From groundbreaking innovations to bold visions, our 2025 Fellows share their predictions on where technology is headed—and the impact it could have on the world.
Advanced generative AI has seen rapid improvements in the last few years. But despite predictions that AI will revolutionize the economy in short order, the adoption of AI will be far slower.
While specific sectors will face rapid changes, the broader economic impact will be tempered by what political scientist Jeffrey Ding calls the innovation fallacy in artificial intelligence deployment: the impact of general-purpose technologies like AI on the economy is realized when AI is adopted across sectors of the economy such as healthcare, education, and finance rather than through innovations in the technology alone.
We have seen this with past general-purpose technologies like the internet, which took decades to be adopted across the economy despite rapid initial innovation. We are already seeing the same for AI adoption: while generative AI capabilities are advancing rapidly, organizations still struggle to effectively integrate these tools into their workflows and processes. This is both due to limitations of the technology (such as the lack of reliability) and the need for institutional knowledge and expertise to adopt AI.
Consider healthcare: AI has long been capable of transcribing conversations, but even deploying simple transcription applications to save time for clinicians while taking patient notes has run into challenges such as "hallucination": AI fabricating incorrect content as part of its output. This makes adoption slow, since hurdles usually need to be solved by domain experts in concert with technical experts.
AI adoption also runs into institutional hurdles. Even when technical challenges such as hallucination are addressed, improving patient outcomes has other roadblocks: privacy regulations, integrating with existing workflows, and training clinicians in using AI. Such technical and institutional hurdles lead many to incorrectly conclude that AI won't be useful in domains like healthcare; the reality is that adoption simply takes far longer than we typically expect.
The gap between the potential of AI and its practical implementation will define 2025. Despite improvements in AI models, using them productively in real-world settings, such as to meaningfully improve healthcare delivery, enhance education, or streamline government services, will remain challenging. The adoption of AI will require creating AI applications for different contexts, improving technical knowhow, and diffusing access and expertise across industries, not just in tech. This will be more time consuming and less striking that building AI that has better capabilities. It might take decades rather than years. But it will be essential for spurring the adoption of AI in useful ways across the economy.
Sayash Kapoor is a 2025 Mozilla Fellow.
From groundbreaking innovations to bold visions, our 2025 Fellows share their predictions on where technology is headed—and the impact it could have on the world.