IIf 2023 was the year of AI fervor, following the late 2022 release of ChatGPT, 2024 was marked by a steady pace of advancement as systems became smarter, faster and cheaper to operate . AI has also begun to reason more deeply and interact via voice and video – trends that AI experts and executives say will accelerate. Here’s what to expect from AI in 2025.
More and better AI agents
By 2025, we will begin to see a transition from chatbots and image generators to “agent” systems that can act autonomously to complete tasks, rather than just answer questions, says futurist Ray Kurzweil in AI. In October, Anthropic gave its AI model Claude the ability to use computers – clicking, scrolling and typing – but that may just be the start. Agents will be able to handle complex tasks such as scheduling appointments and writing software, experts say. “These systems are going to become more and more sophisticated,” says Ahmad Al-Dahle, vice president of generative AI at Meta. Jaime Sevilla, director of the nonprofit Epoch AI, envisions a future in which AI agents function as virtual colleagues, but says that in 2025 AI agents will primarily focus on being new. Melanie Mitchell, a professor at the Santa Fe Institute, warns that agents’ mistakes could have “severe consequences,” especially if they have access to personal or financial information.
Learn more: How the rise of new digital workers will lead to a limitless era
A national security priority
Governments will increasingly view AI through a national security lens, says Dan Hendrycks, director of the Center for AI Safety: “That’s how many big decisions about AI will be made. » The United States has limited China’s access to critical chips, while Meta and Anthropic have forged closer ties with US intelligence agencies by allowing them to use their AI models. “Political developments around the world are moving us towards continued competition,” said the UN secretary-general’s envoy for technology, Amandeep Singh Gill, stressing the need to preserve “pockets of collaboration” between the United States. United and China.
Learn more: How the pros and cons of AI have increased in 2024
The race for governance to catch up
As developers compete to build ever smarter systems, governments around the world are racing to regulate them. The EU is leading with its AI Law. Its code of practice, which is expected to be finalized by April and enforced from August, is one of the first laws targeting cutting-edge AI developers, and many of the EU requirements will likely have a global impact on the way businesses operate, unless they choose to take distinct approaches. in different markets, explains Markus Anderljung of the Center for AI Governance. In the United States, where more than 100 invoices have been brought to Congress, Anderljung predicts that “very little will happen” at the federal level this year, although states can act independently.
Facing the investment test
The coming year “will be a year of taking stock,” Rumman Chowdhury, CEO of Humane Intelligence, told TIME in an email. “With billions invested, companies must now demonstrate their value to the consumer. » In healthcare, this value seems clear: for example, additional AI diagnostic tools are expected to gain FDA approval, and AI could also prove useful in discovering and monitoring long-term impact of various medications. But elsewhere, the pressure to demonstrate results can create problems. “Because of the pressure to make all these investments profitable, flawed models could be imposed on countries in the Global South,” says Jai Vipra, an AI policy researcher, noting that these markets are less monitored than Western markets. In India, she cites trends toward automation of already exploitative jobs, such as call center work, as a source of concern.
AI video is becoming mainstream
In December, Google and OpenAI released impressive video models. OpenAI’s Sora launch was plagued by access delay, while Google’s Veo 2 was released to select users. Seville expects video generation tools to become more widely accessible as developers find ways to make them less expensive to use. Meta’s Al-Dahle predicts that video will also become a key part of AI, envisioning a not-too-distant future in which systems analyze video from smart glasses to offer real-time assistance for various tasks, like repairing a bicycle .
