IEEE Spectrum published its AI at 70 commemorative issue on 22 June 2026, marking the seventy-year anniversary since the Dartmouth Summer Research Project of 1956—the gathering where the term "artificial intelligence" was first coined by John McCarthy and colleagues. The timing coincides with Stanford's Human-Centered AI Institute releasing the 2026 AI Index Report, which provides current data on where the field actually stands in terms of adoption and organisational readiness.

The Papers That Shaped Modern AI

Of the ten papers IEEE Spectrum selected as most influential since 2000, the most structurally important is almost certainly Ashish Vaswani and colleagues' "Attention Is All You Need" (arXiv:1706.03762), presented at NeurIPS 2017. The transformer architecture it introduced—a method for training machine learning systems that can process sequences of data efficiently—became the foundation for virtually every large language model in commercial deployment today. A paper published in summer 2017 restructured an entire industry within five years, which is genuinely unusual velocity even by modern machine learning standards.

The broader selection spans 2000 to present and charts the field's evolution from statistical methods through deep neural networks to current-generation foundation models—the large AI systems trained on billions of examples that can perform many different tasks. For practitioners working in AI today, the IEEE list functions as a concise reading list: these are the papers that explain why contemporary systems behave the way they do.

The historical anchor matters as well. Nils Nilsson's definitive history of AI bridges the early Dartmouth work—which emphasised symbolic logic and formal rules—with the connectionist approach (systems loosely modelled on biological neural networks) that produced today's papers. McCarthy's founding contribution was not simply the name itself, but the claim that intelligence could be formalised mathematically. The field spent six decades testing that idea before data volumes finally allowed it to pay off.

Where AI Is Actually Being Adopted

The 2026 AI Index Report presents adoption figures by country with some striking disparities. Singapore leads at 61%, the UAE at 54%, and the United States ranks 24th globally at 28.3%. For technology leaders inside US enterprises, that ranking warrants attention. The country that produces the most AI research and hosts most frontier model developers is not among the fastest adopters of the technology those developers create.

The report also documents what it calls a widening gap between AI capability—what the systems can actually do—and organisational readiness to govern them responsibly. This appears consistently in enterprise practice: companies have accelerated purchasing AI tools, but have not built the internal structures for model auditing, data governance, and clear liability assignment at a matching pace.

The Singapore and UAE results are telling. Both are relatively small, administratively centralised economies with national AI strategies and the government capacity to execute them quickly. The US figure, by contrast, reflects the friction of a large, decentralised economy where adoption decisions are made firm by firm across millions of organisations with vastly different risk tolerances and technical maturity. The gap between what AI systems can do and how quickly organisations adopt them is not simply ambition. It is partly a structural outcome of scale and diversity.

It is worth noting that the AI Index adoption figures are self-reported and aggregated at the national level, which obscures real variation within countries. A single 28.3% US figure that averages hyperscaler-adjacent technology firms with mid-market manufacturers and rural healthcare providers tells you something, but not everything.

What Comes Next

The IEEE commemoration and Stanford's data converge at a moment the Dartmouth founders could not have foreseen: AI systems are capable enough for large-scale deployment, yet the institutions, regulations, and competencies required to govern that deployment remain incomplete. McCarthy's 1956 question was whether machines could simulate intelligence. The 2026 operational question is different: who is deploying these systems, how prepared are they to do so responsibly, and what does the geography of adoption reveal about where benefits and risks will concentrate.

Seventy years from a New Hampshire campus to a Stanford readiness report. The field has delivered on its technical promise far more substantially than early sceptics predicted. The next decade's challenge will be less about building more capable AI and more about closing the distance between what these systems can do and what the organisations running them are equipped to manage.