By any measure, 2023 was a landmark year for AI, predominantly characterized by the rise of Large Language Models (LLMs) and their chatbot applications. However, significant progress was also made across various dimensions, including image, video, and voice generation technologies.
The convergence of these digital technologies has spawned new use cases and business models, leading to the emergence of digital humans as influential figures and even newscasters, gradually replacing human counterparts.
Adoption of AI in Daily Work
Importantly, 2023 marked a pivotal moment when numerous individuals began to intentionally integrate AI into their daily tasks. This rapid innovation has fueled bold predictions ranging from friendly home robots to the possibility of artificial general intelligence (AGI) within a decade. Nevertheless, progress may face challenges that could hinder these forecasts.
As AI becomes increasingly intertwined with our daily lives, it raises an essential question: What can we expect next?
Robotics on the Horizon
While digital breakthroughs continue to impress, advances in the physical domain, particularly robotics, are also gaining traction. LLMs may offer the critical cognitive capabilities needed for robots, especially when paired with image recognition technologies. This fusion enables robots to better understand and respond to human requests while navigating their surroundings.
Deepu Talla, Nvidia’s VP of robots and edge computing, emphasized that LLMs will enhance robots' ability to interpret human instructions, learn collaboratively, and better understand their environments.
To boost robot performance, researchers at MIT's Improbable AI Lab have developed a framework that employs multiple foundation models, each optimized for specific tasks such as language processing, vision, and action.
"Each foundation model captures a different aspect of the robot decision-making process, collaborating to make informed decisions," the lab noted.
However, simply integrating these models may not be enough for practical real-world application. To tackle existing limitations, Stanford University introduced a new AI system called Mobile ALOHA. This system enables robots to autonomously manage complex tasks, such as sautéing and serving, organizing cooking equipment, using elevators, and rinsing dishes.
The "ImageNet Moment" for Robotics
These developments have prompted Jack Clark to suggest that robotics may be approaching its "ImageNet moment," characterized by a decrease in the costs associated with learning robot behaviors and data acquisition.
The term "ImageNet" refers to a vast dataset of labeled images initiated by Fei-Fei Li in 2006, pivotal in advancing computer vision and deep learning research. Its significance surged following a 2012 breakthrough by researchers who created a convolutional neural network (CNN) architecture resulting in a dramatic reduction in image classification errors. This milestone effectively propelled AI into a new era.
Clark contends that we may now be on the brink of a similar breakthrough in robotics. If realized, bipedal robots could soon collaborate with humans across various settings, from hospitals and factories to homes, transforming how we manage everyday tasks.
Accelerating Pace of AI Progress
The pace of AI evolution is astonishing. Nvidia CEO Jensen Huang recently projected that AGI could be achieved within the next five years. Jim Fan, Nvidia’s senior research scientist, likened the past year in AI to a leap from the Stone Age to the Space Age.
Consulting firm McKinsey estimates that generative AI will contribute over $4 trillion annually to the global economy. According to UBS, the AI market is expected to surge from $2.2 billion in 2022 to $225 billion by 2027, reflecting a remarkable 152% compound annual growth rate (CAGR).
Enthusiasm for AI's potential to enhance our quality of life remains high. Bill Gates noted in his "Gates Notes" at the end of 2023 that "AI is about to supercharge the innovation pipeline." David Luan, CEO of the AI startup Adept, echoed this sentiment, declaring the continuation of rapid AI progress as inevitable.
Given this momentum, it is unsurprising that generative AI currently ranks at the peak of inflated expectations in the Gartner Emerging Technology Hype Cycle, a metric for gauging enthusiasm for new technologies.
Examining Future Challenges in AI
While celebrating AI advancements in 2023, we must also consider the challenges that lie ahead. The momentum behind AI is reminiscent of the internet boom during the dot-com era, which subsequently faced major setbacks.
A Fortune article suggested that 2024 might witness a period of retrenchment, as investors realize many companies lack sustainable business models and larger firms encounter the costs of computation outweighing their benefits. This aligns with Amara’s Law, suggesting that we often overestimate short-term technological impacts while underestimating long-term effects.
Historically, the AI field has experienced elevated expectations followed by "AI winters," periods of stagnation due to unmet promises in building and deploying applications. Two such winters occurred from 1974 to 1980 and again from 1987 to 1993.
As we bask in the current "AI summer," the risk of another downturn is real. The costs associated with computing and the environmental impact of AI model training raise sustainability concerns.
Additionally, the "Four Horsemen of the AI-pocalypse" — data bias, data security, copyright infringement, and hallucination — pose significant hurdles. The lawsuit filed by the New York Times against OpenAI and Microsoft highlights the precarious nature of AI business models, with potential ramifications for the entire sector.
The most pressing concern surrounds the existential risks posed by AI. While some view the advent of AGI as a means to achieve unprecedented prosperity, others, particularly advocates of Effective Altruism, warn of potential destruction.
Recent surveys of over 2,700 AI researchers reveal a sizable portion fears advanced AI could lead to human extinction, with median estimates placing a 5% chance or higher on this outcome.
A Balanced Perspective for the Future
The known and potential challenges serve as a reality check amidst the excitement surrounding AI. Despite this, the forward momentum suggests continued advancements in AI technology in 2024.
The New York Times noted that this year will likely be defined by rapid technological improvements, enabling AI to produce new media, mimic human reasoning, and infiltrate the physical world via innovative robotics.
Ethan Mollick, in his blog "One Useful Thing," expressed a similar belief that AI development will likely accelerate further before facing eventual constraints, whether technical, economic, or legal.
The upcoming year promises significant changes in AI, ideally leading to advancements that greatly enhance our quality of life, such as groundbreaking medical discoveries. However, it’s probable that the most ambitious expectations may not materialize immediately, resulting in a necessary adjustment of market sentiment — a normal aspect of hype cycles. We can only hope that these adjustments do not usher in another AI winter.
Gary Grossman is Executive Vice President of the technology practice at Edelman and serves as the global lead for the Edelman AI Center of Excellence.