From neural to inclusive AI: closing the gap in human-centred AI adoption

The Push for Inclusive AI

Artificial intelligence has moved beyond experimentation to widespread deployment, with neural networks powering decisions in employment, finance, healthcare, and public services. Yet the focus on raw performance—often termed 'neural' in reference to brain-inspired deep learning—has created a widening gap: systems excel in controlled settings but falter when applied to diverse populations, perpetuating biases and excluding underrepresented groups.

Recent global developments have sharpened the urgency. In early 2026, events like the India AI Impact Summit produced the Delhi Declaration, endorsed by dozens of countries, endorsing inclusive, human-centric AI that prioritizes equitable access and development. At Davos 2026, discussions highlighted how trust, fairness, and broad skills determine the next adoption wave. Meanwhile, reports show AI usage rates soaring in places like the UAE and Singapore but lagging elsewhere, creating a digital divide that risks entrenching global inequalities.

The EU AI Act represents the most concrete pressure point. After entering force in 2024, its prohibitions on unacceptable-risk practices took effect in 2025, with high-risk system rules—covering applications in hiring, credit scoring, and medical diagnostics—set to apply fully from August 2026. Providers and deployers must implement risk management, high-quality diverse datasets, technical documentation, and ongoing monitoring for discrimination. Non-compliance carries steep fines, and the extraterritorial reach means global firms serving EU markets must adapt.

Real-world impacts are already visible. Biased AI in recruitment or lending disproportionately affects women, ethnic minorities, and people in low-resource regions. In healthcare, inaccessible AI tools widen gaps in low- and middle-income countries. Workforce anxiety over job relevance and identity further stalls integration, as employees resist tools that feel misaligned with human values.

Non-obvious tensions include the trade-off between speed and inclusion: rapid scaling of frontier models often relies on narrow datasets, while true inclusivity demands costly diverse data and multilingual capabilities. Open-source efforts like DeepSeek have boosted access in underserved markets, yet raise concerns about unchecked proliferation. Balancing innovation with ethical guardrails remains contentious, as over-regulation could hinder progress while inaction amplifies harms.