Beyond Magic: AI in 2025

If any sufficiently advanced technology is indistinguishable from magic, then it’s fair to say 2024 was a magical year.?

From multimodal agents that can think, plan, and remember to experimental machines capable of tasks a traditional supercomputer couldn’t master in 10 septillion (10,000,000,000,000,000,000,000,000!) years, the innovation has been unprecedented.?

Not long ago we were lamenting a slowing pace of scientific progress. But computer science is entering a new realm, with computational tools that are already creating not just scientific breakthroughs, but breakthroughs in how we make scientific breakthroughs.

How did we get here???

Many people don’t recognize how much Search has contributed to the development of Large Language Models. In striving to provide relevant results, our engineers developed the tools to understand and apply the meaning of language at incredible scale:

• 2001: Google developed our "did you mean?" spell-check functionality–one of our first applications of machine learning.?
• 2013: Google introduces Word2Vec (“Word To Vector”)–expressing words as mathematical parameters (think of giant word clouds), leading to fundamental progress in natural language processing.
• 2017: Eight Google scientists introduced the world to our Transformer deep learning architecture with their landmark research paper "Attention Is All You Need,” showing how deep understanding of the context could help us not just predict the next words, but extract the meaning from complex texts.??
• 2024: Google’s Transformer is now not only the foundation of the large language models powering tools like Gemini and ChatGPT, it is also what’s driving every major AI technology breakthrough released today.

In other words, the rise of AI has been an overnight success many years in the making.?

And when we zoom out even farther, we see that thanks in large part to the tireless work of American AI scientists and engineers, plus enormous private-sector R&D investments, democracies now have the advantage when it comes to shaping a golden age of innovation in fields as diverse as healthcare, materials science, and nuclear fusion.

But as geopolitical rivals ramp up investments in AI models, data center infrastructure, new AI companies, and accelerated AI deployment, here are three predictions for what you can expect to see in 2025:

A new priority will emerge: Maintaining democracy’s AI edge.

We’ve been all in on helping protect and promote AI innovation so that America and its allies hang onto their leading position.?

Heading into 2025, many of my conversations with policymakers will focus on the need for a pro-innovation technology agenda that supports the private sector efforts driving today’s AI advances.

We have a generational opportunity to reignite scientific leadership, shape global standards, and confirm America’s position as the global leader in AI innovation.

Shame on us if we let that opportunity slip away.

Economies will have to choose between harmonization and fragmentation.

In October former Italian Prime Minister and European Central Bank Chief Mario Draghi delivered his long-anticipated report on the future of competitiveness.

The top line? The EU’s current regulatory approach has put it at risk of missing out on “future waves of digital innovation.”?

Over 100 European regulations target the digital economy. These regulations, while well-intentioned, are too often conflicting, untested, and inconsistently implemented. The effort to “future proof” regulations in a fast-changing technological environment has inevitably resulted in over-broad rules that are slowing both innovation and adoption.

And in America, we’ve seen early signs of a similar trend, with a patchwork quilt of new laws targeting AI.

We’ve long said AI is too important not to regulate, and too important not to regulate well.?

We shouldn’t reinvent the wheel, but rather focus on the gaps in existing laws. Using a scalpel rather than a broadsword will reduce the chances of unintended consequences or slowing beneficial innovations.

As we round out the year, it’s great to see people’s growing optimism about AI.

But the question remains: Will 2025 be the year countries seize AI opportunities, or hold them back?

To keep the optimism going, the private sector needs to play a leading role in building trust by working with legislators on technology-savvy rules and by creating and sharing norms and best practices for the development and use of AI.

As just one example, leading AI companies joined forces last summer to create the Frontier Model Forum, an organization focused on ensuring safe and responsible development of cutting-edge AI models. It’s also why Google spearheaded the formation of the Coalition for Secure AI (CoSAI)–so that leaders in AI could come together and collaborate on AI security research, share security learnings, and build open-source solutions.

In 2025 we’ll be keeping up the momentum on this work–be on the lookout for updates.

AI will transform the security landscape.

Thankfully, we didn’t see dramatic new AI-powered attacks at scale in 2024. But the threat landscape is constantly evolving. With geopolitical tensions rising, we can expect threat actors to move from experimenting with AI to adopting AI.?

That's the bad news. But here's the good news: For many years, we've kept more people safe than anyone else in the world, and now we're finding new ways to use AI to give the advantage back to defenders.?

Just last month, our Project Zero and DeepMind teams used an AI agent to find a previously unknown, exploitable memory-safety issue in widely used software–we believe this to be the first public example of an AI agent doing this.

That’s a big deal. Finding vulnerabilities in software before it's even released means we’re able to get the jump on attackers–acting before an attack even happens.

AI reduces the toll on defenders, lets investigations run more efficiently, and gives us unparalleled visibility to global threats to critical infrastructure.

The question is whether public and private sector partners will stay ahead of criminal hackers and foreign adversaries in the race to use these new tools.

Final thoughts

The magic isn’t stopping in the new year.?

The pace of AI advances is faster than ever, and we'll be focused on partnering with others to ensure our tools are both advancing the scientific frontiers and making everyday life better.?

We’re at a watershed moment. It’s easy to be excited about the year that was–and to be even more excited about the year that will be.