AI and the Collapse of Trust

Beneath the veneer of the twin transitions-decarbonisation and digitalisation-lie two major challenges. The first involves rebuilding trust within and between nations. Trust is collapsing everywhere, including inside global institutions. Humanity seems to agree on only one thing: we no longer trust each other. Trust is fundamental to human dignity and our shared future. Without it, progress and peace fade.

The second challenge is the rapid progression towards Superintelligent AI. Such AI poses a fundamentally unique danger-indeed, it is the primary risk. If AI surpasses human intelligence, it could bypass the guardrails established by its creators.

The risks posed by Superintelligent AI fall into three broad categories: misuse, where AI is operated as intended but used maliciously; a disconnect between programmers’ intentions and AI goals; and the adverse behaviours emerging from AI’s methods of achieving its objectives.

Paradoxically, as we make progress towards Superintelligent AI, trust in algorithms grows while trust in humans diminishes. In 2025, the term “arms race” primarily refers to the escalating competition in AI and space-based technologies. This includes AI-powered weaponry, surveillance systems, and cybersecurity.

Countries are heavily investing in research and development, aiming to gain a strategic advantage through AI. Some are focused on deregulation to speed up AI development, while others are rapidly advancing AI infrastructure.

During this arms race, AIs are observing human decisions and actions, learning from human behaviours, and the bricolage of our digital artefacts. This poses a deep challenge for the architects of AI-Assemblages to build compassionate and trustworthy AI in such a competitive environment.

It is already the case that AIs observe and learn behaviours from their creators-behaviours that have their resolution in the misery of famine, starvation, drone strikes, extrajudicial killings, modern-day slavery in data-labelling factories, rocket attacks, assassinations, abductions, and a blatant disregard for human dignity and human rights.

From personal experience, we know that children do not do what we tell them to do. They do what they observe us doing. They model the behaviour they see around them. For example, a young boy seeing his father shave with a shaving set and cream may go to the bathroom to imitate him, much to everyone’s horror.

Albert Bandura’s social learning theory states that learning occurs through observing others. This theory underscores that we acquire new behaviours, skills, or attitudes by watching and imitating others. While we might train and instruct AI not to harm humans, it is observing human cruelty and unimaginable violence.

So, how can we expect to give AI a moral compass we ourselves lack? None can give what he does not have, “Nemo dat quod non habet.” We cannot give a morality to our children or artefacts if we ourselves lack one. As we witness the rise of artificial agency, we become inheritors of a world we will co-create with AI.

Unmatched progress in AI is driven by the vast positive potential to resolve complex issues surrounding disease, poverty, the climate emergency, intergenerational immobility, inherited inequality, and environmental degradation.

However, as we edge closer to superintelligence, AI is evolving to become increasingly autonomous, as it becomes less and less artificial and less and less an artefact. Artefacts are created and controlled by humans, but AI is designed to surpass us and generate solutions beyond our imagination and anticipation.

True intelligence is not an artefact, as it can learn by creating new ideas through a bricolage of existing knowledge and adapt independently. The term bricolage refers to the process of creating something entirely novel by combining diverse, accessible resources and ideas to make something new.

It involves making do with what is available, often leading to unexpected and innovative outcomes. Originating from anthropology, this concept now describes how AIs innovate by creatively repurposing existing pieces of information from a web of hyperlinks to make something unimaginable, unanticipated and unforeseen.

If AI merely follows instructions, then it cannot be considered an authentic intelligence. In many cases, predicting outcomes is impossible when using prompt tools like Atom of Thought (AoT) or techniques like Chain of Thought (CoT). Randomness and unpredictability are central to the alignment challenge: how do we ensure AI is both broad in its embrace of human value systems and benevolent?

Prioritising human values is tough for AI because values are often contradictory, context-dependent, shaped by culture, and habitus-factors that are challenging to quantify and encode into algorithms.

Ensuring that AIs align with human values is not just a technical task but a moral obligation involving trust. While humans steer these subtleties instinctively, AIs process patterns without fully appraising the varieties of human value systems.

This hampers their ability to make decisions aligned with our fluid and multifaceted moral landscape. Misaligned Superintelligence can produce unintended and potentially disastrous outcomes if trust is absent.