AI - Part 1: The Good — AI's Genuine Promise
This is the first part of a 4 part post.
Well, AI has certainly arrived, hasn't it? Very few technologies promise to make as much as an impact, however just like the automated loom or the computer before it, AI is nothing more than a collection of tools, and just like any tool, its value depends on who wields it and to what end.
Before examining the harms that the current AI trajectory inflicts on workers and communities, it is worth acknowledging the genuine capabilities these technologies offer. Dismissing AI wholesale would be as naive as embracing it uncritically.
The question is not whether AI can contribute to human flourishing—it clearly can—but whether the benefits of that contribution will be shared broadly or captured narrowly.
Transforming Medical Diagnosis and Treatment
The most compelling evidence for AI's social value comes from medicine. A systematic review and meta-analysis published in Nature Digital Medicine examined 83 studies of generative AI diagnostic performance conducted between 2018 and 2024. The analysis found no significant performance difference between AI models and non-expert physicians, and while AI still lagged behind expert clinicians, the gap is narrowing rapidly1.
This is not a trivial finding: in regions where specialist physicians are scarce, an AI system performing at the level of a competent non-expert could meaningfully expand access to diagnostic care.
The real gains appear when AI is deployed as an assistant rather than a replacement. A multi-reader, multi-case study of 200 emergency doctors in Australian hospitals found that AI assistance increased correct diagnosis rates by 5.9% and improved patient management decisions by 3.2%, with the largest benefit accruing to senior residents, who saw an 11.8% improvement2. Critically, these improvements came without meaningful increases in interpretation time. This is augmentation in the truest sense: the technology makes existing workers more effective rather than rendering them redundant.
In drug discovery, the promise is even more dramatic. Traditional pharmaceutical development is notoriously slow, expensive, and failure-prone. AI-driven approaches now streamline target identification, lead optimization, and drug repurposing. Deep neural networks, convolutional neural networks, and generative adversarial networks are being used to design novel drug-like compounds with desired properties. Case studies include DDR1 kinase inhibitors designed through generative models and CDK20 inhibitors developed via structure-based methods—therapeutics that might have taken years longer to identify through conventional screening alone.3Models like Curate AI can optimize personalized treatment dosing in real time, predicting toxicological risks with high accuracy.
Climate Adaptation and Disaster Response
The same pattern-recognition capabilities that power medical diagnosis are being applied to climate extremes. AI systems now improve weather forecasting, model emulation, and the prediction of floods, droughts, bushfires, and heatwaves. A comprehensive review published in 2024 found that AI can help identify and explain extreme climate events more effectively, improving disaster response and communication. The technology is not a substitute for emissions reduction, but for communities already living with the consequences of a warming planet, better prediction means more time to evacuate, more targeted resource allocation, and ultimately fewer deaths.
These applications are not speculative. Machine learning models are being integrated into operational forecasting systems, and AI-driven analysis of satellite imagery is helping humanitarian organizations preposition supplies before disasters strike. The key challenge identified by researchers is not technical capability but trust: models must be transparent and interpretable to gain the confidence of the stakeholders and emergency managers who rely on them4.
The same logic applies in agriculture, where AI applications in precision farming, crop modelling, and early disease detection can improve resilience and food security, particularly for smallholders in climate-vulnerable regions.
Cybersecurity as Collective Defence
For working people, cybersecurity is not an abstract concern. Ransomware attacks shut down hospitals, disrupt supply chains, and compromise the personal data of millions. AI offers a meaningful advance in collective defence. Large language models and machine learning techniques can now detect malicious behaviour in real time, identify patterns of intrusion, and classify previously unknown threats—capabilities that traditional signature-based approaches cannot match. A 2024 survey of LLM applications in cyber threat detection catalogued approaches ranging from automated malware analysis to network intrusion detection, noting that these tools can process threat intelligence at a scale and speed that no human security team could achieve alone4.
This is not about replacing security analysts but about equipping them to defend networks against adversaries who are themselves using AI. In a context where cyberattacks increasingly target public infrastructure and small organisations with limited security budgets, AI-driven defence can help level a playing field that is currently tilted heavily toward attackers.
Freeing Workers from Drudgery
Perhaps the most widely cited promise of AI is liberation from repetitive, meaningless work. The argument has genuine merit. AI can automate data entry, routine report generation, scheduling, and countless other tasks that consume working hours without engaging human creativity or judgement. One Harvard Business Review analysis describes the fork in the road clearly: "Some organisations see it as a way to replace employees and to squeeze more productivity out of the ones they have... But others see GenAI as an opportunity to free workers of mindless tasks and rote drudgery, so people can engage with more complex problems".
However, the evidence complicates this sunny narrative. A 2025 study published in Scientific Reports across four experiments with 3,562 participants found that while human-AI collaboration enhanced immediate task performance, the performance gains did not persist when participants returned to independent work. More troublingly, the transition from AI collaboration to solo work produced significant decreases in intrinsic motivation and increases in feelings of boredom5. The researchers identified a "psychological deprivation effect": when the AI handles the engaging parts of a task, workers are left with the fragments, and their motivation deteriorates.
This finding is crucial. It suggests that the design of human-AI collaboration—who controls which parts of the work, and whether the human retains meaningful agency—determines whether AI augments or diminishes the experience of labour. The technology itself can serve either outcome; the determining factor is the structure of power within which it is deployed.
Educational Access and Personalised Learning
In education, AI-powered adaptive learning systems can tailor content to individual learners in real time, emulating the benefits of one-on-one tutoring. A systematic review of 142 peer-reviewed studies found that AI-driven tools significantly improve personalised learning, provide timely feedback, and enhance student engagement.
Intelligent tutoring systems can track student progress and dynamically adjust difficulty—a capability with particular relevance for regions where student-to-teacher ratios are high and individualised attention is scarce. AI-powered assessment tools also reduce teacher workload by using natural language processing to analyse written and spoken responses6.
Technology Is Not Destiny
Across medicine, climate science, cybersecurity, drug discovery, and education, a consistent pattern emerges. AI tools can genuinely augment human capability—improving accuracy, expanding access, and accelerating discovery. But in each domain, the benefit depends on design choices about how the technology is integrated, who controls it, and whether productivity gains are reinvested in workers or extracted from them. The same generative model that helps a junior doctor in a rural clinic spot a fracture could, under different economic logic, be used to justify replacing that doctor with a cheaper telehealth alternative.
The positive applications are real. What remains unsettled is whether they will be developed as public goods, accessible to the many, or as proprietary assets whose benefits flow upward. That question turns not on the technology itself but on the social arrangements governing its deployment—arrangements that the following sections will examine in detail.
This is the first post in a 4 part series.
Next part is here.
