Artificial Intelligence & Arrogance and Ignorance
So much has been written on AI and a board’s many roles relating to it. I’ve held back as there are so many people far more qualified than me doing such a good job. Yet I have become increasingly exercised by something which I feel is being underacknowledged when considering AI investments and taking decisions augmented by AI. It is also something which Artificial Intelligence, and in particular the application of Machine learning and Large Language Models, could help us more with.
To put it bluntly, it’s about another “AI” - Arrogance and Ignorance. These frequent and fiendish travelling companions are slayers of judgment and destroyers of value and impact. Heavy doses of critical thinking are required to keep them at bay as well as to minimise their destructive capabilities as we work towards achieving the awe-inspiring potential of artificial intelligence.
In an increasingly algorithmic-driven driven world, those who write the code, whether human or artificial, have considerable power. Their ability to system think, to have the right objectives, to ensure the integrity of their models and the robustness of the assumptions driving them, determines outcomes. As will the levels of arrogance and ignorance, and quality of critical thinking amongst everyone in the chain, whether they be creators, investors, users or those who are responsible for the governance of organisations.
The real rules of this emerging AI game are being made up as we go along, in an increasingly rule-less, winner takes all environment. Shortening attention spans combined with a lack of desire or patience to understand underpinning logic or underlying assumptions adds further risk. Especially when biases lurk well beneath, rather than near or on the surface.
We hear the cry so often that the development of AI is just like the development of one of its key enablers, the internet. And indeed, we are seeing some similar patterns. Goldrush and FOMO (Fear of Missing Out) investing, the appearance of snake oil salesman all around us and, as with many social and culture-shifting technological developments, huge demand for unpredicted use cases.
Yet it is also profoundly different. In part because of the environment and atmosphere between now and 2- 3 decades ago. In part because of the speed of its adoption but mostly because of the fear of its potential for us to lose control. For some, high levels of arrogance and ignorance may create an almost hallucinatory illusion that they will gain greater control through something they don’t understand.
Making investments in AI is also something of a developing art and science. With so many fundamental variables shifting so significantly and unpredictably, our trusty and familiar Return On Investment (ROI) approaches with more easily identified risks and more predictable payoffs are hard to apply. It’s also hard and perhaps a little early to precisely join the dots. Yet we need to.
The recently issued SpaceX prospectus, with its sci-fi narrative and fantastical fusion of facts and interpretations being the most spectacular example of an investment case so far. A summary of its 377 pages is essentially that “This is a huge punt but in Elon we trust and even if all our investment shrinks to dust, then we’re happy for him to decide whether he stays”. It worked and the IPO was oversubscribed.
It made Nvidia’s presentation ahead of its June 2026 AGM, with estimates that spending on AI infrastructure will reach $3-4 trillion dollars a year by 2030, look sober by comparison with only a 40% compound annual growth rate over 5 years. NVIDIA justify their estimates more conventionally, citing examples of the scaling of different use cases. Yet when you stand back, the major factor that will sustain such investment will be whether it delivers commensurate returns for those making that investment and ultimately on those being charged for using it. On that, it is silent.
The world’s five largest technology firms apparently spent more than $400 billion building data centres in 2025, and they are forecast to spend c $750 billion in 2026. Given the risks, you would expect them to want to get at least 3-5 times their money back over a five-year period. That’s $2.25- 7.75 trillion dollars just on that year’s investment. It’s interesting to note that much of this investment is not from their own balance sheets.
According to the OECD in 2025, venture capital (VC) investments in AI firms globally made up 61% of all VC investment (USD 427.1 billion), double its 2022 share. As with the dotcom investing frenzy, not all of this hastily spent cash will be well spent. The overconfident and uninformed investors will become apparent sooner than the smart ones, and the deflated froth will look a mess.
More positively and practically when it comes to being specific and embedding AI operationally in small and medium businesses Cassie Petrie, Managing Director SMB gave me a good example of how SAP Concur is embedding AI.
"For us, AI rubs the rough edges from travel and expenses. Our Expense Automation Agent automatically creates and populates expense reports for employees so all they have to do is review, refine and submit. It transforms expenses from a reactive chore, ensuring real-time policy compliance, automating repetitive tasks to eliminate manual errors and freeing staff to focus their time on other areas.”
Finance and travel teams also benefit from more timely, accurate expense data that supports stronger compliance, reporting, and visibility into travel spend. All these efficiency gains help lift some of the administrative burden and address pain points for businesses and employees alike.”
At school our rather eccentric Irish maths teacher Paddy B, drilled the power of critical thinking into all of us. His starting point was not trusting anything coming out of the mouth of a chemistry or physics teacher. He taught us not to take anything at face value and always seek out alternative hypotheses. The principles of linear programming, were of great joy to him. Having multiple objectives, ensuring that all of the variables and constraints are identified and putting the effort into understanding how they interrelate and interact seemed so logical. Then, most importantly, remember the aim is to come up with a maximising solution but understanding and to test the underpinning assumptions as well as the consequences and potential downside of every decision. Simulations are a great way of doing this.
As for Bayesian statistics, so central to the maths driving decisions with uncertainty and using live data to improve predictability. Well, that was something he would only share glimpses of as a treat. Even if, as with so many things he taught us, it wasn’t on the curriculum. He was a genius who captivated, inspired and terrified us, and we loved him madly. He raised our ambitions and mindsets. He loved fooling us by giving us problems that he said he couldn’t solve. Motivating us by saying that if we could solve them then he’d help us escape from “playing” chemistry and perhaps even save us being electrocuted by the rather dodgy circuit boards we had in Physics.
I imagine if he were around today, he’d be asking us to “work out if all this fe***** investment makes any sense lads”. He’d probably start by asking us to remember Enrico Fermi, a physicist he did have respect for. Fermi, the famous estimator, would use every day relatable things to try and make a best guess. For example, if you want to calculate the height of a tall building, look at how many floors it has. Then work out how many people standing on each-other’s shoulders would be a height equivalent to a floor, and you can then calculate a reasonable guess. Lest you think that’s a bit basic, Fermi also created the world’s first artificial nuclear reactor.
Today, we crave increased predictability of outcomes and speed of learning in relation to AI investment. We need the answer to “What is the current and predicted return on AI investment in X?”. Yet answers are understandably patchy, and even glimpses of them are based upon incomplete and inconsistent data.
Perhaps a Fermi approach powered by machine learning, combined with large language models fed with investment papers and decision review analysis for AI investment, will eventually provide it.
Until then, our collective investment will be sub-optimal and we will miss many “failure signals” in the investment papers for our strategic choices. Better we recognise that. Do all we can to increase our critical thinking capabilities. Reduce our ignorance. Be incurably curious to learn more and avoid arrogance the judgement slayer all with the aim of maximising the potential of the awe inspiring developments in Artificial Intelligence.
Patrick Dunne OBE
Experienced Chair and author of the award-winning book Boards
This article has been sponsored by SAP Concur industry-leading technology that automates travel, expense and invoice spend processes. The fee will be donated to EY Foundation an independent charity which supports young people from low-income backgrounds to maximise their potential through getting paid work experience, employability skills training and career guidance.









