Your AI is a high speed intern, not a partner. The IQ data explains why.
Updated: Aug 21
In educational and developmental psychology, we are trained to look past the surface of "IQ." When we administer the Wechsler Adult Intelligence Scale, currently in its fifth edition for Australia and New Zealand, we are not chasing a single score. It tells us very little, really. What we are looking for is much deeper than that. The current standard rests on decades of theoretical work that lets us identify cognitive strengths and weaknesses, the actual parts of the mind that combine to produce what the public calls an "IQ" score.
Understanding how those domains work, and what they mean, is more important than ever.
The world is turning to generative AI to solve complex problems, and the psychology of intelligence can explain, with precision, why that turn is going to disappoint a lot of people who do not yet know it.
The Full Scale IQ, or FSIQ, is what most people picture when they hear the word intelligence. It is a single composite, drawn from five distinct domains of cognition that the field has spent decades isolating. Verbal comprehension, our capacity to think with language and concepts. Visual spatial reasoning, the part of the mind an architect uses to see a building before it exists. Fluid reasoning, the ability to detect patterns and solve genuinely novel problems, the domain most people mean when they say insight. Working memory, the capacity to hold and manipulate information in real time. And processing speed, the rate at which the mind executes simple cognitive tasks under pressure. Average those five and you get the number the public remembers. It is a useful single figure. It is also a misleading one, because the five domains it covers tell very different stories about how a mind actually works.
This is why psychologist's look past the FSIQ, the "IQ" to a quieter composite called the General Ability Index, or GAI. The GAI strips out speed and short term memory load and asks the question that actually assists with predicting good professional judgement. What is this person capable of when the stopwatch is removed?
It is exactly the question we should be asking about AI, and at the moment, we are not.
Right now, we are clocking it. Benchmarking it. Measuring tokens per second and time to first answer and how quickly it returns something passable. We are, in clinical terms, fixating on its processing speed and working memory and calling the result intelligence. We are treating AI as if a high FSIQ is the prize. For the kind of thinking that matters in legal, medical, financial and executive work, it is not. What we actually need from AI, and from the humans who use it, is something closer to a high GAI. Depth of reasoning. Conceptual fluency. The capacity to hold a problem still long enough to interrogate it.
This is the cognitive profile of an intern, not a partner.
A bright intern almost always has strong processing speed and reasonable working memory. That is precisely why we give them the research, the first drafts, the turnaround work. What an intern has not yet developed is the verbal and fluid reasoning that come from years of pattern exposure, ethical seasoning and conceptual depth. We do not promote them to partner because we already understand that speed is not judgement. And yet that is exactly what is happening, quietly, in organisations across the world. The AI draft arrives so fast and so well formed that it suppresses the productive friction that used to surface a better answer. The room agrees more, and the room thinks less.
Albert Einstein is a great illustration of why this matters. By many accounts his processing speed was likely unremarkable. He was not necessary the fastest person in any room he sat in. What he had was the capacity to stay with a problem for years, reasoning verbally, spatially and fluidly about ideas no one else could hold steady. That is GAI. It is also the reason people still talk about him all this time later.
When leaders judge their teams, or their AI tools, on speed to response alone, they are optimising for the lowest order of cognitive functioning. They are rewarding the domain that least predicts good decisions in complex, ambiguous, ethically loaded environments, which is to say most of the decisions that actually matter.
If you would not promote your fastest intern to managing partner, you should not be outsourcing partner level decisions to your fastest tool either.
The implications of getting this wrong are not just a compliance problem. They are a reputational one, and they are a current concern, not a future one. Every regulated profession, including law, medicine, financial services, accounting, engineering and education, operates under a duty of care that assumes a competent human is exercising independent judgement at every material decision point. AI does not change that duty. It changes how easy it is to quietly fail it.
A solicitor who signs off on AI-generated advice without genuinely interrogating it has not fulfilled their professional obligation. A clinician who follows an AI-generated triage recommendation without applying their own diagnostic reasoning has not met the standard of care. A board that approves a strategy drafted by AI without testing the assumptions has not discharged its fiduciary duty. And in every one of those cases, the public failure that follows lives in the press and the public record long after the regulator has finished with it.
The risk is arguably greater in the industries that have no regulator at all. Marketing, communications, technology, consulting, recruitment, real estate, construction, hospitality and the broader corporate sector employ millions of Australians whose decisions affect customers, clients, employees and shareholders every day, but who answer to no professional body and no compulsory standard of care. There is no regulator to write case law for them. There is no tribunal that will examine the quality of their reasoning. The only safeguard between an AI-generated decision and the public is the professional integrity of the individual making it. When that integrity erodes, and it does erode in the presence of fast and authoritative AI output, the failure does not get caught upstream by a complaints process. It lands directly on the customer, the contract, the campaign, the brand. And it lands in the press all the same.
There is no instrument that brings clinical psychology and management theory together to measure how leaders actually work with AI. The closest available tools sit in two separate literatures. One assesses general digital literacy or technical fluency. The other assesses leadership style or emotional intelligence in the abstract. Neither captures the question I kept returning to in my doctoral research and consulting work, which is what happens to a professional's ethical agency when the machine starts producing answers faster than the human can interrogate them.
That is not a productivity question. It is an integrity question. The fields most exposed to AI right now, including law, medicine, finance and education, are also the fields where professional ethics is the entire reason the work is trusted.
Lose the ethical centre of gravity inside those professionals and the technology problem becomes a public trust problem. So I built the Artificial Intelligence Management Positioning Inventory, or AIMPI, drawing on personality psychology, ethical decision science and the literature on professional autonomy, to give that moment of integrity a structure leaders can actually examine.
The AIMPI assesses four domains:
Ethical Agency, whether you can steer the machine morally as well as technically.
Systemic Fluency, whether you understand how AI fits into the architecture of your work, your team and your sector.
Augmented Command, your willingness to push back on the output, question it and refine it.
Empathic Synchrony, whether you can hold on to your own attuned, curious thinking in the face of rapid fire answers designed to feel authoritative.
The regulators have not yet written the case law on this. They will. The journalists already have. When the first major reputational event lands, and it will, the organisations that survive scrutiny will be the ones that can demonstrate, in writing, that their people were trained to maintain their own reasoning in the presence of AI. The ones that cannot will find that "the AI suggested it" is not a defence any tribunal will accept, and it is certainly not a position any reputation can recover from quickly.
True intelligence is not how fast you answer. It is the depth of the problem you are willing to sit with.
If we want to lead well in the age of AI, we need to stop confusing speed with substance. Step back into the room. Take as long as the problem deserves. And reclaim the part of the mind that machines cannot rush.
If this has named something you have been quietly observing in your own work, or in the teams you lead, the next step is to look at it directly. The AIMPI is available now. It will show you where you sit across the four domains and what the data is telling you about how you are working with AI in practice.
