AI may boost productivity, but we are already becoming reliant on it, and that’s a problem

A Stanford report says that as AI continues to develop, navigating its benefits and potential drawbacks will be crucial.Just ten years ago, AI systems couldn’t even classify images as well as humans. Now, they’re routinely outperforming people on a range of tasks, according to a new report from the Stanford Institute for Human-Centered Artificial Intelligence (HAI). The 2024 AI Index is the latest update to the annual analysis of trends in artificial intelligence. Led by a team of experts from academia and industry, it’s one of the most in-depth reports on the technology and its impact available. This year’s edition tracks research, development, technical performance, responsible practices, economics, policy, public opinion, and more.

The report aims to provide decision-makers with a clear picture of what’s happening in the world of artificial intelligence. This includes policymakers and business leaders who need to understand this rapidly evolving technology.

Studies by Microsoft, Harvard Business School, and others found AI tools help workers complete tasks faster and improve the quality of their work. Generative AI, a powerful new technology, even shows promise in assisting with legal tasks like contract drafting.

Interestingly, the report suggests the technology might help bridge the skill gap between low-skilled and high-skilled workers. Research from Harvard Business School found that while high-skilled workers with AI assistance still outperform their low-skilled counterparts, the performance gap narrows significantly when AI is available.

However, the report also warns against over-reliance on artificial intelligence. When workers become complacent and overly trusting of AI’s results, their performance suffers compared to those who scrutinize the tech’s output more critically. Additionally, many executives surveyed believe AI will lead to job cuts, with 43 percent anticipating staff reductions.

The report also highlights a trend noted last year: training models is getting expensive. New estimates show cutting-edge systems like OpenAI’s GPT-4 cost a staggering $78 million to train. Google’s Gemini comes in even higher at $191 million. Compare that to just a few years ago, where state-of-the-art models cost a fraction of the price. The original transformer model (2017) cost around $900 to train, while RoBERTa Large (2019) was $160,000.

The US remains the global leader in AI. In 2023, American institutions produced the most significant models (61) compared to the European Union (21) and China (15). The US also attracts the most investment in AI, with a total of $67.2 billion in private investment in 2023, nearly nine times more than China.

But China is catching up. They’re the world leader in robot installations, putting in more robots in 2023 than the rest of the world combined. Additionally, China dominates AI patents, holding 61 percent of the world’s total.

There’s a shift towards open-source AI models, with over half (65.7 percent) released in 2023 being open-source compared to just 44.4 percent in 2022. This trend coincides with a significant rise in open-source AI projects on platforms like GitHub.

The report highlights the growing excitement around multimodal AI, which can handle various data types like text, video, and audio. This paves the way for robots that can function more effectively in the real world. However, comparing and evaluating these powerful models remains a challenge, as developers primarily use different responsible AI benchmarks.

With AI’s growing capabilities, regulations are also on the rise. The US saw a jump from just one AI-related regulation in 2016 to 25 in 2023. Public perception is also shifting, with over half of people expressing more concern than excitement about AI’s impact. A separate survey suggests two-thirds believe AI will significantly affect their lives within the next few years.