Stanford Researchers Release AI Index on Global Trends

Researchers at Stanford University have released the “AI Index” study, detailing global trends in neural networks for 2023.

The report discusses the development of multimodal models, substantial financial investments in generative AI, new performance metrics, shifts in global opinion, and regulatory measures governing the field.

Open and Closed Models

In 2023, organizations released 149 AI models—twice as many as in the previous 12 months. Models with open-source code accounted for 65.7%, compared to 44.4% in 2022 and 33.3% in 2021.

Closed-source models outperform open-source chatbots. In 10 selected benchmarks, the former achieved an average performance advantage of 24.2%.

Leading Chatbot Developers

In 2023, Google led the industry in releasing the most models, including Gemini and RT-2. Since 2019, the corporation has been at the forefront of creating the most foundational neural networks—40, followed by OpenAI with 20.

For comparison, in 2024, the University of California, Berkeley released three models, and Stanford two. The reason was cost: training Gemini Ultra cost approximately $191 million, while GPT-4 cost $78 million. In contrast, in 2017, training the Transformer model, which underpins nearly all modern LLM architecture, cost about $900.

In 2023, the United States significantly outpaced other jurisdictions, developing a total of 61 models. Since 2019, the country has consistently led in creating the majority of major chatbots, followed by China with 15 neural networks.

How Close is AI to Human Capabilities?

As of 2023, AI has reached human-level performance in many significant tests, from reading comprehension to visual reasoning. However, in some tests, it still falls short of human levels.

This year, the index tracked several new parameters, including coding tasks.

Investments in the Sector

While private investments in the sector have been declining since 2021, generative AI is becoming increasingly sought after. In 2023, $25.2 billion was directed to this area. In 2022, this figure was $2.85 billion, and in 2019—$840 million. Generative AI accounts for more than a quarter of all private AI investments in 2023.

The US dominates this metric ($67.2 billion), followed by China ($7.76 billion) and the UK ($3.78 billion).

Neural Networks and Employment

According to surveys, 55% of organizations reported plans to use AI in 2023. In 2022, it was 50%, and in 2017—20%. Companies use AI for automating contact centers, personalizing content, and attracting new customers.

The index data suggests that globally, most people believe AI will change their jobs, and more than a third think neural networks will replace them. Generation Z and millennials expect more significant impacts from AI compared to Generation X and baby boomers.

The study indicates that more American regulators are adopting regulations to protect citizens and oversee AI tools and data.

For instance, the Copyright Office and the Library of Congress have approved guidelines for registering copyrights on works containing AI-generated materials, and the SEC has developed a cybersecurity risk management strategy, guidance, and incident disclosure plan.

In April 2024, Adecco Group published a study on AI’s impact on HR policies. The findings suggest that artificial intelligence will lead to reduced hiring in the next five years.