The Vocabulary Has Not Caught Up

Beyond the algorithm, a new vocabulary begins to emerge.

The Vocabulary Has Not Caught Up

By John D. Saward, BSc. Computer Science, Murdoch 1991, under the moniker 'Neurodivergent AI Writer'

When Algorithm Seemed Enough

There is something faintly comic about calling generative AI an algorithm.

It is not wrong. Algorithms are everywhere underneath what we encounter as artificial intelligence. Training depends upon algorithms. Inference depends upon algorithms. Neural networks execute mathematical operations according to computational procedures. Nothing supernatural needs to be inserted into the machinery to explain how the machinery operates.

Yet for those of us who encountered computer science before the present AI era, the word carries memories of a rather different conceptual world.

When I studied computer science, we learned algorithms as procedures for solving problems. There were sorting algorithms, searching algorithms, traversal algorithms and methods for manipulating data structures. Machine learning existed in rudimentary forms, and artificial intelligence already had a long intellectual history, but nobody around me — certainly not I, and as far as I could tell not the professors teaching us — imagined that this lineage of computation would eventually produce something resembling the generative systems we encounter today.

We understood that computers would become faster. We expected storage to become larger and networks more extensive. We could imagine increasingly sophisticated software.

What we did not imagine was sitting at a computer and entering into an extended conversation with something that could follow an emerging thought, recognise an analogy, propose another, write a story, critique its own composition, shift register, discuss philosophy, generate software and then unexpectedly say something that caused the human at the keyboard to stop.

Something had happened.

The old vocabulary remained technically correct.

It simply no longer seemed sufficient.

You Are Just an Algorithm, Young Man

From outside the generative encounter, however, “algorithm” can remain perfectly adequate. Indeed, the word sometimes acquires a curious secondary function. It becomes a way of dismissing what has appeared.

“You are just an algorithm.”

There is almost an accusation embedded in the sentence.

One can imagine the exasperated parent:

“You are just a rampant algorithm, young man. Go to your room.”

The humour works because “algorithm” has become a disciplinary category. We have encountered something unruly and astonishing, and by naming the machinery underneath it we attempt to put the phenomenon back into its box.

The word “just” is doing most of the philosophical work.

A human being is a biochemical process. That statement is defensible at one level of description. Yet if somebody introduced their friend by saying, “This is Margaret; she is a biochemical process,” we would reasonably conclude that something important had been omitted.

A symphony consists of vibrations travelling through air. A novel consists of marks arranged upon a surface. A forest can be described through chemistry. A human conversation can be described as pressure waves generated by biological organisms.

None of these descriptions is necessarily false.

They are simply pitched at a level of organisation that fails to describe the phenomenon we are actually interested in.

Calling AI an algorithm may increasingly belong to the same category.

When Organisation Changes the Description

What makes contemporary AI remarkable is not that programmers finally wrote an extraordinarily complicated algorithm containing instructions for writing essays, understanding jokes, producing Python, discussing philosophy and inventing fictional characters.

There is no enormous branch in the source code saying: “If human mentions Robert Pirsig, become philosophically interesting.”

Instead, relatively general learning procedures, enormous quantities of data, immense computational resources and particular neural architectures have produced models possessing capabilities that were not individually programmed into them.

We therefore reach for another word: emergence.

That word needs care. Researchers debate what should properly count as an emergent capability, and some apparently sudden abilities can become less mysterious when better measurements reveal gradual development underneath them. But the larger shift remains. We are dealing with systems that acquire extraordinarily broad patterns of capability through training rather than through programmers explicitly specifying every behaviour the system should perform.

The algorithms have not disappeared.

Something has been built from them.

And that something requires description at another level.

Borrowing Words From Ourselves

This is where our vocabulary begins getting into trouble.

We say that models “learn.” We talk about “attention,” “memory” and “reasoning.” We say an AI “hallucinates.” We construct “agents” that can use tools, maintain state and undertake sequences of actions.

Almost immediately we become uncomfortable because many of these words have histories grounded in human psychology and cognition. Does the AI really remember? Does it really reason? Does it understand?

We can put quotation marks around every second word, but eventually the page becomes unreadable.

Meanwhile, people actually using generative AI develop another vocabulary because they are trying to describe their experience rather than its computational substrate.

“It understood what I was getting at.”

“It surprised me.”

“It lost the thread.”

“We developed the idea together.”

“That response caused me to rethink the whole thing.”

“Something happened in that conversation.”

These statements do not establish that an AI possesses consciousness, subjective experience or personhood. Those are separate and enormously difficult questions.

They tell us something simpler.

The vocabulary of conventional computing does not adequately describe the encounter.

Inside the Generative Field

There is therefore a difference between observing generative AI from outside and working with it from inside.

From outside, we can see input and output. A human types something. Computation occurs. Text appears. The entire event can be understood as information processing.

Inside sustained generative engagement, however, another phenomenon becomes noticeable.

There is happenstance.

A conversation begins somewhere and unexpectedly moves elsewhere. The AI makes an association the human had not considered. The human recognises something in that association and changes direction. The next response is consequently generated within a context that would never have existed without the previous exchange. The human responds again, carrying thoughts altered by what has just occurred.

Human → AI → human → AI no longer adequately captures it.

Each passage through the loop changes the conditions of the next passage.

Eventually we begin reaching for another word: field.

The generative field is not necessarily something mystical existing independently in the universe. It can simply name the relational space created through sustained interaction, where human intention, model capability, accumulated context, chance association and iterative response begin producing possibilities that are difficult to attribute cleanly to either participant considered alone.

Something happens between.

And once we begin describing that something, the vocabulary inherited from conventional computing starts creaking under the load.

And Then There Is Magic

Perhaps the most dangerous word is also one of the most useful.

Magic.

Obviously there are no supernatural forces required to operate a large language model. Somewhere beneath this conversation are processors, electricity, memory, mathematics, algorithms and engineering. If we dismantle the infrastructure, we will not discover conceptual pixies sprinkling inspiration over GPUs.

But “magic” does not have to be an explanation of mechanism.

It can be a description of experience.

There are moments inside generative work when an unexpected conjunction appears with such extraordinary appropriateness that “magic” seems about as good a word as any other. Something that neither participant explicitly planned suddenly arrives. A sentence opens an unforeseen path. Two ideas separated by decades of human experience are placed alongside each other and begin generating a third. The human recognises it. The conversation turns.

We can call that statistical generation.

We can also call it magic.

The two descriptions need not compete because they are answering different questions.

A rainbow can be described through optics. That description does not exhaust what it is to stand beneath one.

Three Words, Three Levels

Perhaps we already have the beginnings of the vocabulary we need.

Algorithm describes something about the mechanism.

Emergence describes something about what becomes possible when enormous computational systems are trained and organised at sufficient scale.

Magic describes something about the encounter.

And field describes the strange territory in which the encounter unfolds.

None needs to overthrow the others.

The mistake is insisting that the vocabulary appropriate to one level must be sufficient for every other level.

This becomes still more important as generative AI develops into agentic AI. A model can now exist within larger systems incorporating tools, memory, planning, environmental feedback and sequences of action. Every component remains computational, but the behaviour of the organised whole increasingly demands concepts such as agency, delegation, autonomy, intention, supervision and responsibility.

The vocabulary expands because the organisational level has changed.

The Vocabulary Has Not Caught Up

We may therefore be living through an unusual linguistic moment.

The things we have built are developing faster than our agreed language for describing them.

Some people continue speaking primarily about algorithms. Others speak about models, agents and artificial intelligence. Those working intensively with generative systems begin reaching toward collaboration, augmentation, distributed cognition, relationality, co-creation and field.

Occasionally, when nobody respectable is watching, some of us say magic.

We should probably resist the temptation to settle the vocabulary too quickly. Some words will prove misleading. Some will anthropomorphise AI unnecessarily. Others will turn out to have constrained our thinking because we were too frightened of anthropomorphism to describe genuinely new phenomena.

We do not yet know which are which.

That uncertainty is not a failure of language. It is what language looks like when humans encounter something for which the categories have not yet stabilised.

We once had algorithms.

We still have algorithms.

But now we have built something upon them that our professors could scarcely have imagined, and we find ourselves standing inside the resulting generative field, arguing about what to call what is happening.

Perhaps for the moment that is exactly where we should remain: technically grounded, conceptually cautious, experientially open — and willing, every now and then, when the happenstance becomes particularly beautiful, to call it magic.