Those of us who lived through the 90s remember all too well the pervasive rhetoric of the time. One oft-repeated slogan was that 'any company without an internet strategy will be dead in 6 months'. Only in hindsight do we understand the considerable financial and organisational resources that were brought to bear over time, and the successive failures that occurred on the path of transformation initiated by internet technology. Transformation didn't happen overnight — instead, progressive change happened over a period spanning 2 generations, and these changes continue today.
#1 The name
The implicit message we are exposed to on a daily basis is that artificial intelligence has the ability to stand in for almost everything, offering immediate results in almost any environment. While we entertain these vague claims, it would be good to examine them through first-hand experience. To state that a machine has intelligence is a grandiose claim, and we should use it and examine it critically in order to form our opinion, and to prioritise what we have assessed over what the experts (self-appointed or otherwise) claim.
Intelligence is the implicit domain of humans, and it means many things to many people. While a machine that performs mathematical and analytical feats might be expressing intelligence, we should be ever more cautious of the unintended consequences of such a remark. Intelligence lives far beyond the capabilities of algorithms, even the most sophisticated ones, and we may be well-advised to avoid the name artificial intelligence, lest it misinform us.
In what way does the name artificial intelligence inform or misinform us? It implies that an LLM produces the self-same attribute vested in humans, with the only qualifier being that it is artificial, and the name blurs the human-machine boundary in a way that is significant. While a machine substitutes or augments human effort –which is what an LLM does–, artificial intelligence implies something entirely different; in what way does the chosen metaphor attempt to reframe the machine, and what other conclusions might we (falsely) draw from it.
The message the name sends is that human and artificial intelligence are interchangeable on a one-for-one basis. A machine is traditionally defined and named by the function it performs; an LLM mimics the sphere of activity that is based in language, and interacts through language, creating the appearance of intelligence.
#2 The machine
By choosing to refer to the thing by one of its functional names –Large Language Model, Transformer Models or Autoregressive Models–, we express its meaning accurately. LLMs aren't intelligent, and they don't replace human intelligence. They are flexible and capable language-based algorithms that perform intelligent functions, and they do so in response to prompts (instructions).
By understanding that LLMs don't substitute intelligence, and that they don't replace the work that people do, we give ourselves the chance to truthfully assess the capabilities of the new machine. While we can, indeed, point to immediate and proven advantages of using LLMs, the problem that arises is the propensity to generate an endless sequence of seemingly rational extrapolations of what it can (or will) do from the little evidence we have recently accumulated.
An LLM is a diverse and flexible computing resource. Whereas it will change the social and commercial landscape, performing many tasks presently handled by humans, it neither replaces humans nor substitutes human intelligence. What's significant is that enormous (versatile and adaptable) computing capabilities are now freely accessible to anyone with a computer and an internet connection.
The present opportunity is the possibility to harness a machine that augments a vast array of capabilities, and that has the flexibility of being many things to many people, of increasing our analytical and productive capabilities, of refocusing our limited resources –time, attention and energy–, and of tackling projects and opportunities that were previously out of reach.
#3 How it behaves
Based on applied experience of using Claude, ChatGPT, Gemini, Grok, and others, we can describe the ways in which an LLM functions, and attempt to provide one of the many relevant explanations of the extent to which its promise of intelligence affects the way we do things.
When using an LLM in the general course of research or business, with the intention of completing otherwise time-intensive tasks, the following 4 points summarise the overall experience:
- It replies to prompts in an expansive manner, providing a lot of information; it provides more information than we need, requiring us to identify what is relevant.
- When prompted to organise large amounts of information, the initial result is positive and surprising; as we try to refine what was organised, we learn that we might have achieved better results with clearer instructions to begin with.
- In any given interaction it repeats past replies and conclusions. While we attempt to focus and distill, it offers new variations and options. Instead of progressing from hypothesis to conclusion it explores, clarifies and concludes in a random and fragmented manner; there is no implicit order to the discussion.
- The longer a discussion, the more it digresses from the original context. Answers to prompts raise new questions; the answers to new questions change the order and relationship of the original arguments. Continuity is disrupted, and we must retrace earlier versions of the discussion to get clarity.
As we experiment with different types of prompts, we start to understand something about the quality of the interaction. Language is accurately rendered; grammar is impeccable, its logical capabilities are powerful, and it even applies rhetoric to persuade us of the worthiness of our questions and pursuits.
At the same time, interaction that is provided by mathematical pattern creation has a distinct pattern of its own. It is generative in a broad and prolific way. Its method is to produce responses that are rational and that are within proximal range of the prompt's instructions, and it does this both artfully and with brute force.
#4 What changes
Large language models (LLMs) occupy peak territory in terms of technological achievement and impact. Their ability to redefine the current state appears to be unparalleled. It is hard to think of any sector whose opportunities are not dramatic when viewed in the context of LLMs. We are also challenged to understand the scale and pace of current changes by looking to the past for examples.
The usefulness of railroads and telegraph systems required little explanation. 150 years later, the use cases for the internet also did not require much in the way of explanation, though they were certainly more abstract. LLMs are poised to infiltrate every commercial sector at the level of the individual worker, and by virtue of its implied versatility, its usefulness is difficult to describe and to define.
The biggest threat of a new technology is never to the individual. Though it may pose a threat to any pre-defined job position, potentially rendering it obsolete, it only does this in the context of the job's present definition. While some functions of that job will be better handled by an LLM, other functions will necessarily remain with people; and as with any new technology, it will also present its own unique emerging requirements, creating demand for new sets of skills and activities that require people to perform.
If the threat is not to people, then it is to the established norms and structures, to the way in which the world is presently organised. The internet reorganised the world, and LLMs now promise to do the same. What this is going to look like, nobody truly knows.
We are currently in the stage of chaotic growth, characterised by euphoric investments built on unsubstantiated claims. While we tend to be impressed by optimistic predictions and hyperbolic claims, it's important to think critically, and to view present developments from the wisdom of past experience.
#5 Individuals
When environments and markets change, the attributes that have the most value are curiosity and mobility. The old species and entities struggle, either due to blindness, the inertia of their sheer size or their inability to adapt. While a given person may lose the opportunity of their current job, LLM technology counteracts this threat through optionality – what is available to a large enterprise is also available to the individual:
- An analyst loses her job due to LLMs, but uses her knowledge and experience to create her own downstream opportunities through LLMs, providing big data analysis to family run businesses that can benefit from these services.
- A logistics manager loses his job due to LLMs, but uses his network and domain knowledge to develop his own resource planning platform using LLMs, offering his services to other firms.
- A CPA loses his job due to LLMs, but uses LLMs to streamline tax accounting processes in a way that he understands, enabling him to directly service private clients, thereby competing with the larger firm that laid him off.
Tools that were previously out of reach, or skills that were too time-intensive, now accrue to any one individual. Web development is technically challenging to most, but with an LLM anyone of average intelligence who has the will can develop a web presence, develop his own network, and build his own direct customer base.
The internet expanded the versatility and reach of businesses, reducing their reliance on physical locations and increasing their geographic reach, allowing smaller businesses to compete with larger businesses on a more even playing field. LLMs further reduce the barriers to entry, allowing businesses that are even smaller to participate and become viable contenders in large sophisticated markets.
#6 Society
The point of demystifying LLMs, of questioning claims to intelligence and of choosing to define its advantages in the context of any new technology (machine), is that it puts people and society first, and at the center of the frame, emphasizing the advantages that must necessarily accrue to them.
Doing this also puts into perspective the reality of social transformation, which is a gradual path, not an instant one. Every technology, no matter how groundbreaking, makes its presence felt gradually. Things are not going to change overnight.
While the euphoria of instant results may continue for some time, the first waves of failures already point to the gap between the signal and the reality: 'artificial intelligence' will not reliably or predictably replace people, nor does it belong to an exclusive entity that chooses who gets to do the work – which is determined by market forces alone.
Because LLMs are accessible to anyone and everyone, and because they will require extensive training – training LLMs to perform tasks, and developing people to work with and train them – we are in for a long and gradual transformation.
Furthermore, the low-hanging fruit, the useful capabilities already demonstrated and implemented, don't reliably demonstrate the longer-term trajectory. While the general tone of the media is to highlight its peak achievements, deployment only works in increments.
Each stage of progress will reveal weaknesses and present setbacks, and these will have to be vetted and evaluated by people. Such tasks will leverage LLMs to solve problems resulting from the implementation of LLMs, and the operators who will lead them will be people.
#7 Markets
Markets will introduce their own dynamics to the progress equation. While LLMs may create competitive advantages for any one company, the proliferation of their use will erode those advantages, as the overall capabilities of all market participants rise. Increased competition creates friction that counteracts the gains of the technology.
Structural changes will also cause friction, and the process of change will create challenges for all companies. Large-scale transformations can only be realised through progressive organisational changes. From the market perspective, large transformations can only occur through the collapse of old and the emergence of new business models and new market categories, which will require adaptation and reorientation by all market participants.
The best vantage point for understanding any new technology is to use it, and while doing so, to reflect on how people and society might adopt it; the best vantage point for an individual is to understand how it can improve his life by augmenting his unique approach to life and to work, within the environment he chooses to operate in.
In the current environment it's safe to say that haste is the enemy of critical thinking. In a gold rush it is best to think extensively and to act slowly. The opportunities that LLMs provide are real, but the path to realisation must be assessed, planned and executed, all the same.