AI, Philosophy, and the Many Shapes of Thinking

In this episode of Agile Software Engineering Deep Dive, we take a more reflective look at artificial intelligence.

When we discuss AI, we often compare it to human intelligence, as if humanity were the only valid model for thinking. But is that the right comparison?

This episode explores how AI brings old philosophical questions back into modern software engineering. We look at Turing’s imitation game, Searle’s Chinese Room, Descartes’ ideas about language and reason, and Plato’s distinction between knowledge and appearance.

We also discuss why intelligence may not have one single architecture. Symbolic AI, neural networks, neuro-symbolic AI, embodied AI, neuromorphic computing, and biological computing all suggest that thinking may have many possible forms.

The central question is not whether AI is already human. It is not.

The deeper question is whether AI helps us understand that intelligence was never only one thing.

Transcript

SPEAKER_00 0:29

Welcome to the Agile Software Engineering Deep Dive, the podcast where we unpack the ideas shaping modern software engineering, technology, and leadership. My name is Alessandro Guida, and I've spent most of my career building and leading software engineering teams across different industries. Today I want to take a slightly more reflective turn. We often talk about artificial intelligence as if there is one final destination: human intelligence. The closer a machine comes to writing like us, reasoning like us, speaking like us, or making decisions like us, the more intelligent we say it is. But perhaps that is already the wrong starting point. As software engineers, we know that the same problem can often be solved through very different architectures. The same external behavior can be produced by very different internal mechanisms. A bird flies, an airplane flies, a helicopter flies, a drone flies, but they do not fly in the same way. So why should intelligence have only one valid architecture? In this episode, we explore the relationship between AI and philosophy. Not because Plato, Descartes, Turing, or Searle had direct answers to modern machine learning, large language models, or data centers full of GPOs? They did not. But they asked questions that have suddenly become practical again. What is thinking? What is understanding? Is intelligent behavior enough? Can language create the feeling of mind? Can a system produce correct answers without knowing what those answers mean? And when we say artificial intelligence, do we mean fake intelligence? Or intelligence produced by a different mechanism? We will look at Turing's imitation game, Searle's Chinese room, Descartes' ideas about language and reason, Plato's distinction between knowledge and appearance, and the many possible architectures of AI, symbolic, neural, neurosymbolic, embodied, neuromorphic, and perhaps even biological. The point is not to claim that today's AI is human. It is not. The point is also not to claim that AI is conscious. We have no reason to say that. But maybe the most interesting question is not whether AI will become a perfect copy of us. Maybe the more interesting question is whether AI helps us discover that thinking was never only one thing. So let's dive in.

SPEAKER_02 3:17

Oh yeah. It uses our words, it mimics our tone.

SPEAKER_01 3:20

Right. So it is incredibly easy for you to picture a tiny, highly efficient scholar just spinning its wheels inside that server somewhere. But uh, what if our entire definition of intelligence is just, well, too narrow?

SPEAKER_02 3:34

Aaron Ross Powell, we constantly measure the machine by how well it reflects us in the mirror. I mean, it's a very human trap to fall into, right? Totally. We assume that because the output looks familiar, the internal process just has to be identical to our own.

SPEAKER_01 3:46

Aaron Powell So it becomes this game of projection. We're projecting our own humanity onto the machine.

SPEAKER_02 3:51

Exactly.

SPEAKER_01 3:52

And that fascinating tension is exactly what we're getting into today. We're pulling from issue number 35 of the Agile Software Engineering newsletter. It's a great piece. Yeah. It's an incredible piece titled AI, Philosophy, and the Many Shapes of Thinking. And our mission for you on this deep dive today is to explore how ancient philosophy actually helps us decode modern AI.

SPEAKER_02 4:13

Which sounds wild, but it makes so much sense.

SPEAKER_01 4:16

It really does. And why intelligence is not just a single final destination. But uh before we jump into the deep end, a quick favor to ask you make sure to read and subscribe to the full newsletter article.

SPEAKER_02 4:28

I highly recommend it.

SPEAKER_01 4:29

Yeah, there are fantastic illustrations and even more in-depth content waiting for you there. Also, this deep dive is completely free, but it is a massive help to us if you press like on this track and subscribe to receive all our future issues. It really is the best way to keep learning with us.

SPEAKER_02 4:46

So to set the stage for you here, the author brings up a really practical analogy from software engineering. In software, there is never just one correct architecture to solve a problem. Like sometimes you might build a monolith.

SPEAKER_01 4:59

Right, where everything is just baked into one massive centralized block of code.

SPEAKER_02 5:03

Exactly. But other times you might use uh an event-driven architecture where dozens of separate independent pieces react to specific triggers. You simply match the structure to the problem at hand.

SPEAKER_01 5:14

Makes sense.

SPEAKER_02 5:15

And the core premise we're exploring today is that intelligence should be viewed through that exact same lens.

SPEAKER_01 5:20

Aaron Powell Intelligence as an architecture, I love that. But to understand where we are going with these different shapes of thinking, we really have to look back at where this whole dream of artificial intelligence actually started.

SPEAKER_02 5:33

And that origin point was a very specific, incredibly bold engineering conjecture back in 1955.

SPEAKER_01 5:39

The Dartmouth proposal, right?

SPEAKER_02 5:41

Yes, the famous Dartmouth proposal. It was written by some of the absolute titans of early computer science. I mean, people like John McCarthy, Mervyn Minsky, Nathaniel Rochester, and Claude Shannon.

SPEAKER_01 5:52

Heavy hitters.

SPEAKER_02 5:53

Absolute legends. And they planted a flag and formally proposed that every aspect of learning and, you know, any other feature of intelligence could be described so precisely that a machine could simulate it.

SPEAKER_01 6:04

Just imagine the sheer audacity of these scientists in 1955. I mean, computers at the time were the size of entire rooms.

SPEAKER_02 6:11

Right, running on fragile vacuum tubes.

SPEAKER_01 6:13

Exactly. They use punch cards, and they could barely manage basic algebra. And here are McCarthy and Minsky essentially saying, uh, give us a summer and we can map out the entire human mind as a mathematical equation.

SPEAKER_02 6:26

A two-month study, they called it?

SPEAKER_01 6:27

Yeah. It wasn't just an engineering goal, it was immense hubris.

SPEAKER_02 6:31

Hubris, definitely, but also a profound philosophical shift. They were basically asserting that intelligence isn't magic. Right. It's not some mystical vapor floating around in our skulls. It is a mechanical process. Learning, reasoning, language, these are describable mechanics. Extremely complex, sure, but fundamentally not beyond analysis. The dream was simple. If we can understand intelligence precisely enough, we can build it.

SPEAKER_01 6:56

Okay, let's unpack this. Because taking that massive philosophical dream and actually making it testable is where Alan Turing enters the picture.

SPEAKER_02 7:04

The imitation game.

SPEAKER_01 7:05

Right. Five years earlier, in his 1950 paper, Turing asks the big question, can machines think? But he realizes very quickly that trying to define the inner invisible essence of thought is just a bottomless trap. So he pivots, he creates the imitation game.

SPEAKER_02 7:19

He essentially sidesteps the entire metaphysical debate entirely. Turing argues that we should purely look at observable behavior.

SPEAKER_01 7:27

Right. Forget what's inside the box.

SPEAKER_02 7:28

Exactly. If a machine can participate in a text-based conversation so convincingly that a human judge cannot tell it apart from a real person, well, at what point do we just admit the machine is thinking?

SPEAKER_01 7:41

And functionally, that makes a lot of sense to me. I mean, look at a modern chess engine.

SPEAKER_02 7:46

Oh, yeah.

SPEAKER_01 7:46

It doesn't sweat, it doesn't get nervous about its rating, and it certainly doesn't think about strategy the way a human grandmaster does. But it beats the grandmaster every single time.

SPEAKER_00 7:55

Clawless.

SPEAKER_01 7:56

Or like a satellite navigation system. It doesn't know the smell of a city or the history of the streets like a local cab driver, but it routes you around a traffic jam perfectly. If the functional output works, the internal method almost feels irrelevant.

SPEAKER_02 8:11

It does feel that way. But Turing's highly practical approach has a massive, dangerous blind spot. It assumes that acting smart is the exact same thing as being smart. And the danger really arises when we confuse a machine's useful simulated output with actual human comprehension.

SPEAKER_01 8:30

Oh, the black box problem.

SPEAKER_02 8:31

Mm-hmm.

SPEAKER_01 8:38

It looks incredibly capable, so we naturally assume it fundamentally understands what it's doing.

SPEAKER_02 8:43

Precisely. And in 1980, a philosopher named John Searle exposed exactly why that's such a dangerous assumption.

SPEAKER_01 8:50

The Chinese room, right?

SPEAKER_02 8:51

Yeah, using one of the most famous counterweights to Turing's practical approach, the Chinese room argument. I want you to picture this scenario. Imagine a person locked inside a closed room. This person speaks only English and does not understand a single syllable of Chinese.

SPEAKER_01 9:07

Got the picture. A monolingual English speaker in a box.

SPEAKER_02 9:10

Now people outside the room slide pieces of paper with Chinese symbols under the door, and the person inside the room has this massive, incredibly detailed rule book written in English.

SPEAKER_01 9:19

Like a lookup table.

SPEAKER_02 9:20

Exactly. The rule book says if you see this specific squiggle, find this other specific squiggle from the pile on your desk and slide it back under the door.

SPEAKER_01 9:28

So they're doing nothing but matching shapes based on a manual.

SPEAKER_02 9:31

Pure pattern matching. But from the outside perspective, the people sliding the notes are receiving perfectly coherent, brilliant, and contextually accurate responses in Chinese. Wow. They believe absolutely that the person inside that room is a fluent Chinese speaker. But inside, there is zero comprehension. Zero. Searle's central point is that syntax, the arrangement of symbols, is not semantics. Correctly manipulating signs according to rules does not mean you understand what those signs actually mean.

SPEAKER_01 10:01

Okay, let me play devil's advocate here for a second because I think this is where the theory hits the real world.

SPEAKER_02 10:06

Yeah.

SPEAKER_01 10:06

Humans make catastrophic mistakes constantly.

SPEAKER_02 10:09

Fair point.

SPEAKER_01 10:10

A tired doctor misreads a patient's chart. An overworked paralegal misses a crucial liability clause in a dense contract. If this room full of squiggles can summarize a 200-page legal document perfectly or get a medical diagnosis right 99% of the time, isn't that still a massive upgrade over a sleep-deprived human? Why should you, the listener, care if it possesses inner understanding as long as the output is highly accurate?

SPEAKER_02 10:36

If we connect this to the bigger picture, that is a very fair comparison on the surface. But the nature of the error changes entirely when there's no underlying comprehension.

SPEAKER_01 10:46

What do you mean?

SPEAKER_02 10:47

Well, when a human doctor makes a mistake due to fatigue, they still possess a fundamental conceptual model of human biology. They know a leg bone is connected to a knee.

SPEAKER_01 10:58

Right. They have a grounding in reality.

SPEAKER_02 11:00

Exactly. When an AI makes an error, what we call a hallucination, it might invent a medical condition that defies the laws of physics or cite a legal precedent that simply does not exist. It confidently slides the wrong squiggle under the door because it has zero conceptual grounding in reality. And when we use these systems to advise judges, structural engineers, or physicians, the lack of actual understanding becomes a critical safety flaw.

SPEAKER_01 11:26

Because it doesn't know when it's crossed the line from fact into total absurdity.

SPEAKER_02 11:30

Exactly.

SPEAKER_01 11:31

And that illusion of understanding is so incredibly potent right now because modern AI has breached our ultimate fortress, which is language. We always assumed language was our unique territory.

SPEAKER_02 11:42

We did. If we look all the way back to the 17th century, Rene Descartes wrote about machines that might one day physically imitate human actions, but he argued that the ultimate unbeatable test would be open, flexible language.

SPEAKER_01 11:56

Because language requires thought.

SPEAKER_02 11:58

Right. We could accept a machine lifting heavier stones or calculating numbers faster, but language felt intrinsically tied to the soul, to consciousness itself.

SPEAKER_01 12:08

And today, language is the primary interface we use to interact with the machine.

SPEAKER_02 12:12

Which is wild.

SPEAKER_01 12:13

We used to click rigid buttons on a screen or type highly specific code commands. Now we just casually talk to the software. But we have to remember the mechanism underneath. A large language model is not a tiny scholar reading your prompt.

SPEAKER_00 12:26

Not at all.

SPEAKER_01 12:27

It is essentially an incredibly advanced form of autocomplete. It predicts the most statistically probable next word in a sequence based on trillions of examples it's processed. It isn't reading your question, it's calculating token probabilities.

SPEAKER_02 12:40

Yet, because the interface feels so organic, and because it uses the open, ambiguous language Descartes thought was our sacral domain, it triggers our deepest anthropomorphic instincts.

SPEAKER_01 12:51

I can't help it.

SPEAKER_02 12:52

No, it creates an eerie sensation of a mind at work behind the glass.

SPEAKER_01 12:57

So what does this all mean for you as you navigate this technology? The source text brings in Plato here, which I thought was a brilliant lens for this specific problem.

SPEAKER_02 13:06

Plato was intensely obsessed with the difference between mere appearance and true knowledge. In The Republic, he gives us the allegory of the cave.

SPEAKER_01 13:14

A classic.

SPEAKER_02 13:14

Right. Prisoners are chained inside a dark cave facing a blank wall. Behind them is a fire, and objects passing in front of the fire cast shadows onto the wall. Because these two-dimensional shadows are the only things the prisoners have ever seen, they mistake the shadows for the entirety of reality.

SPEAKER_01 13:32

And modern AI is casting some extraordinarily detailed shadows on our screens. You get generated text that cites complex concepts, uses the perfect academic tone, and sounds wildly confident.

SPEAKER_02 13:43

But it might be completely empty underneath.

SPEAKER_01 13:45

Exactly. It's confidently wrong. We mistake the linguistic shadow of knowledge for actual, verified knowledge.

SPEAKER_02 13:53

Which is exactly why we must break away from this relentless obsession with human imitation. If we stop demanding that AI think exactly like a human biological brain, we open ourselves up to a vast new landscape of problem-solving architecture.

SPEAKER_01 14:08

Stop looking in the mirror.

SPEAKER_02 14:09

Yes. We need to look beyond the metaphor we've trapped ourselves in.

SPEAKER_01 14:13

I want to use an analogy directly from the newsletter because it completely changed how I view this. Think about flight. A bird flies, an airplane flies, a helicopter flies, and a drone flies. Right. They all achieve the exact same functional goal moving through the air, but the physical mechanisms are entirely different. A Boeing 747 does not flap its wings to cross the Atlantic.

SPEAKER_02 14:33

It definitely doesn't.

SPEAKER_01 14:34

A submarine does not swim by swishing a tail like a fish. So why on earth do we expect artificial intelligence to have only one acceptable mechanism that perfectly mirrors our own brains?

SPEAKER_02 14:45

We shouldn't expect that. And the broader field of computer science does not. The term neural networks dominates the headlines today, but neural nets are just one single province on a much larger map of intelligence.

SPEAKER_01 14:57

Okay, let's break down a few of these alternative architectures.

SPEAKER_02 15:00

Sure. First, there's symbolic AI. This approach builds intelligence through explicit rules, logic, and highly structured reasoning.

SPEAKER_01 15:13

So that is the ultimate if this then that engine, like uh heavy-duty tax software. It knows every single rule and deduction perfectly, but it possesses zero intuition.

SPEAKER_02 15:24

Yes, exactly. It's exceptional at rigorous math and strict logic, but it's incredibly brittle when faced with nuance or ambiguity. Now, neural AI, which powers the chat bots and image generators we use today, is the exact opposite.

SPEAKER_01 15:38

It's all intuition.

SPEAKER_02 15:39

Right. It's brilliant at recognizing messy patterns like identifying a cat in a blurry photo, but it struggles with strict explainable logic. It cannot easily show its work. So now researchers are exploring the third wave, which is neurosymbolic AI.

SPEAKER_01 15:53

Bringing the two together, how does that practically work?

SPEAKER_02 15:56

Imagine a self-driving car. The neural network component is used to process the messy visual data from the cameras. It recognizes the pattern of a red octagon covered in rain.

SPEAKER_01 16:06

The stop sign.

SPEAKER_02 16:07

Right. But then it hands that recognition over to the symbolic component, which holds the absolute unbreakable logical rule. If a stop sign is detected, you must halt the vehicle. It attempts to combine the pattern recognition intuition of neural nets with the strict safety-critical logic of symbolic systems.

SPEAKER_01 16:25

That makes perfect sense. It bridges the gap between seeing and reasoning, but the text also dies into embodied AI.

SPEAKER_02 16:31

Yeah. Embodied AI operates on the philosophical premise that true cognition requires a physical body interacting with a physical environment.

SPEAKER_01 16:39

Okay, wait, so a robot, basically.

SPEAKER_02 16:41

Essentially. You cannot truly comprehend the concept of heavy, hot, or fragile purely by processing text descriptions of those words. You have to bump into things. You have to feel gravity pulling against a joint.

SPEAKER_01 16:52

You need spatial reasoning that only comes from physics. Okay. But the architecture that absolutely floored me in this deep dive was neuromorphic computing and specifically organoid intelligence.

SPEAKER_00 17:04

Oh man.

SPEAKER_01 17:05

We are talking about actual biological computing.

SPEAKER_02 17:09

The frontier of wetware.

SPEAKER_01 17:10

Literally growing human brain cells, neural tissue in a petri dish, and integrating it directly into computing systems. How does that even execute a computational task?

SPEAKER_02 17:20

It relies on translating signals. Biological neurons communicate via electrical and chemical impulses, right? Right. In organoid intelligence, researchers grow these neural networks on microelectrode arrays. They translate digital data into electrical stimuli, feed those tiny shocks into the biological cells, and the cells naturally react, form connections, and process the input. Wow. The array then reads the resulting electrical output from the cells and translates it back into digital data. We are essentially hijacking the natural computational power of biology.

SPEAKER_01 17:52

Which proves that silicon chips running predictive algorithms are absolutely not the only way to build a functional mind.

SPEAKER_00 17:57

Not at all.

SPEAKER_01 17:58

And we desperately need these strange alternative architectures because the current silicon path is hitting a massive physical wall. I've read that the energy required just to cool the data centers, running modern AI models, is starting to rival the power grids of small cities. We are burning enormous amounts of electricity.

SPEAKER_02 18:17

The power problem is the silent bottleneck of the AI revolution. Modern data centers require specialized hardware, staggering amounts of electricity, and millions of gallons of water just to keep the servers from melting.

SPEAKER_01 18:30

Yeah, they're energy leviaths.

SPEAKER_02 18:32

Exactly. Now, contrast that massive infrastructure with the human brain.

SPEAKER_01 18:36

The human brain runs on what, about 20 watts of power, a few slices of toast, and a cup of coffee can power the most advanced cognitive engine on the planet for an entire day.

SPEAKER_02 18:44

It's a biological miracle of efficiency. I mean, the human brain might occasionally forget where it left the car keys, but its energy consumption is incredibly optimized. If we intend to weave artificial intelligence into every single layer of global society, the sheer laws of physics and engineering demand that we find vastly more efficient architectures.

SPEAKER_01 19:03

Here is where it gets really interesting for me. It all comes down to a fundamental realization about the word artificial. When we hear artificial, our cultural instinct is to assume it means a lesser fake substitute.

SPEAKER_02 19:17

Right, like artificial leather is fake leather.

SPEAKER_01 19:19

Yeah, or artificial flavor is fake flavor. So we naturally assume artificial intelligence is fake intelligence simulating the real thing. But think about light. Artificial light isn't fake light, it genuinely illuminates the room so you can read a book at midnight.

SPEAKER_02 19:34

That's a great point.

SPEAKER_01 19:35

Artificial flight isn't fake flight. A helicopter genuinely lifts you into the air and takes you to your destination.

SPEAKER_02 19:42

That is a vital distinction. It is a very real phenomenon. It's just being produced by an alternative mechanism.

SPEAKER_01 19:48

So AI isn't a parlor trick simulating intelligence. It is actual intelligence, just generated differently. And if we can accept that reframing, it completely shifts our ultimate goal for the technology.

SPEAKER_02 20:01

It moves us decisively away from the trap of imitation and toward the concept of extension. Think about the physical tools that have truly advanced human civilization over the centuries. A heavy-duty construction crane does not try to imitate the complex bones and muscles of a human arm. It simply extends our ability to lift heavy objects. A scanning electron microscope does not imitate the biological structure of the human eye. It extends our vision into the cellular realm.

SPEAKER_01 20:30

It allows us to see things the human eye was never evolutionarily designed to see.

SPEAKER_02 20:34

Precisely. So why are we so fixated on making AI pass human bar exams or write passable human poetry? The true monumental purpose of artificial intelligence shouldn't be to just mimic human thought. It should be to extend it.

SPEAKER_01 20:49

To see we can't see.

SPEAKER_02 20:51

Yes. We should be using it to spot hidden correlations and millions of medical records patterns the human brain would naturally filter out. We should use it to relentlessly challenge our assumptions and simulate a thousand alternative architectural models in a second. It shouldn't replace human thinking. It should radically elevate the baseline of what we can conceptualize.

SPEAKER_01 21:09

Aaron Powell So bringing it all together, what are we actually dealing with when we log onto these systems? AI is not a person. It is not a human mind trapped inside a glowing box. It is certainly not a soul living in a server rack. It is a completely new class of cognitive tool. And what is so beautiful about this entire endeavor is that the very act of building this alien tool is forcing us to turn the mirror around and reconsider what thinking, understanding, and knowledge actually mean in the first place.

SPEAKER_02 21:39

It's reviving philosophy as a hard engineering necessity. Asking what constitutes true knowledge or how we define responsibility, these are no longer abstract academic questions debated in a classroom. They are mandatory design requirements for the future.

SPEAKER_01 21:54

It is such a powerful way to look at the tools we use every day. Listeners, we cannot recommend this newsletter issue enough. Please remember to go read and subscribe to the full article, AI, Philosophy, and the Many Shapes of Thinking. You will find the visual breakdowns of these architectures incredibly helpful. And before you go, remember that while this deep dive is completely free, it is a massive help to us if you press like on this track and subscribe to the feed so you receive all our future issues. It truly allows us to keep doing this work for you.

SPEAKER_02 22:21

As we step away from the mic today, I want to offer you one final thought to mull over. We've spent this time discussing AI as a tool for extending our minds, much like a microscope expends our eyes. Right. But cognitive tools have a unique way of intimately blending with our own identities over time. If artificial intelligence eventually becomes a natural, seamless extension of our daily thought process, a few decades from now, will we even be able to tell the difference between our own original thoughts and the ones we co-created with our artificial extensions, who ultimately is really doing the thinking?

SPEAKER_01 22:54

Are we the ones genuinely asking the profound questions, or is the machine just quietly helping us find the words? It certainly makes you look at that blinking cursor on your chatbot in a whole new light. Until next time, keep diving deep.

SPEAKER_00 23:25

If you found this episode valuable, feel free to share it with someone who might benefit. A colleague, your team, or your network. You can access all episodes by subscribing to the podcast and find their written counterparts in the Agile Software Engineering newsletter on LinkedIn. And if you have thoughts, ideas, or stories from your own engineering journey, I'd love to hear from you. Your input helps shape what we explore next. Thanks again for tuning in, and see you in the next episode.