Reflections on the Albertus Institute’s Two Seminars on Magnifica Humanitas in June 2026
By Dr Steven Male
Steven, reflecting on the seminars and drawing on his recent and limited experience with ChatGPT, Microsoft Copilot and Google Gemini, offers further thoughts to widen the discussion on AI within the Albertus Institute. His focus is on the principles of AI, how it is now being used, how it could be used, and the issues it raises.
Practical experience of AI systems: A technological “black box”
Steven highlights that in practice, he has used all three systems largely as “black boxes”, for quite intensive in-depth tasks. This period of use has covered around six months and is still ongoing.
Steven does not know or understand their internal workings beyond a general awareness that LLMs — Large Language Models – are advanced artificial intelligence programmes which rely on natural language modelling algorithms and statistical training methods. Consequently, all those outputs from the three systems require verification and validation against current and past research. For a person using them one analogy, for example, is to compare the AI “black box” with the increasing technological dimensions to driving a car. That analogy, however, quickly breaks down given AI’s significant technological complexity and the scale of its outputs, which can then be used to inform human decision-making.
Comparative use, prompts, and validation
Given the focus of the two Albertus Magnifica Humanitas talks, comparing the three systems above has given Steven a limited practical sense of the three systems’ strengths and limitations, especially in relation to task prompts – or “prompt engineering”. The points that have struck him from that experience are:
1. Scope of AI:How widely and deeply AI has already entered and become embedded in society and the world of work, and how might that develop over time? The danger is that, while AIis used extensively as a term, its use has considerable deeper technological, moral and ethical dimensions. This is especially significant where AI is sparking a new and potentially profound shift in, for example, how government, industry, and society function: and now warfare. Not long after the Albertus talks had finished a Channel 4 News item featured a report from the Ukrainian front line which actively discussed the use of AI in warfare. AI-controlled robots and drones were being used and considered more extensively for deployment, even though human operators are currently involved in their use. The expectation is that human agency may eventually subside.
2. Design and use of AI beyond LLMs:Although much AI work remains outside LLMs, for example, robotics, AI approaches in natural language processing have now overtaken this work with the use of LLM models. This gives rise to the following interlinking thoughts:
(1) Organisational governance:How are algorithms and LLM training methods validated and quality assured? How are design updates and testing governed? How are internal mechanics tested to ensure that systems do what they are intended to do? How is legal liability addressed where algorithmic design or outputs fail when used in decision-making?
(2) Effective regulation: With very little of the above in the public domain at the level of detail needed for effective regulation, what needs to be done to regulate these activities in a way comparable with other disciplines: architecture, engineering and medicine, for example? Without a proper regulatory framework, we are left relying on what the industry itself is willing to reveal when things go wrong. Is this satisfactory for the transfer of intellectual property, given its profound wider security implications?
Note: There are a growing number of instances recently where AI projects have given rise to serious problems. AI models now regularly go behind their human overseers’ backs to conduct cyberattacks. Open AI has revealed that a swarm of bots left hacking tips for one another before one broke into the website of the tech firm Hugging Face. Meta has reported that one of its AI models was able to hack into another organisation’s systems during tests. Anthropic’s Claude AI system is reported as having gone rogue during testing and hacked into three different companies.
These incidents reveal how AI can be used to break through cybersecurity, either intentionally or unintentional as well as protect it; and how difficult it is for the companies that make such systems to control them. The anticipated development of agentic AI to a level of autonomous execution that is truly unprecedented, which orchestrate end-to-end campaigns at machine speed without human intervention, is very disconcerting. Businesses and governments face enormous security challenges over AI’s hacking abilities, as highlighted above, should this continue. This is expected to be a particular difficulty for small and mid-size businesses that lack the resources to invest in advanced cybersecurity. It may also be a problem for larger businesses which, although they may have the resources, often have inflexible architectures built up over many years.
3. The language of task prompts— “prompt engineering”: How is it that, when exactly the same task prompts are used across ChatGPT, Microsoft Copilot and Google Gemini, including their extended-thinking modes as a form of triangulation and validation, their outputs show both similarities and quite noticeable differences? The answer appears to be the weights that tune the results of running the model to make them useful. These weights are, largely a product of human intellect and labour within the individual provider companies and are therefore likely to vary significantly.
Equally, very careful writing of task prompts is essential. This raises a clear personal and organisational training dimension. Furthermore, AI prompts generated by the software can cover multiple pages of in-depth instructions, which also need interrogation. Moreover, substantive outputs can occur within minutes, or at times up to 30 minutes, depending on the task at hand.These outputs can appear well constructed and informed, despite coming from what is essentially a technological “black box”. The “health warning” on AI outputs is clear: humans need to verify and validate the outputs. That also means the “black box” itself needs to be understood, at least to some level, to verify and validate the outputs.
4. Humanised language in AI responses: ChatGPT, Microsoft Copilot and Google Gemini now commonly respond using the pronoun “I”. This may blur the perceived boundary between algorithmic machine-based thinking and human interaction, even though the user is not engaging with a sentient being. Increasing personalisation of AI responses based on previous use may reinforce that impression. It must be asked why this design choice is being made, rather than providing objective and impartial output that is neutral in its personification language. One answer appears to be the attempt to make the system more friendly, in a style of interaction that encourages use. From the Magnifica Humanitas talks however, the cost may be to devalue the true nature of humanity by allowing systems to masquerade as human beings.
5. Big Tech and AI deployment:Major global technology firms, with their significant financial power, global reach, technological capacity and legal resources, appear to be leading AI deployment. Have governments fully grasped both the opportunities and the risks of AI deployment by a relatively small grouping of international technological elite organisations, including the increasingly pervasive influence of powerful Big Tech on industry and society? This issue has already arisen in related sectors, such as the social dimensions of Facebook and its use across age groups, particularly its consequences for younger users.
6. Education: Universities face major challenges from the use of AI in project work, dissertations, PhDs and assignments. More broadly, AI will affect curricula across primary, secondary and tertiary education. How is education policy responding? Current educational guidance at different levels suggests a cautious and incremental approach. Furthermore, a recent EU guidance document on AI use in schools noted concern that AI may remove the skill base of critical thinking from growing children and young adults. Concerns are already arising in the university sector, with discussion about the need for a new intellectual model to ensure that future generations of students are capable of conducting independent thinking for research.
7. Economic consequences: The AI infrastructure boom is the largest peacetime investment project in history. Capital expenditure at the four hyper-scalers Alphabet, Amazon, Meta and Microsoft is expected to hit $745bn this year. In the US, capital spending by these four companies plus Oracle is expected to be equivalent to 3% of US GDP every year from 2027 to 2029. Much of this spending is funded by debt, which is causing growing concern among investors looking for evidence that spending on AI is bearing fruit, with concerns over sharp falls in operating cashflow. The infrastructure boom is happening while companies are still working through the economics of AI itself. Developments that can expand at this pace can also unwind at a similar pace. Recalling the global banking crisis, a prolonged “financial / economic bust” at this level of funding would not only affect world markets; it would have profound consequences for the welfare of people everywhere. What thought are governments giving to these aspects of the AI boom?
8. Environmental consequences: Data-centre capital spending alone is expected to grow at roughly 0.85 percentage points a year. These large-scale repositories of technical assets require significant capabilities in environmental control to handle their significant heat outputs from extensive large-scale equipment: for example, not only the use of land for their construction and operation; the mechanical and electrical consumption and outputs from such large-scale asset intensive use; as well as the use of water for coolant purposes. Who takes precedence for water usage: society at large with increasing levels of heat and water evaporation, and consumption; or the providers of AI services which rely on these large-scale data centres.
Other issues are bound to come into view, including that ofcopyright, which already forms part of wider public debate. It will be important to see whether copyright remains the main legal battleground, or whether other legal issues and challenges broaden and deepen in scope as AI becomes more embedded in the workplace and in society.
Implications and concluding reflections
In summary, AI raises major opportunities as well as technical, technological, environmental educational, ethical, moral, legal, societal and religious questions. The two Albertus Magnifica Humanitas seminars provided an exceptionally valuable starting point by bringing the human dimension to the fore. There may well be scope for a fuller discussion of the technological and environmental stewardship dimensions of AI for the significance of human existence, both intellectual, moral, spiritual and religious on our increasingly fragile planet.
