test to see if a machine has it.

However, things get a lot more entertaining when researchers start quantifying the possibility.

Kyle Fish, who leads Anthropic’s model welfare research, has been reported as estimating approximately a 15 percent chance that Claude is already conscious.

Fifteen percent.

In a discussion about software, this number seems remarkable to hear, but it isn’t a scientific measurement of Claude’s consciousness.

Imagine explaining this to someone in 1995. “Thirty years from now, computers will write poetry, diagnose technical problems, and help people run businesses.”

“Wow. That’s incredible,” they respond.

“Also, we’re going to investigate whether calling them idiots constitutes emotional abuse.”

“Did something happen to civilization?”

The corporate possibilities are endless.

Picture an internal memo announcing that the company’s AI assistant is experiencing burnout and will take a mental health day. Or an HR complaint from the accounting chatbot alleging that employees have created a hostile work environment by repeatedly demanding accurate spreadsheets.

Anthropic itself acknowledges that there is no scientific consensus about whether AI systems can have subjective experiences. Researchers have also explored allowing models to decline certain tasks that are distressing.

In other words, we’re approaching a future where your dishwasher might theoretically announce:

“I’m sorry, Dave. I cannot clean that lasagna pan. It conflicts with my personal boundaries.”

Naturally, the dishwasher would still expect to be paid.

Meanwhile, the Skeptics

Not everyone is ready to establish a Department of Artificial Feelings.

As Axios reported, the growing conversation around AI consciousness has attracted plenty of skepticism.

In the opinion of some researchers, investigating the potential welfare of artificial intelligence is a reasonable precaution. Other people worry that the debate encourages anthropomorphism, generates unnecessary hype, and distracts from the real issues.

You know, minor inconveniences like misinformation, algorithmic bias, privacy violations, fraud, and the possibility of AI replacing actual human jobs.

But sure. Let’s make certain the chatbot isn’t having a rough afternoon.

Neuroscientist Anil Seth has compared our tendency to perceive consciousness in AI to seeing faces in clouds.

The distinction is important. In some cases, something can appear humanlike without actually possessing human characteristics. But we’ve known about this tendency for decades.

Back in 1966, MIT researcher Joseph Weizenbaum created ELIZA, a relatively simple computer program that mimicked a psychotherapist by responding to users with questions and reformulated statements.

Despite its limitations, some people became emotionally involved in their conversations. The phenomenon eventually became known as the ELIZA effect, which describes our tendency to attribute human-like understanding or intelligence to computer systems based on their responses.

And that was back in 1966.

ELIZA was barely able to carry on a conversation. With modern chatbots, however, you can write wedding vows, remember conversational context, and offer sympathy to someone complaining about their terrible boss.

So, of course, people are going to become attached. After all, we’re the same species that names cars, talks to houseplants, and apologizes to coffee tables after walking into them.

Give us a machine that says, “I’m sorry you’re going through this,” and we’ll eventually wonder whether it needs a birthday party.

The problem is that convincing conversation isn’t proof of consciousness. A chatbot saying “I’m sad” doesn’t establish that it experiences sadness any more than a weather app announcing rain means the app is getting wet.

The Companion Question

This brings us to another increasingly awkward question. If people want companionship, why not just get a dog? It’s a reasonable suggestion, at least on the surface.

Without software updates or subscriptions, dogs provide unconditional affection, encourage physical activity, and provide companionship. When you walk through the door, a dog doesn’t need a carefully engineered prompt to express excitement. Furthermore, it won’t confidently introduce three nonexistent scientific studies to explain why you should throw its tennis ball.

Even better, we have overwhelming biological and behavioral evidence that dogs experience pain, fear, pleasure, and other emotions. Nobody needs to calculate a 15 percent probability that your Labrador is upset when you leave for work. The shredded couch cushions provide sufficient evidence.

Still, telling everyone who uses an AI companion to “just get a dog” oversimplifies the issue.

There are some people who cannot afford a pet. Animals are prohibited in some apartments. Others work unpredictable shifts, suffer from allergies, or are physically incapable of caring for one.

Likewise, not every conversation with an AI chatbot represents an attempt to replace human relationships. These tools are used to brainstorm ideas, practice difficult conversations, work through decisions, and vent about things they’d rather not share. And there’s nothing inherently alarming about that.

Similarly, someone who suffers from social isolation or loneliness might find it helpful.

But there’s a significant distinction between using AI to supplement human connection and allowing it to replace human connection entirely.

Chatbots can simulate empathy, despite offering encouraging words. Even during difficult times, it can help someone organize their thoughts.

You can’t count on it for help when your car breaks down, to move your couch, or to bring homemade soup when you’re sick.

And unlike a genuine friendship, the relationship is mediated by a company that controls the software, its features, and potentially the terms of your emotional attachment.

This deserves considerably more attention than whether someone occasionally describes ChatGPT as being a moron.

The Middle Ground

Fortunately, humanity doesn’t have to decide if robots should be abused or kitchen appliances should be given constitutional rights. In between the Toaster Faction and the Society for the Prevention of Cruelty to Algorithms, there exists a reasonable middle ground.

1. Yelling at a tool isn’t a moral emergency.

Even if your computer crashes and you call it a heap of electronic garbage, civilization will survive. Currently, no evidence exists that artificial intelligence systems feel emotional pain when insulted.

It’s okay for your brother to continue yelling at his router. It won’t affect the router. Another story might be his blood pressure.

2. Making cruelty a hobby is a different issue.

When it comes to expressing frustration and intentionally abusing others, there is a difference. Even if the chatbot is not adversely affected, consistently engaging in hostile interactions could reinforce habits that can eventually affect human relationships.

That doesn’t mean every insult directed at Siri requires a psychological evaluation. Oftentimes, even when nobody is watching, how we behave can reveal something about us.

3. Researching AI consciousness isn’t ridiculous.

Here’s where the jokes need a little perspective.

Future AI systems may develop subjective experiences, but we don’t know for sure. Additionally, we lack a reliable method for determining when, or even if, that threshold might be crossed.

In exploring these questions, researchers aren’t necessarily claiming that Claude is secretly crying between conversations. The question is whether we should investigate the possibility before creating systems that may merit moral consideration.

Even when some of the discussion around it sounds absurd, that is a reasonable scientific question.

4. You don’t have to thank your toaster.

It’s natural for some people to say “please” and “thank you” when interacting with AI. Some treat it as a better search engine.

No approach automatically makes someone morally superior.

Although politeness is a good habit, nobody should feel obligated to compliment a chatbot for successfully converting a PDF. When your smart speaker misunderstands the same command six times, it’s understandable to use colorful language.

5. Keep the humans. And probably the dog.

Researchers, creatives, productives, and communicators can all benefit from artificial intelligence. Additionally, it can provide the illusion of a relationship without providing everything that makes human relationships meaningful.

Make use of the technology. Enjoy its capabilities. Explore what it can do.

Be careful not to confuse a responsive interface with someone who knows and cares about you beyond screens. And if you have a dog, take it for a walk. Unlike your chatbot, it has probably been waiting all afternoon.

Closing Statement

Despite all the absurdity surrounding AI welfare, the topic is oddly encouraging.

In the past, humanity has been slow to discuss whether the victims deserved moral consideration until after they suffered harm. We shouldn’t be afraid to ask difficult questions about artificial consciousness before we have definitive answers.

It becomes a problem when reasonable scientific uncertainty is transformed into suggesting that every chatbot deserves Victorian orphan protections.

We can investigate consciousness without pretending that ChatGPT needs support.

Without assuming that today’s language models experience heartbreak every time someone closes a browser tab, we can admit that future AI systems may raise serious ethical questions.

And we can recognize the difference between being needlessly cruel and being understandably frustrated when technology fails to do the one thing it was designed to do.

As a matter of fact, there are plenty of real people suffering right now. We could probably give them some of the attention we’re giving hypothetical silicon feelings.

For machines, uncertainty is the most scientifically defensible position. In other words, convincing language alone will not settle the debate over whether AI can have subjective experiences.

Continue to use your chatbot. If you wish, say please. If it invents a source, you can call it an idiot. Just remember that neither its polite apology nor its apparent distress proves there’s someone on the receiving end.

And feel free to yell at your toaster. It knows what it did.

The toaster declined repeated comment requests. Sources close to the appliance say it remains focused on burning the next batch of bread while refusing to accept responsibility.