Originally published on Forbes
The term “agent AI” hype is the latest trending buzzword for repackaging pies into Sky AI ambitions, but it does not imply any particular advancements that could achieve them. It amplifies the excesses of stories of us rapidly heading towards a major leap in autonomy. Towards the most extraordinary, all the boldest goals, Artificial general informationa speculative idea of a machine that can automate virtually all human work.
Setting unrealistic expectations can undermine true value. Generation and Predictive AI It offers opportunities for ever-growing concrete, but the claim that technology will soon retain its “agents” is a microcosm of steam machines. It sets the industry for misleading, costly, avoidable disillusionment.
Most high-tech terms such as machine learning, predictive modeling, and autonomous driving are legal. They represent one of two new goals of a particular technological approach or technology. However, the terms “agent” and “agent” fail in both respects. 1) Most uses of “agents” do not refer to new technological methodologies, but 2) the ambition of increasing autonomy is not new. Here is a breakdown of these two obstacles and their impact:
1) “Agent” does not refer to any particular technology or advancement
“Nothing attracts crowds like a crowd.” – 19th century circus showman, Barnum, famous for hoaxes
“Agent AI” pretends to be reliable, short-term capabilities, but it represents only the most obvious goals possible for technology (increasing automation). Certainly, we would like a large-scale language model to complete monumental tasks in itself, such as information collection and assimilation, and completion of online tasks and transactions, but labeling such ambitions as “agents” makes them more feasible.
The term “agent AI” is inherently misleading. Its pure popularity broadens the belief that technology will soon be able to operate more autonomously, but buzzwords do not refer to any particular technical approach that could lead us there. That trend helps institutionalize the notion that we are approaching a very new level of automation. “Agent AI” is so ubiquitous that it can sound “established” and “real”. This means the existence of groundbreaking advancements that are not actually possible.
Despite the fact that the vast majority of reports on “agent AI” only promote this hype narrative and encourage it to support it, autonomy itself is often a valuable goal, and researchers do valuable work to pursue it. For example, recently Carnegie Mellon University and Amazon collaboration We curate a large testbed of modest tasks to assess the degree to which LLMS can be managed autonomously. The study focuses on information retrieval tasks such as “Get articles discussing recent trends in renewable energy from the Guardian” and “Get published research papers on quantum computing from the MIT website.” This study evaluates clever approaches to using LLM to navigate websites and perform such tasks automatically, but does not say that these approaches constitute groundbreaking technology. Rather, they are ways to leverage LLMS, which is already groundbreaking. As this study revealed, cutting edge is currently failing at 43% of the time on these modest tasks.
2) “Agents” do not present new goals or objectives
“Agent AI” spotlights the autonomy of the machine as if it were a new ambition, but it is an old, self-evident goal. There is no new, innovative thrust. There are some buzzwords Adaptability and fuzzywhich generally refers to the desire for increased autonomy. “Agent AI” means a virtual machine that can perform substantial tasks on its own. This has always been a core fundamental purpose. The very purpose of any machine is to automate some or all of what is otherwise performed by people and animals. Put another way, we build machines to do things.
By repeating the innate desire to automate, the “agent” is simply stating what is obvious. Certainly the more machines can be safely for us, the better. However, there are rather stubborn limitations to the scope of tasks that can be fully automated in the absence of humans in a loop. For example, predictive AI is instantly Decide whether each credit card fee is permittedOn the other hand, wholesale exchanges for doctors are at best a long way. “Agent AI” is just as redundant as “The Lord of the Evil Sith”, “The Library”, and “Data Science.”
To be clear, autonomy is often a valuable goal, and LLM may be better, at least when the scope of automation is somewhat modest. Economic benefits put pressure on increasing autonomy, and various social concerns put pressure on them both direction. However, the range of autonomy for a released machine increases very slowly. One reason is that the technology doesn't improve as quickly as it was promoted. Another is that cultural and social inertia tend to slow adoption.
The loud concept of a machine “agent”
There is another problem with using the terms “agent” and “agent” to evoke the goals of an autonomous machine. Creating a machine in an “agency” is fantastic. This doubles into the core myths of AI and original sin, the personification of machines. Machines are no longer human tools. Rather, they are promoted to have their own human-level understanding, goal setting and will. It's our companion. Essentially, it is alive.
The voluntary goal setting that accompanies the agency, and the resulting bottlerability, has permeated the AI narrative for many years. “Working AI doesn't just stay in the lab.” Kevin Loose writes in New York Times. “It appears in the weapons used by the military and software used by children in the classroom.” Another articlehe writes: Similarly, Elon Musk, one of the most effective AGI Hype transmitters in the world, has announced a safety guarantee that skillfully implies intentional or dangerous AI. He says that future humanoid robots in his company will stick to obey every time everyone says it.Stop, stop, stop. ”
The story of technology taking on its own life is an old drama. We need to see this high-tech myth about what it is: a more persuasive, streamlined ghost story. If she's well-versed in algorithms, Mary Shelley wrote novels. Incredible and unsupported concept that we are actively progressing in AGI – Alias Artificial Man – It's at the root of many hype (and often Explicitly overlay (same). “Agents” evoke this story.
Despite unprecedented capabilities, and creepy, Apparently human nature – Of generative AI, the limitations of being able to fully automate human work continues very slowly and slowly. I think we generally need it Calm down for partial autonomy.
Do not buy or sell “Agent AI.” It is an empty buzzword that is overkill for most uses. The AI industry is primarily operating – not entirely – but certainly not in hype. To the extent that it continues to over-inflate expectations, the industry will ultimately face the disastrous disillusionment and unmet debt that arise from unmet promises.
I followed up on this topic in my second article.Agent AI's hype cycle is out of control, but widely normalized. ”
About the author
Eric Siegel is a leading consultant and former professor at Columbia University, helping companies deploy machine learning. He is the founder of the long-term Machine Learning Week Conference series, an instructor for the highly acclaimed online course, Machine Learning Leadership and Practice – End-to-End Learning, and an executive editor of the Machine Learning Times and a frequent keynote speaker. He wrote a predictive analysis of bestsellers. Used in hundreds of university courses to predict clicks, purchases, lies, or death, and AI playbook: the acquisition of the rare art of machine learning deployment. Eric's interdisciplinary work fills the stubborn technology/business gap. In Columbia, she received a renowned faculty award while teaching graduate computer science courses in ML and AI. He later served as a professor of business school at UVA Darden. Eric also publishes an OP-EDS on Analysis and Social Justice. You can follow him on LinkedIn.
