36 Book Reviews
Unlock your personal power
This book by executive coach and Dragon’ s Den pitcher for healthy snacks Claire Brumby is based around five premises with which it is hard to argue.
Firstly, the contention that leadership has fundamentally changed. Employee disengagement is at historic highs while traditional‘ command and control’ leadership no longer works. Secondly, empathy and authenticity now drive performance meaning that modern leaders are expected to display trust, emotional intelligence and selfawareness.
Non-linear
Women now comprise nearly half the workforce and careers are increasingly non-linear. Leadership guidance needs to reflect lived experience, not outdated norms. And finally the incidence of burnout has made intuition a leadership necessity. Leaders need clarity and self-trust to make confident decisions and lead sustainably.
Human approach
This is a bold and practical guide to modern leadership. Brumby has created a
MAGIC framework – the five pillars of Mindset, Awakening, Gumption, Intuition and Charisma.
The book is described as“ ideal for women in business, professionals moving roles, or anyone seeking a more intuitive human approach to leadership.” Ultimately, how to unlock your personal power!
Forget Normal, I Want MAGIC Author: Claire Brumby Kogan Page www. koganpage. com
A Smarter City
Why AI needs to explain itself
Vallikat Peethamber on causal AI and the case for trustworthy automation
Artificial intelligence has moved from the experimental edge of business into its operational core. Banks assess credit risk with it. Manufacturers predict equipment failure. Healthcare providers triage patients. Yet as AI takes on more consequential decisions, a problem once academic has become commercial: most models cannot explain why they reached a conclusion, only that they reached one.
A model that denies a loan or flags a transaction as fraudulent is making a decision with real consequences. When a regulator or board member asks " why," the honest answer from most current systems is that the model found a statistical pattern, not a reason. That gap is becoming a board-level issue across financial services, healthcare, and manufacturing alike.
What causal AI actually changes
Most machine learning is built on correlation: a model notices that certain inputs tend to appear alongside certain outcomes and repeats that association at scale. It cannot reliably answer " what if we had intervened differently " – only " what has tended to co-occur before."
Causal AI represents the structure of cause and effect, not just association. Decisions become explainable in terms a human can audit. Not " the model scored this 0.83," but " this outcome changed because of this specific upstream factor." Causal models are also more robust when conditions shift, because they are not simply re-describing historical patterns that may no longer hold
From theory to practice
The deeper question is not architectural but mathematical: is there a form of inference grounded enough to require no free parameters at all? Most AI systems inherit their opacity from degrees of freedom introduced during training – weights, hyperparameters, regularisation choices. Each is a place where the model silently encodes an assumption the designer cannot easily retrieve later.
Harmonic analysis and kernel theory point toward a resolution. A kernel fully determines an inference method if and only if it is characteristic – uniquely identifying probability distributions in the space it operates over. There exists a unique extension of this principle to the adelic number system, the mathematical structure running simultaneously over the real line and every prime-based number system at once. That object has no free parameters by construction: it is the only function satisfying a self-symmetry condition across all scales simultaneously. The regularisation parameter – the last
remaining degree of freedom in kernel methods – can be eliminated analytically, yielding a complete inference pipeline where every output follows from mathematical structure rather than training history.
This is the basis of Canon +. An inference engine with no free parameters produces the same output for the same input regardless of when it ran or who configured it. Auditability follows from reproducibility. Explainability follows because the reasoning structure is the mathematics – not a post-hoc approximation of it.
What this means for London businesses
The Bank of England, the FCA, and EU regulators are all moving toward explicit expectations of explainability and auditability for AI-driven decisions. Firms that adopt causal, explainable approaches now will find compliance far less disruptive than those retrofitting explainability onto opaque systems after the fact.
The broader lesson is simple: ask not only how accurate a model is, but whether it can explain itself. That question is becoming the dividing line between AI an organisation can trust with significant decisions and AI that requires a human to quietly double-check everything regardless.
Vallikat Peethamber is co-founder of VectorPeak
www. vectorpeaktech. com