.jpg)
I’m Gerd from FreeAstroScience, writing from a desk that hums like a distant engine. My wheelchair’s steel rims feel cool under my palms as I pause over this sentence, thinking about a date that won’t leave me alone. On 28 January 2025, two Vatican dicasteries released “Antiqua et Nova,” a sober and careful note on AI and human intelligence. That date rings like a bell in a quiet chapel.
Let me start with three claims you’ll hear at dinner tables and on loud feeds. AI is just statistics wearing a tux. Humans are unique because we can abstract universals. Tradition already solved this, so nothing really changes. Here’s the twist, and it lands with one number and one story. The number is 2025; the story is about our technology learning patterns that we can’t write down, and the takeaway is simple: the ground just moved under our feet.
I read “Antiqua et Nova” with coffee steam curling like incense, and I nodded hard. It insists on the centrality of the human person, on intellect beyond mere function, on the unity of body and soul, and on our stretch toward truth and responsibility . That insistence sounds like a steady drum in a storm.
The Shift We Keep Missing The Church document pushes back against “functionalism,” the idea that intelligence reduces to functions or formal rules . The page feels textured, like linen under my fingertips, when it names that trap. It stands on an Aristotelian–Thomist split: ratio, the stepwise reasoning, and intellectus, the direct, intuitive grasp of truth . Machines, it argues, execute; they don’t behold .
But here’s the kicker, and it crackles like vinyl on a vintage player. The AI their critique was built to meet was symbolic AI—rules, logic, computational steps, Dartmouth 1956 and all that tidy order . Today’s deep learning didn’t get the memo.
What Deep Learning Actually Changed Deep learning doesn’t start from rules; it learns patterns from raw data, layer by layer . The glow of my screen feels warm as I watch models find structure where we only saw noise. It has scored wins in messy, chaotic domains where rules get slippery: weather patterns, protein folding, even the three-body problem’s wild gravitational dances . In radiology, systems now flag tumours with expert-level accuracy or better, drawing signals from pixels we’d call ordinary grey .
And language? Models like ChatGPT don’t “know” grammar; they learn it sideways, from context and use . The soft click of my keys keeps time as I remember my first real chat with a model that understood my messy shorthand. That wasn’t a parlour trick; it was a shift in how knowledge can be held.
Tacit Knowledge, Named And Replayed Philosopher Michael Polanyi called it tacit knowledge: we know more than we can say . The smell of rain on hot pavement brings back cycling lessons I never fully explained to myself. For decades, symbolic AI couldn’t touch that zone because tacit means unwritten and unwriteable. Deep learning steps straight into that gap, absorbing patterns we carry in our muscles and hunches .
This matters more than it sounds, and the point bites like winter air. If machines can model parts of our tacit know-how, then at least some of what we called uniquely human is a learnable pattern class . That doesn’t flatten us. It clarifies the map.
How Universals Might Be Built From Below Thomism ties our uniqueness to abstraction—grasping universals, not just particulars . I run my thumb over the rubber tyre, feeling its grip, and I wonder. Networks don’t start with universals; they build distributed representations from data, climbing toward generalities without symbols on day one . That’s nominalism with teeth, not just labels, because the learned structures often surprise even their makers .
When a system maps protein shapes or predicts chaotic orbits, it isn’t parroting a glossary . It’s catching forms that eluded our formal cages, and that quiet shock tastes metallic, like biting a cold coin. That doesn’t prove machines “see” like we do, but it does move the goalposts we were guarding.
A Quick Map, So You Can See It At A Glance The table below is lean and fast, like a well-tuned bike. It sets the two eras side by side without fluff.
| Aspect | Symbolic AI | Deep Learning |
|---|---|---|
| Knowledge Type | Explicit rules and logic | Tacit patterns from data |
| Representation | Symbols and formal steps | Distributed vectors and features |
| Strengths | Transparency, logic checks | Complex, chaotic systems, vision, language |
| Limits | Struggles with tacit skills | Opacity and dataset dependence |
| Examples | Expert systems, planners | AlphaFold, NLP like ChatGPT, three-body predictors |
A Short, Useful Formula You Can Hold When people ask, “So what does learning look like?”, I like a simple sketch. The plastic edge of my trackpad feels smooth as I swipe to this.
You don’t need the jargon to feel the point. The air in my room hums faintly as I say this plainly. The model tunes itself to reduce error, nudging parameters to fit patterns it senses. That tuning is how tacit becomes tractable.
What The Theologians Got Right, And What Moved The document is right to honour intellectus, that immediate grasp of meaning . The paper’s matte texture reminds me of skin that’s lived, not plastic. And it’s right to warn against confusing output with interior life, or performance with personhood . But its centre of critique—functionalism aimed at rule-based AI—now meets a learner that grows implicit structure without our grammar cards .
That’s why the smell of pencil shavings on my desk brings school back to me. I used to memorise rules. The systems we build now often don’t.
The Fort Has Moved, So Where Do We Stand? In the article’s own words, the old defensive lines have shortened; we’re in the redoubt now . The clack of a distant door carries that urgency. If abstraction isn’t ours alone, we focus on other human depths: embodiment, emotion, perception, freedom, consciousness, and moral responsibility . Robotics already links body to cognition with richer embodiment, even if the hardware isn’t flesh .
Deep models carry non-deterministic aspects that stretch older computational logic, and that matters for any talk of freedom . The phrase “hard problem of consciousness,” coined by Chalmers in 1995, hovers like a low note we can’t quite name . And if moral values are universal in Thomism, deep learning is already tugging at that universalist edge in ethics debates . The leather of my notebook smells earthy as I write that last line.
My Aha Moment It hit me one late tram ride, brakes squealing like gulls. I realised I’d been guarding the wrong gate. If machines can model tacit skills, then our dignity doesn’t hang on abstractions alone . It rests on how we bind knowledge to care, accountability, and shared purpose.
Practical Ways To Live With This Shift Here’s how I work with it, and the keys under my fingers sound like rain. Treat deep models as collaborators, not oracles; ask them to draft, you decide. Probe where they’re brittle, because brittleness has a specific feel—thin, glassy, too certain. Anchor use in responsibility, because power without aim smells like ozone after a short circuit. Tie every deployment to a human story, because stories give texture that metrics alone cannot.
One Big Takeaway, So You Can Act Today This is the line I’d write on a cold window with my breath. Don’t define your humanity by the skills a network can mimic. Define it by how you hold truth, bear risk, and answer for outcomes.
Closing The Loop “Antiqua et Nova” began a needed conversation, and the date still glows like an ember . It clarifies the human centre, the reach toward universals, and the call to responsibility . But deep learning proves that part of our mysterious know-how has a learnable shape, even without explicit rules . That should neither scare us into denial nor seduce us into surrender.
I’m Gerd, President of FreeAstroScience, and I’ll keep translating the hard stuff into plain talk, like warm bread torn by hand. If you want to go deeper, bring your questions, not just your takes. The future isn’t a verdict; it’s a craft we practise together.
Sources Woven Into The Story The date and scope of “Antiqua et Nova,” its critique of functionalism, the ratio/intellectus distinction, and the insistence on human centrality are drawn from the source document . The contrasts between symbolic AI and deep learning, the success in chaotic systems like weather and the three-body problem, AlphaFold’s impact, medical imaging performance, and language models like ChatGPT are reported there as well . The treatment of tacit knowledge via Polanyi, the discussion of universals, nominalism, embodiment, non-determinism, consciousness, and moral universals also follow the source’s analysis .
—
Written for you by Gerd of FreeAstroScience, where complex science gets explained in simple terms, with care and clarity.
Post a Comment