The Challenging Future of AI: Recent Setbacks and Mistrusts in Industry

 

Joon Young Kim

Assistant Professor, School of AI Convergence

Director, Institute for Digital Mobility and Robotics Research

 

 

Let’s imagine a time, you are so exhausted that you wish someone would bring you a mac-n-cheese or BLT sandwich, complete your calculus assignment, or drive you to a cool indie coffee shop without moving, asking, telling, or explaining anything. Well, now imagine all of this being handled by some"thing" not some"one," thanks to Artificial Intelligence.

 

The essence of Artificial Intelligence (AI) is simple: a computing device or computer that performs tasks typically requiring human intelligence. The rise of deep neural networks and learning in the early 2010s spurred the recent popularity of AI and sparked technical advancements across various fields, including language processing, vision and sound recognition, and autonomous systems. Today’s Large Language Model (LLM) demonstrate the potential for machines to process and tackle complex tasks and solve analytical problems much like humans do. The potential of AI seems unlimited with a future full of promising advancements.

 

However, recent industry performances in AI do not align with these potentials and are facing serious challenges. Autonomous driving, one of the leading technologies that heavily relies on AI, has not shown significant progress, and companies have pulled the plug on their autonomous research. Apple announced the cancellation of its car project, and Samsung discontinued its autonomous driving development in 2024. Even this October, the Telsa robot taxi showcase brought a huge letdown and further doubts about self-driving technologies.

 

LLMs are no different from autonomous systems. Their primary issue is their high operational costs, and current perceptions question whether their investment is worth it. Sources from The Information, one of the central technology digital media based in San Francisco, claim that OpenAI is facing an estimated $5 billion loss in 2024 due to its infrastructure, operational, and labor costs, cutting into their profits. Given that other LLM companies, such as Meta and Google, are in a similar situation as OpenAI, Wall Street investors have recently begun to question the possibility of monetizing AI, including LLMs, according to CNN.

 

Let me say that AI is still a promising technology and can have a significant impact on various fields as every researcher continues to discuss AI for their research progress and advancement. However, the recent setbacks in AI, especially in LLMs, are complex cases and may raise issues from various perspectives, including overcapacity and operations. As published in Fortune magazine recently, AI has become a general-purpose technology in the view of international governments and organizations, and they put their heavy bills on AI for future investments. It is obviously a logical and understandable move, and no one wants to fall behind this trend. However, in the past, we have always seen that such over-investment can cause the excess supply over demand, which may result in significant losses and instability in the related industry. As a prime example, steel has suffered from overcapacity, leading to issues like poor-quality production and even contributing to climate change. Unless significant improvements and profit generation happen, AI risks facing the same fate, potentially becoming the new "Steel."

 

Operational cost is another issue with AI. To achieve current expectations, AI needs to be trained, evaluated, and managed more extensively, making AI systems operable. This leads to an exponential cost increase in operation and infrastructure investment. PCMag reported that Meta and Microsoft requested approximately 350,000 H100 shipments from NVIDIA, which translates to the equivalent of several data centers in terms of computing powers, and other companies have  followed the suit. This operational demand also drives the heavy consumption of resources and electricity directly linked to climate change. It is too early to say that AI will cause a climate crisis, but it is close to becoming one of the leading environmental concerns.

 

The fundamental risk of AI is the difficulty in evaluating bias and correctness within its systems. Deep learning has become highly complex, with massive parameters, and how they are connected and interact remains in great difficulty to interpret and understand in theory. This “black box” issue can result in biased and incorrect outcomes, of certain content and understanding, making human intervention inevitable. The false classification of a specific ethnic group as an animal by Google Photos in 2015 highlighted the instability and potential risks that operators must constantly monitor AI systems. In other words, slight alterations in the deep learning system can occur easily, possibly resulting in serious or even fatal consequences.

 

Although many unresolved issues remain, emerging technologies, especially AI, always have the opportunity and potential to get back on track. For instance, Bitcoin, the foundation of all cryptocurrencies, has experienced ups and downs and suffered during the cryptocurrency crisis due to the Terra-Luna crash and the FTX collapse in 2022. However, the strong support from its community and the promising values of blockchain technology helped Bitcoin survive, and it even became a part of major financial assets with the Exchange-Traded Product (ETP) approval by the United States Security and Exchange Commission (SEC) in January 2024.

 

It is clear that current market and industry conditions are unfavorable to AI, and they are unlikely to improve soon. However, AI holds value and potential for contributing to society and community and making the world a better place. The EU AI Act and the White House Executive Order on AI from the United States indicate that governments are preparing for the future of AI at the highest levels. Even three recent Nobel laureates in Physics and Chemistry are associated with artificial intelligence. How to unlock and flourish AI's potential is the unsolved question that we have been trying and will try to figure out for a long time. It will be a challenging process, but we will find that answer eventually, giving us a clear direction and hope for AI’s role in a better future.