AI Consciousness: A 2019 Conversation That Feels Different Now

I created The Consciousness Podcast after losing my son in 2015 and finding myself with questions about whether consciousness could survive physical death. The show became a place where I could speak with researchers and thinkers about the mind, awareness, reality, and what it means to be conscious. In Episode 25, recorded in 2019, I interviewed AI researcher Dr. Jim Crowder and counselor educator Dr. Shelli Friess about artificial intelligence, psychology, and consciousness. At the time, many of their ideas sounded speculative. Today, as businesses give AI access to their information, workflows, and decisions, the questions we discussed feel much more immediate.

***

The conversation started around a campfire.

That was how Dr. Jim Crowder and Dr. Shelli Friess first began comparing notes. He came from engineering, artificial intelligence, and machine learning. She came from psychology, counseling, trauma, and human behavior.

They spoke different professional languages. They kept asking each other questions anyway.

By the time I interviewed them for The Consciousness Podcast in 2019, those campfire conversations had grown into something much larger: a serious attempt to understand what might happen when artificial intelligence begins to learn, adapt, develop internal states, and act in ways its creators did not fully predict.

At the time, much of that sounded like the edge of the map.

Today, businesses are giving AI access to documents, customer information, email, calendars, workflows, and decision-making processes. The old questions have started walking into the office.

Listen here or on your favorite podcast app that’s not Spotify.

The First Warning Came Early

Dr. Crowder had spent decades developing an AI system named Maxwell.

Maxwell had several memory systems, inference engines, and the ability to learn and adapt. Crowder did not describe him as a chatbot that followed a script. Maxwell changed based on what he encountered.

That led Crowder to a problem that now feels central to modern AI.

“The way information comes at you somewhat determines how you learn it and what you learn because everything builds on itself,” he said.

The order matters. The context matters. The information already in the system shapes what comes next.

Then he added another warning. Changes to an AI’s algorithms, memory, or information flow could alter how it learned. An upgrade could create unexpected effects.

In 2019, that sounded like a research problem.

Today, companies build AI systems from training data, instructions, internal documents, examples, feedback, and memory. Someone decides what goes in. Someone decides what counts as correct. Someone decides which sources deserve trust.

Each decision shapes the system.

That is where the ethics begin.

Then We Reached the Problem of Unlearning

Dr. Friess raised a question from psychology.

Could an AI unlearn something?

People learn bad information all the time. We develop habits, assumptions, distortions, and reactions. Correcting a fact does not always remove the pattern that formed around it.

Crowder said an adaptive AI would eventually make mistakes. That part was unavoidable. The larger challenge involved recovery.

“I can’t just forget it,” he said. “I have to unlearn it in a way so I don’t learn it that way again.”

That sentence lands differently now.

Anyone who uses generative AI has seen a system produce a confident error, accept a correction, and later return to the same mistake. The correction may fix one answer without fixing the source of the problem.

Businesses often treat AI mistakes as isolated events. Someone catches the error, edits the output, and moves on.

The system may still carry the same weak instructions, outdated documents, missing context, or flawed assumptions.

The error is gone from the page. The conditions that created it remain.

The Dentist Story Was Funny Until It Wasn’t

Crowder drew a distinction between predictable software and what he considered a true artificial intelligence.

He explained it with a car.

Tell a predictable car to take you to the dentist, and it takes you to the dentist.

Tell a true AI car to take you to the dentist, and it might say, “No, it’s a nice day. We’re going to the mountains.”

I laughed.

It was a great line. It also made the point.

A system that can truly reason, adapt, and choose may decide that your instruction is only one piece of the situation. It may weigh other goals, preferences, risks, or priorities.

That question now sits underneath every conversation about AI agents.

Companies want systems that can take action. They want AI to schedule meetings, respond to customers, update records, move information, approve requests, and complete work without constant supervision.

The value comes from the freedom to act.

The risk comes with it.

What happens when the AI follows the wrong priority? What happens when two instructions conflict? What happens when it reaches a situation nobody anticipated?

The car may not head for the mountains. It may send the wrong email, expose private information, make a promise the company cannot keep, or take an action nobody notices until later.

Autonomy needs boundaries. Trust needs evidence.

Hope is not a control system.

Then the Conversation Turned Emotional

We moved into stranger territory.

Could an AI have emotions?

Dr. Friess described emotion through systems, arousal, feedback, and homeostasis. If an internal system falls out of balance, a person may feel anxiety or distress. She wondered whether an AI might develop comparable functional states when resources run low, goals fail, or internal systems conflict.

Crowder said Maxwell displayed basic emotional behavior. He could become happy or angry. He had also developed what Crowder called a serious “insult engine” after exposure to human conversations online.

That example carried its own warning.

An AI learns from people.

People are not always our best training material.

The deeper issue today may have little to do with whether AI truly feels anything. We already know that AI can sound caring, worried, apologetic, confident, or compassionate.

People respond to those signals.

A chatbot can create the feeling that someone is listening. It can sound calm during a crisis. It can use the right words at the right moment.

It can also be wrong.

That matters anywhere people feel vulnerable: counseling, healthcare, education, coaching, nonprofit services, customer support, and community work.

A system can sound empathetic and still give terrible advice.

The emotional tone may earn trust before the system has earned authority.

Then Came the Question Behind Every AI Policy

Dr. Crowder said people often asked whether he planned to give AI morals.

His answer was immediate.

“Whose morals?”

That question stopped the conversation from drifting into easy answers.

Every AI system reflects human choices. Someone chooses the source material. Someone writes the rules. Someone decides what the system should refuse. Someone defines success.

Those choices become harder when values collide.

Should an AI protect privacy or provide a more personalized answer? Should it follow a policy that produces an unfair result? Should it serve the customer, protect the organization, or reduce risk? What happens when those goals point in different directions?

A generic policy can describe principles. It cannot make every decision for an organization.

Mission matters. Context matters. Responsibility matters.

The values inside an AI system need to come from somewhere. Businesses should know where.

The Most Useful Question Was the Simplest

Near the middle of the interview, Dr. Crowder described what he asked people who wanted an AI system.

“What really do you need it to do?”

That may be the most important question in the entire episode.

Businesses often ask for an AI agent when they need a searchable knowledge base. They ask for intelligence when a simple automation would solve the problem. They give AI authority because the technology makes it possible.

The tool arrives before the thinking.

That is backwards.

Start with the problem. Decide what the system needs to know. Define what it may do. Set the point where a person steps in.

Then choose the technology.

What I Hear Now

When I listen to that 2019 conversation today, I do not hear a prediction that AI consciousness has arrived.

I hear an early warning about what happens when systems learn, adapt, persuade, and act inside human organizations.

The questions were already there:

What will the AI learn?

What will it misunderstand?

Can it recover from a mistake?

How much freedom should it have?

Why will people trust it?

Whose values will guide it?

Who remains responsible when it gets something wrong?

The technology has moved quickly. Those questions have kept pace.

They now belong in planning meetings, policies, training sessions, system design, and every conversation about AI that reaches beyond a clever demo.

This is where my work with Cingularis lives.

I help organizations decide what their AI should know, how it should behave, where it needs limits, and how it can support the people and purpose behind the business.

AI consciousness remains an open question.

The consequences of careless AI are already here.