Author’s Note: This is the first in a four-part series that will explore aspects of innovations in Artificial Intelligence (AI). Part I below will outline an approach to assessing technological development in general followed by a contrast of artificial with human intelligence. Part II will explore aspects of business and consumer use of AI, with a prognosis of these developments. Part III will examine the possibilities for AI upon human flourishing, while the concluding fourth instalment with examine the potential dangers.

*     *     *     *     *

Technological innovation has a habit of breeding a backlash. From the luddites and the saboteurs to violence in video games, each wave of invention tends to invite anything from mild disdain to fervent opposition.

A streak of this opposition is the allegation that technology “causes” bad things to happen – to which our digital age is no stranger. For instance, economist Paul Craig Roberts pulls few punches with this recent lament:

From observation, I conclude that the costs [of the digital revolution] are sharply rising, and the benefits are declining. Indeed, many claimed benefits, such as students using AI to do their assignments and, thereby, never learning any skills, such as how to write a theme, solve a math or physics problem, in fact create an ignorant population devoid of ability to function independently of technology. They have no ability to even know if the information provided to them by AI is correct. Their minds are totally controlled by whoever programs the software.

Such thinking romanticises the past in order to plea for a return to “simpler” living that will “save ourselves” from the alleged vicissitudes of too much tech. Roberts may harken back to the “analogue age” when, according to him, problems were solved with a “three-minute telephone call answered on the third ring”. But were people any less docile and stupid during this time? Were they not “totally controlled” by whomever wrote the newspapers or ran the broadcasting stations? Seen in this light, Roberts’ complaints wind up being little more than the twenty-first century equivalent of a parent moaning about the kid’s calculator.

The real “ugly truth” about technology and inventions is that they are responsible for nothing, as they have no agency. One can use a knife either to prepare dinner or to stab someone to death. The knife makes no demand either way. If technological shortcuts make people lazy it’s because they were already inclined to be so. If social media makes people vain and vacuous it’s because their lives were already devoid of meaning – they have simply found an outlet for it. Far from creating problems, the most any development can do is amplify or bring out into relief that which was already there. Our true fear, therefore, is what technology might unleash within ourselves.

The faux-nostalgic “return to simplicity” – much like veganism and “green” energy – is often just another luxury belief dressed up as moral clarity. Such advocates typically crave not anti-materialism, merely curated materialism. A seafront cottage in North Norfolk with artisan bread, reclaimed oak, and a Peloton beside the Aga. In the same way that constructing wind turbines is possible only through burning hydrocarbons, romanticised simplicity is achievable because it is subsidised by what it claims to reject. Looking pious about eschewing progress is easy when you’re parasitically adjacent to all the soot and smog that makes such living possible. Even the Amish – always the poster child of technology sceptics – benefit from their immersion in industrial society: safe, stable infrastructure, emergency healthcare systems, agricultural supply chains, legal protections, and economic spillover.

If there’s any rewinding of the tape at all, it’s only to a point at which the complainer happens to feel comfortable. Roberts peers back thirty to fifty years. His father may have preferred the 1940s or 50s, his grandfather the 1910s or 20s. Moreover, I was reading Roberts’ article on a tablet. Most of his output is published online. So apparently that part of the digital revolution gets a pass?

In contrast, true simple living is eighteen hours workdays just to scrape enough food to eat. Rotten teeth, premature death, and zero backup if your crops fail. No electricity, no central heating, no supplied gas or water. No cute mugs saying “live laugh love.”

Escaping from such hellish conditions has always been crushingly difficult to resist. One need cite only the developing countries who don’t give two hoots about “climate change” in their race to industrialise. Those who turn their nose up at the material world swiftly discard their intangible or ethereal values as soon as acute suffering takes hold. For instance, Gandhi – always the philosophical tourist – wasted no time in checking himself out of sackcloth simplicity to undergo a hospital appendectomy.

For practical purposes, therefore, technological development is “inevitable”, and there is little use in complaining about it. But the extent and nature of its influence can be determined not by looking at the technology itself, but at ourselves. Grit or gold, what is already within us will be magnified by the innovation and ultimately determine the latter’s influence.

The Nature of Artificial Intelligence

It is with these thoughts in mind that I wish to assess the apparent “revolution” in Artificial Intelligence (AI) – an innovation which, at the extremes, is either pooh-poohed or spreading panic over the prospect of human extinction.

Before we dive in, a full disclosure: I am an active AI user, having used generative AI, both as a businessman and as a consumer, for a little over a year now. Overall, I regard myself as a cautious enthusiast. Much of what follows – in this and later parts – is therefore not merely academic, but is unavoidably a product of my own experiences, focussing mostly on large language models (LLMs). I am not, however, an out-and-out tech expert who has examined exhaustively every single detail of this topic. Nevertheless, I hope everything in this, and in later parts, is a reasonable and representative discussion of how AI works, and what it can and cannot do.

The first port of call in our examination is to recognise, first off, that AI’s significance lies not so much in its ability to undertake repetitive or routine tasks. Machines have been replacing humans in that regard for centuries, freeing up labour for new endeavours and industries that were previously unimagined let alone explored. As such – in spite of local disruption at the time – railways, motor cars, computers, aeroplanes and the like have always been a boon to humans in the long run. AI’s similar accomplishments in labour saving advances may themselves constitute a revolution. But however giant the leap, it would be one of degree rather than of kind.

Moreover, any danger from these advances stems less from the technology itself than from the system absorbing it. AI isn’t replacing human labour in a free, organic marketplace. Rather, it’s being weaponised within a rigged order: where inflationary finance distorts capital allocation, where regulatory moats shield monopolies, and where IP laws grant permanent gatekeeper status to firms that own the means but serve no ends but their own. The real question is not whether machines will do work, but whose work they’ll do. Will AI be directed toward industries that serve genuine consumer wants, or hijacked for wasteful boondoggles and coercive agendas? Will it expand individual flourishing and allow labour to move freely into new sectors? Or will it tighten the leash on “useless eaters” through a blend of cheap goods on welfare, mindless entertainment and algorithmic obedience? AI becomes a threat not when it automates tasks, but when it automates power – concentrating control over knowledge, production and value in the hands of firms already bloated with political protection.

The more interesting issue, however, is AI’s (alleged) ability to undertake non-repeatable actions: those involving creativity, imagination, judgment and decision-making. Acts whose essence cannot be reduced to a communicable process or list of instructions, and seem to define what it is to be human. Acts which – if taken over by a superior intellect – could reduce humanity to the level of a toddler in a room full of professors.

AI’s abilities in this regard largely rely upon the field known as machine learning. While the execution of machine learning maybe a complex endeavour in terms of programming and coding, we’d gained a theoretical understanding by the 1960s. We simply lacked the necessary hardware and processing power required to undertake the tasks at scale. Today’s AI breakthroughs therefore owe themselves more to developments in infrastructure than in computer theory. In fact, given AI’s gargantuan thirst for energy, raw materials and water, the limits to its capabilities may end up being physical rather than conceptual – a bit like wanting to play a compact disc when you can only afford a cassette player.

Anyhow, machine learning involves a program – an AI model – being fed a data set before it analyses that data set for patterns. The larger the data set, the more accurate the analysis of those patterns. (Hence the desire to “train” AI models on as much of the world’s information as possible). When the model is subsequently fed new data – say, you ask it a question – the model analyses the new input to assess its pattern. It will then identify a response that most closely aligns with the patterns it recognises in the existing data. Such patterns extend not only to the identification of cold, hard facts, but also style, tone, voice, humour and so on. An added layer of randomness means that individual outputs usually sound sufficiently varied, inventive and nuanced, and is responsible for much of AI’s “creative” abilities.

For instance, if you ask a model to write a joke in the style of Ricky Gervais about gym goers, it will scan your request, identify patterns in how Ricky Gervais typically speaks, how gym culture is discussed, what kinds of punchlines people respond to – and then stitch together a response that best matches those patterns. What results is something that resembles that which tends to make Ricky Gervais funny when certain words, tones and structures are involved.

The interaction, by a user, with an AI model over time also constitutes fresh data which can be used for pattern analysis and shape future outputs. Hence, a model is usually able to converse with you in your own tone of voice, and more accurately provide you with your specific responses based on how you speak or write, or what you tend to prefer. An important part of this is the model’s ability to absorb pragmatics, or contextual meaning. That is, the meanings you intend to convey stated not in the literal words you use, but in the context of the specific conversation.

Now, all of this is, of course, a gross simplification of the complexity involved in making AI a reality. Beneath the surface lie countless other advances: turning words into navigable maths; tracking context across long passages; fine-tuning via human feedback; reinforcement learning to clarify outputs; and so on.

But the basic point remains: machine learning – and its most potent variant, deep learning – remains a feat of computation: a sophisticated data processor, not a living thing. And the implications of AI cannot be understood correctly without first grasping this framework.

AI vs Human Intelligence

The most tempting response to the description above is to suggest that AI neither knows nor understands what it is saying, nor has any comprehension of why the answers it gives are correct or appropriate.

For instance, if you were tell it a joke, then the model would spot the structure, the setup, the punchline, the semantic twist – even which types of people statistically laugh at it. It might then respond “Haha! That’s hilarious!” with perfect timing. But it wouldn’t actually find the joke funny because it doesn’t feel surprise, irony or absurdity. It just calculates, from a billion examples, that “Haha! That’s hilarious!” is statistically the right noise to make here.

Machine learning is therefore little more than training a parrot with a photographic memory. You feed it with examples until it starts guessing patterns that we (humans) will recognise as correct. But as far as the parrot is concerned, they’re just the noises that get the peanut. It’s neither thinking nor understanding as much as statistically bullsh*tting with confidence. And the more data you shovel in, the better it gets at sounding clever – but it still has zero clue why anything it says means anything at all.

In short, AI is prediction that mimics the appearance of comprehension without actually comprehending. It looks alive but it’s dead behind the eyes – useful to minds, but not a mind itself. And as we’ll see in Part III, that distinction may prove critical to our approach to AI and the impact it may have.

However, while all of this is cogent and relatable, the assumption that statistically probable responses aren’t, at least, a category of “knowing” ends up begging the question. One could, after all, point out that much of human understanding also involves surface pattern mimicry rather than a deep grasp of logic or causality. If someone says “good morning” to you, you are likely to say “good morning” back without necessarily comprehending why – you just do it.

So what really is the essence of how AI fails to know, understand or comprehend? What does it really mean to “know” something in the first place? And can we answer these questions without disappearing down a metaphysical rabbit hole?

The difference is that human knowledge is structured around the fact that we are actors – beings experiencing dissatisfaction, and using scarce means to remove that dissatisfaction.

At the heart of every one of our actions is the simple fact that something about the world isn’t right. Not just obvious physical needs like hunger or tiredness, but an unceasing train of itches to alter the configuration of reality. If no such dissatisfaction existed, there’d be no reason to learn, choose or act. All human knowledge – even abstract knowledge – is therefore ultimately anchored in this inescapable endeavour: reducing discomfort and securing satisfaction through grasping cause and effect. We then structure and comprehend that knowledge accordingly – even down to the way in which we arrange sentences by subject (actor), verb (action) and object (means/ends).

These feelings of dissatisfaction leading to knowledge and action cannot (yet) be reduced to a definitive cause. We do not know which (if any) parts of our physiology or external stimuli cause the choices we make. Such choices are an original fact which cannot be subject to further analysis. This is not true of AI output – everything it does can be traced back to its training data and programming.

The real gap between humans and AI, therefore, is one of value-infused agency. Left alone, ChatGPT never says: “Right, here’s what I need to tell you,” because it has no inner pressure, no curiosity, no discomfort it’s trying to resolve, no itch to reach out, no urge to be understood. It just waits, like a loaded gun, until you pull the trigger.

When it does respond, it doesn’t choose its responses in the Misesian sense. It doesn’t say “I want to say X to achieve satisfaction Y.” It just runs a prediction. Here, then, is the difference with our “good morning” example. Say “good morning” to an AI model and it replies “good morning” back because that’s the pattern – a silicon oracle parroting statistical empathy. But when you respond with “good morning”, that’s not just mimicry but a choice with purpose. It has value: maintaining the social fabric, smoothing a relationship, and fitting a custom that promotes social co-operation – even if you couldn’t necessarily articulate that as your reason. And if you didn’t want that outcome – if you despised the other person, or wanted to rupture the ritual – you could just as easily say “f*ck off.”

Either way, the essence of human comprehension is not just copying the right shape of words, but feeling the weight of choosing them. Everything we see, hear and say connects – even if only imaginatively – to some form of dissatisfaction and its potential resolution. If a person says to you “I am hungry,” and you respond with “have some bread”, our comprehension followed by response isn’t based on a mere calculation of statistics. Rather, we’re connecting directly to a fear of pain, starvation or death, and what bread does to remove that fear. AI can simulate how we respond to those signals, or map what humans tend to value. But it neither cares about nor feels the pull of hunger and satiation. It doesn’t want – and without wanting, there’s no comprehension, only correlation.

This extends also to our understanding of abstractions. The second law of thermodynamics, for instance, describes the tendency toward entropy (disorder). But “order” and “disorder” are value judgments. A lump of coal is ordered because, as concentrated energy, it’s useful to us. Scattered ash dancing in the wind is disordered because its energy is no longer usable. Or, rather, there is no value for us in being able to distinguish each possible configuration of scattered ash particles, so we lump them all together under the heading of “random”. So even scientific laws, though value-free in content, are interpreted through a value-laden lens.

This has significant ramifications for what AI can achieve. Suppose, for instance, you were to ask an AI model “Is the Mona Lisa beautiful?” That model could analyse all of the world’s opinions and judgments on this question, and produce a reasonable answer. It could tell you in a heartbeat all of the criteria deemed by humans to be relevant in determining whether the Mona Lisa is beautiful – factors that you, as a single human, may be unaware of. That data may sharpen your own perception and understanding of beauty and where to see it. But the model itself will never be able to give its own original, gut felt opinion on whether the Mona Lisa is beautiful, let alone say why. It will never be able to craft a new theory of beauty based on what it recognises and finds beautiful in the painting (or does not) because neither the concepts of beauty nor ugliness produce in it any inner dissatisfaction in need of adjustment. Nor, as a corollary, could it ever says whether it prefers viewing the Mona Lisa to looking up in the Sistine Chapel. Or to eating a hamburger, for that matter.

To any one person, therefore, AI can seem like a god – churning out knowledge, brilliance, wit and insight faster than we can blink. But its inability to produce original value means that, to humanity as a whole, it’s a lagger. It invents nothing nor creates anything new. It just reshuffles what we already gave it following syntactic patterns designed by, and suitable, for human actors. The scope may be vast, but the source is always us.

True enough, humans build on past knowledge. Sir Isaac Newton, for instance, drew on Kepler, Galileo, Descartes and others. Each of us stands on the shoulders of giants. Theoretically, AI could take what we already know and produce something more wonderful and brilliant.

But there’s one critical difference:

Human accomplishment consists not only of accumulation, but direction. The sum total of all knowledge in the universe – and everything that’s physically possible – is value-free. But the pursuit and use of that knowledge is not. It consumes time, attention, energy, risk – all of which are scarce. Some paths are worth more to us than others, not because they’re more complex, but because they matter more. We didn’t harness the laws of propulsion by accident – we did it because we wanted to move. We had a reason, a goal, an urge that sorted which truths to chase and which to ignore.

Human discovery isn’t, therefore, a repetitive mix of eggs, flour, butter and sugar to find a tastier cake. It’s knowing whether to use those resources to bake a cake at all. Less “how do we move forward?” and more “are we even on the right rails?” And answering that question is not something that can be reduced to the sum total of everything that has been recorded. Rather, it changes moment to moment with each new situation we face. That’s why humans can break patterns, contradict mentors, hallucinate, misinterpret, dream up wrong answers that somehow become “right” – because “right” isn’t just about what’s technically correct. It’s about what’s valuable – what serves to best remove the most urgent, inner feelings of discomfort with what we have available.

AI, however, never experiences the pangs of scarcity. It does not know what it means to gain something more important by discarding something less. It is essentially a wraith that never has to feel the import of choices that humans have to make. AI could wile away the hours slicing, dicing and recombining all information in existence in the hope of imagining something new. But it would never know to what purpose that effort should be directed or the value of any result. Such efforts could discover a technically able cure for cancer. Yet it would have no idea whether that suggested solution is worth the cost of harnessing it.

AI can therefore only ever blend the known according to purposes granted to it by humans (and may do that very well). It stays within the data it was trained on, which means it’s bound, always, to the sum total of existing human output and values. So even if it “goes farther,” it can only ever travel on rails we laid down. And unless it develops its own sense of value, preference and purpose, it won’t ever leave those rails.

Of course, we may one day be able to program a model to behave like it values something, or maybe we could manufacture that inner sense of discomfort. But this would still be at our direction, and owe itself fundamentally to causes that we control. AI can follow the dance, but it never makes the music.

Now, I’m not saying that the above accounts for every difference between AI and human intelligence. One may well wish to discuss whether humans are infused with a soul, or to explore more deeply the nature of consciousness. That deeper understanding may well be necessary in order to remind ourselves that AI’s performative abilities can mask its true nature. However, this praxeological approach has clarified the key problems without needing to resort to metaphysics or mysticism.

It’s these clarifications that have brought us full circle: that if AI is reliant upon us for its knowledge, direction and values, then the true nature of the AI revolution can be understood primarily by looking at ourselves – and to the motivations of states, innovators, businesses and consumers – rather than the technology. And that is what we shall begin to explore in Part II.

==> Go to Part II.