Advances in AI capabilities

In 2024, artificial intelligence systems have notably surpassed human performance in several key areas: image classification, visual reasoning, and English understanding. This achievement is a result of the ongoing refinement of AI algorithms and the integration of more sophisticated training data.

Image classification, which is powerful in applications ranging from medical diagnostics to autonomous vehicle technology, benefits greatly from AI’s ability to process and analyze vast quantities of images far more quickly than human analysts.

In visual reasoning, AI systems apply complex algorithms to interpret and make decisions based on visual data, an area increasingly utilized in security systems and manufacturing.

Improved performance in English understanding underscores advancements in natural language processing, enhancing AI’s role in customer service, content generation, and comprehensive data analysis.

Limitations in complex areas

Despite these advances, AI still falls short in several complex domains, primarily in competition-level mathematics, visual commonsense reasoning, and strategic planning. These areas require raw computational power and an intricate understanding of abstract concepts and creative problem-solving strategies, which remain challenging for AI.

  • Competition-level mathematics: The ability to conceptualize and solve novel problems involves a level of intuitive knowledge that AI systems, reliant on existing data, struggle to emulate.
  • Visual commonsense reasoning: Demands an understanding of everyday objects and scenarios that humans generally take for granted—capabilities that AI has not yet fully mastered.
  • Planning: Especially in unpredictable and dynamic environments, AI systems often cannot match the human ability to anticipate unknown variables and adapt strategies accordingly.

Surge in AI model production

2023 was a major year for AI research and development, particularly highlighted by the production of 51 machine learning models by private companies, a sharp increase compared to the 15 models produced by academic institutions. This indicates a robust engagement from the private sector in pushing the boundaries of AI technology.

Companies are investing heavily in developing new AI models while playing a key role in the practical application of these technologies in the market.

Collaborations between industry and academia have also been fruitful, yielding 21 high-profile models. Partnerships leverage the theoretical and experimental strengths of academic researchers combined with the practical and financial resources of industry players.

These types of collaborations are essential for driving AI innovation, facilitating a cross-pollination of ideas, and making sure that theoretical advancements are translated into practical applications that can benefit society at large.

Escalating costs of AI training

Training cutting-edge AI systems such as OpenAI’s GPT-4 and Google’s Gemini Ultra requires a massive financial commitment, evidenced by the enormous costs incurred.

OpenAI’s GPT-4, a language model renowned for its ability to generate human-like text, required an estimated $78 million in computing resources for its development. Google’s Gemini Ultra model, which is part of a new generation of even more complex AI systems, needed a staggering $191 million to train.

These numbers spotlight the scale of investment necessary to push the boundaries of AI tech. They reflect the high costs associated with the extensive computational power and energy required, as well as the data acquisition and the skilled personnel needed to develop and refine such sophisticated models.

Geographic and institutional dominance in AI

The United States continues to lead in the production of state-of-the-art AI models, with 61 notable systems developed in 2023 – surpassing that of other regions by a large margin.

China and the European Union are still key players, producing 15 and 21 notable models, respectively.

The U.S. dominance is facilitated by larger funding pools and a robust tech industry ecosystem, and strong collaborations between academia and the private sector, while China and the EU are accelerating their efforts to close the gap through increased investments and government support.

Funding shifts and priorities

Despite a general downturn in private AI investment for the second consecutive year, the sector saw a dramatic surge in funding for generative AI technologies, which nearly octupled, reaching $25.2 billion in 2023.

This is largely due to the growing commercial interest and potential revenue streams that generative AI promises, particularly in industries like media, entertainment, and advertising, where the ability to generate realistic images, videos, and text can transform content creation processes.

Contrast between the overall decline in AI investment and the booming funding for generative AI indicates a strategic shift in investor focus, prioritizing technologies with immediate practical applications and high growth potential.

Challenges evaluating AI systems

Current evaluations of AI systems are a challenge due to the absence of standardized testing protocols across the board for measuring responsibility, safety, and security. Lack of uniformity makes it difficult for industry leaders to assess and compare the inherent risks and limitations of various AI models systematically.

For AI to integrate safely into societal frameworks—whether in healthcare, finance, or transportation—developers, regulators, and users must have consistent benchmarks. Standards would promote transparency and foster trust among consumers and businesses alike by making sure that all AI systems adhere to agreed-upon norms for performance and ethics.

Emerging risks and public perception of AI

New risks and concerns

One of the more pressing concerns in the AI field is the creation and circulation of political deepfakes—synthesized media in which a person in an existing image or video is replaced with someone else’s likeness.

Deepfakes are becoming increasingly easy to produce and yet are difficult to detect, posing complex challenges for maintaining political integrity and public trust.

Ongoing research is revealing complex vulnerabilities in language models that could potentially be exploited to produce misleading or harmful content. These could lead to misinformation spreading or even malicious use in cyber-attacks, highlighting an urgent need for more robust security measures in AI development.

Changes in public opinion

Public perception of AI is shifting, with more individuals globally recognizing the profound impact AI is likely to have on their lives.

Data indicates that the share of people who believe AI will dramatically affect their lives within the next 3-5 years has increased from 60% to 66%.

In the United States, the change in sentiment is even more pronounced. The proportion of Americans who feel more concerned than excited about the proliferation of AI has risen sharply from 37% in 2021 to 52% in 2023 – suggests growing apprehension about the role of AI in daily life. This also points out an increasing awareness of the potential consequences of AI technologies, both positive and negative.

Tim Boesen

June 4, 2024

5 Min