[ad_1]
AI agents are supposed to be the next big thing in AI, but there is no precise definition of what they are. Up to this point, people cannot agree on what exactly constitutes an AI agent.
At its simplest, an AI agent is best described as an AI-powered program that performs a series of functions on your behalf that a human customer service agent, HR employee, or IT help desk employee may have done in the past, although it may eventually include Anything important. You ask it to do things, and it does it for you, sometimes crossing multiple systems and going beyond just answering questions. For example, last month Perplexity released an AI agent that helps people do their holiday shopping (and it’s not the only one). Last week, Google announced its first AI agent, called Project Mariner, which can be used to find flights and hotels, shop for household items, find recipes, and other tasks.
Sounds simple enough, right? However, the matter is complicated by a lack of clarity. Even among tech giants, there is no consensus. Google views them as task-based assistants depending on the job: programming assistance for developers; helping marketers create a color scheme; Help an IT professional track down the issue by querying the log data.
For Asana, an agent may act as an additional employee, taking care of assigned tasks like any good coworker. Sierra, a startup founded by former Salesforce co-CEO Brett Taylor and Google vet Clay Bavour, sees agents as customer experience tools, helping people achieve actions that go far beyond chatbots of yesteryear to help solve more complex sets of problems.
This lack of a coherent definition leaves room for confusion about exactly what these things will do, but regardless of how they are defined, agents help complete tasks in an automated manner with as little human interaction as possible.
It’s still early days and that could be why there’s no agreement, says Rodina Cecere, founder and managing partner at Glasswing Ventures. “There is no single definition of what an ‘AI agent’ is. However, the most common view is that an agent is an intelligent software system designed to autonomously perceive and reason about its environment, make decisions, and take actions to achieve specific goals.”
She says they use a number of artificial intelligence techniques to achieve this. “These systems integrate multiple AI/ML techniques such as natural language processing, machine learning, and computer vision to work in dynamic domains, independently or alongside other agents and human users.”
Over time, as AI becomes more capable, AI agents will be able to do more on behalf of humans, and there are already dynamics at play that will drive this development, says Aaron Levy, co-founder and CEO of Box.
“With AI agents, there are multiple components of a self-reinforcing flywheel that will dramatically improve what AI agents can achieve in the near and long term: GPU price/performance, model efficiency, model quality and intelligence, and AI frameworks and AI frameworks,” Levy wrote. : “Infrastructure improvements.” On LinkedIn newly.
This is an optimistic view of technology that assumes growth will occur in all of these areas, when in fact this is not necessarily a given. MIT robotics pioneer Rodney Brooks noted in a recent TechCrunch interview that AI will have to deal with problems much harder than most technologies, and won’t necessarily grow in the same rapid way that chips grow under, say, Moore’s Law.
“When a human sees an AI system performing a task, they immediately generalize it to similar things and estimate the efficiency of the AI system; “Not just performance, but efficiency related to that,” Brooks said during that interview. “They are often overly optimistic because they are using a model of a person’s performance on a task.”
The problem is that systems are difficult to cross, and this is further complicated by the fact that some legacy systems lack basic access to an Application Programming Interface (API). While we’re seeing the steady improvements that Levy alluded to, getting software to access multiple systems while resolving the issues it might encounter along the way may be more difficult than many think.
If so, everyone may be overestimating what AI agents should be able to do. David Cushman, research leader at HFS Research, sees the current crop of bots as looking a lot like what Asana does: assistants that help humans complete certain tasks in order to achieve some sort of user-specified strategic goal. The challenge is to help the machine deal with emergencies in a truly automated way, and we’re clearly not there yet.
“I think this is the next step,” he added. “It’s where AI operates autonomously and effectively at scale. So, this is where humans set guidelines, guardrails, and apply multiple techniques to take the human out of the loop — when everything is about keeping the human in.” in “The episode with GenAI,” he said. The key here, he said, is to let the AI agent take charge and apply true automation.
This will require creating an AI agent infrastructure, a technology stack specifically designed to create agents (however you define them), says John Turo, partner at Madrona Ventures. In a recent blog post, Toro Illustrated examples of artificial intelligence agents Currently operating in the wild and how they are built today.
In Turo’s view, the increasing prevalence of AI agents — and he also acknowledges that the definition is still a bit elusive — requires a technology stack like any other. “All of this means our industry has work to do to build the infrastructure that supports AI agents and the applications that depend on them,” he wrote in the article.
“Over time, the logic will gradually improve, leading models will come to guide more workflows, and developers will want to focus on the product and the data – the things that set them apart. They want the underlying platform to work only with scale, performance and reliability.”
Another thing to keep in mind here is that it will likely take multiple models, rather than a single MBA, to make agents work, which makes sense if you think of these agents as a set of different tasks. “I don’t think at the moment that any single big language model, at least publicly available, monolithic big language model, is capable of dealing with agentic tasks. I don’t think they are yet able to do the multi-step thinking that would make me excited about the future of agency,” he said. “I think we’re getting close, but we’re not there yet,” said Fred Havemeyer, head of AI and software research at Macquarie US Equity Research.
“I think the most effective agents will probably be multiple combinations of multiple different models with a routing layer that sends requests or claims to the agent and the most effective model. I think it will be an interesting (automated) supervisor, delegating some sort of role.”
Ultimately, for Havemeyer, the industry is working toward that goal of agents working independently. “As I think about the future of agents, I want to see and hope to see agents that are truly autonomous and able to take abstract goals and then think through all the individual steps in between completely independently,” he said. TechCrunch.
But the reality is that we are still in a transition period with regard to these clients, and we do not know when we will reach this end state that Havemeyer described. While what we’ve seen so far is a promising step in the right direction, we still need some progress and breakthroughs for AI agents to be able to function as they are envisioned today. It is important to understand that we are not there yet.
This story was originally published on July 13, 2024, and has been updated to include new clients from Perplexity and Google.
[ad_2]