IFollowing the late 2022 release of ChatGPT, 2023 was the year of AI craze. 2024 saw steady progress as systems became smarter, faster, and cheaper to run. AI is also now reasoning more deeply and interacting via audio and video. AI experts and leaders say this trend will accelerate. Here’s what to expect from AI in 2025.
Better and better AI agents
AI futurist Ray Kurzweil predicts that by 2025 we will see a shift from chatbots and image generators to “agent” systems that can not only answer questions but also work autonomously to complete tasks. states. In October, Anthropic gave its AI model, Claude, the ability to use a computer to click, scroll, and type, but this may be just the beginning. Experts say it will enable agents to handle complex tasks like scheduling appointments and writing software. “These systems will become increasingly sophisticated,” says Ahmad Al-Dahle, vice president of generative AI at Meta. Jaime Sevilla, director of AI predictive nonprofit Epoch AI, envisions a future where AI agents act as virtual colleagues, but by 2025, AI agents will be primarily focused on novelty. It says that it will be. Melanie Mitchell, a professor at the Santa Fe Institute, warns that agents’ mistakes can have “huge consequences,” especially if agents have access to personal or financial information.
Read more: How the rise of the new digital worker will lead to unlimited age
national security priorities
Dan Hendricks, director of the Center for AI Safety, said governments will start looking at AI from a national security perspective, saying, “It’s about how many big decisions are made about AI.” While the U.S. has restricted China’s access to critical chips, Meta and Anthropic have forged close ties with U.S. intelligence agencies by allowing them to use their AI models. Amandeep Singh Gill, the UN Secretary-General’s special envoy for technology, said that “political developments around the world point us in the direction of continued competition” and that the United States and China have “no room for cooperation”. ” emphasized the need to maintain this.
Read more: How the benefits and harms of AI will grow in 2024
Governance races to catch up
While developers race to build smarter systems, governments around the world race to regulate them. The EU is leading the way in AI legislation. Its code of practice, due to be finalized by April and come into force from August, will be one of the first pieces of legislation targeting frontier AI developers, and many of the EU’s requirements will require companies to take a clear approach. is likely to have a global impact on how companies operate unless they choose to do so. “In a variety of markets,” says Markus Anderjung of the Center for AI Governance. In the United States, where more than 100 bills have been introduced in Congress, Anderjung predicts that “very little will happen” federally this year, although each state may act on its own.
facing investment challenges
Ruman Chaudhry, CEO of Humane Intelligence, told TIME in an email that next year “will be a year of reckoning.” “With billions of dollars invested, companies must now demonstrate value to consumers.” In health care, that value seems clear. For example, additional AI diagnostic tools are expected to receive FDA approval, and AI could also prove useful in discovering and monitoring the long-term effects of various medicines. But elsewhere, the pressure to prove benefits can cause problems. “The pressure to get money back from all these investments could push flawed models into the Global South,” said Jay Vipla, an AI policy researcher. says the market is less supervised than in Western countries. She points out that the trend in India to automate already exploitative jobs, such as call center work, is a cause for concern.
AI video becomes mainstream
In December, Google and OpenAI released an impressive video model. OpenAI’s Sora release suffered from access delays, while Google’s Veo 2 was released to a limited number of users. Sevilla hopes video generation tools will become more widely available as developers find ways to make them cheaper to run. Meta’s Al-Dahle predicts that video will also become an important input for AI, with systems analyzing video from smart glasses to provide real-time assistance across a variety of tasks, such as bicycle repair, not far away. I envision the future.