In 2023, the fascination with artificial intelligence surged. As observers turned their attention to the practical applications of this technology, 2024 emerged as a pivotal year aimed at harnessing these AI marvels for tangible benefits without incurring exorbitant costs.
Arvind Narayanan, a computer science professor at Princeton University and co-author of the book “AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell The Difference,” noted that the focus has shifted from merely releasing models to actually developing functional products. The initial group of around 100 million users who engaged with ChatGPT upon its launch two years ago found mixed utility: while it excelled in some areas, it fell short in others.
Today, generative AI is becoming integral to a growing array of technological services, often without users consciously seeking it out. Examples include AI-generated results in Google searches and innovative features in photo editing software.
Narayanan remarked that the primary shortcoming of generative AI in the previous year was the public release of powerful models lacking clear applications for everyday users. This year, however, there has been a focus on developing products that effectively leverage these AI capabilities for practical use.
After OpenAI unveiled GPT-4 in March 2023, and other competitors launched similarly advanced AI language models, the pace of improvements has slowed, debunking the unrealistic expectations of rapid advancements towards superhuman intelligence. The discourse around AI has transitioned from existential threats to viewing it as a common technology.
While discussing the financial implications of AI, tech executives have faced scrutiny from Wall Street analysts regarding the return on significant investments in AI research and development. Creating AI systems that power generative tools like ChatGPT demands substantial financial commitment toward energy-intensive computing systems with costly AI chips. To combat this high energy consumption, some tech companies have pursued agreements to source nuclear power for their operations.
According to Kash Rangan, an analyst at Goldman Sachs, the capital invested in this technology amounts to hundreds of billions of dollars. An analyst from the same bank raised concerns that AI has yet to tackle complex issues that would justify its expenses, questioning whether AI models will ever replicate the nuanced capabilities inherent to humans. Nonetheless, Rangan maintains a positive outlook about AI’s potential, stating these tools have already shown incremental productivity improvements in various fields such as sales and design.
As artificial intelligence evolves, some workers have expressed anxieties about how these tools might affect their jobs. For instance, Borderless AI, a tech company, employs a chatbot from Cohere to draft employment contracts in countries like Turkey and India without involving attorneys or translators. Meanwhile, video game actors from the Screen Actors Guild raised alarms during their July strike, fearing that AI could diminish future job prospects by mimicking performances without consent. The transition to AI also influenced last year’s four-month strikes in the film and television sector.
Concerns regarding AI utilization have also been vocalized by musicians and authors worried about their likenesses and works being replicated. Walid Saad, an electrical and computer engineering professor at Virginia Tech, emphasized that generative AI struggles to create truly original pieces. He illustrated this challenge using a humorous example, noting that when asked to depict salmon in a river, the AI produced an image featuring cut supermarket salmon instead of the intended scene.
Saad argued that AI still lacks the common sense that humans possess, which hinders its reasoning capabilities. This reasoning is crucial for developing more consumer-friendly AI tools, as Vijoy Pandey, a senior vice president at Cisco’s innovation branch Outshift, highlighted. AI developers are now exploring generative AI agents capable of handling complex inquiries and proposing detailed solutions.
Looking ahead to 2025, Pandey predicts that AI agents will be capable of collaborating as teams to resolve issues, rather than simply functioning as standalone tools. In future technologies, including Bitcoin software, these agents will likely specialize in specific tasks, ensuring a comprehensive approach to problem-solving.
In the realm of medicine, AI tools have made notable strides, assisting healthcare professionals significantly. The Nobel Prize in Chemistry this year recognized AI-related research led by Google aimed at new medicine discovery. Saad confirmed that AI has streamlined diagnostics, allowing physicians to quickly identify potential care paths for patients. AI doesn’t identify diseases independently, but it effectively processes data, highlighting areas for doctors to examine further.
Despite these advancements, AI possesses limitations, such as the tendency to fabricate information. An example is OpenAI’s transcription tool, Whisper, which has been criticized for inconsistencies in output. Pandey mentioned that clients in the pharmaceutical sector have noted AI’s pivotal role in bridging experimental research with data analysis, expediting pharmaceutical development timelines from years to mere days.
“That has been the most dramatic use,” he concluded, showcasing the transformative potential of artificial intelligence within the medical field.
