Author: Esther Shittu
Translator: Baihua Blockchain
For many, the AI agents of the future are like the character J.A.R.V.I.S. in the Marvel Cinematic Universe.
J.A.R.V.I.S., short for "Just A Rather Very Intelligent System", was initially created by the fictional industrialist and renowned investor Tony Stark as a natural language computer system. It later became an AI system that served as Stark's assistant. Even later, J.A.R.V.I.S. acquired a synthetic body and became the robot "Vision".
While AI agents - autonomous and semi-autonomous generative AI systems capable of independent action - may still be far from having physical embodiment, they may approach or even surpass J.A.R.V.I.S. at some point next year.
In the second half of 2024, the surge in popularity of AI agents will be similar to the rapid rise and transformation of the AI market by ChatGPT and other generative AI systems in 2022. Vendors seem to be shifting quickly from developing the latest large language models (LLMs) and AI chatbots to creating agents and action models.
For example, Salesforce launched Agentforce, a low-code agent-building tool, last fall. Microsoft has introduced the AI Agents Service, a community platform to help developers build AI agents.
Other vendors have also introduced AI agents into enterprises, automating various business processes. Analyst firm Forrester Research lists 400 vendors currently building agents.
"The excitement around them is very high right now," says Forrester Research analyst Craig Le Clair. "But there are also some risks, because you're unleashing an automation process that can execute without human oversight and balance."
The coexistence of excitement and risk means AI experts and vendors have high expectations for AI agents by 2025.
1. Eliminating Confusion Through Real-World Applications
One expectation is that while 2024 laid the groundwork for AI agents, 2025 will be the year they are ready for enterprise adoption, say AI market experts.
This means the confusion around agents will dissipate, says AJ Sunder, co-founder and CIO of Responsive, an AI-driven proposal and response software provider.
"There's a lot of confusion between agents and automation, agents and RPA (robotic process automation)," Sunder says. "Most of that confusion will go away. Then we'll start to see more agents deployed and used in the real world."
RPA uses robots or bots to automate repetitive tasks without relying on AI, while agents involve AI technology. RPA is deterministic and predictable, while agents are not.
"The similarity is that they're both digital colleagues," Le Clair says. "The difference is that when you add AI into the digital colleague, we call that an AI agent, and it's smarter, can understand context, and knows how to avoid getting stuck."
Sunder says some real-world applications of agents will emerge in customer service; others will be in finance or fraud detection.
"Any complex task that requires AI memory, planning, and executing multi-step, complex tasks, I think agents will play a huge role in that," Sunder says.
One complex application is video creation.
"A lot of these agent AI solutions can actually be deployed in a way that assists the video creation process," says Shahzaib Aslam, head of research at Colossyan, an AI video platform.
AI agents can help create an engaging video, provide compelling arguments, and include a call to action to encourage customers to take action, such as purchasing a product, Aslam says.
"It becomes a very powerful tool because it's going to help you create a more engaging, more successful video," he says.
Agents will not only play a role in different application scenarios like video creation, but many will also start using them to solve scalability problems, says Gartner analyst Tom Coshow.
However, there are different levels of application and use of AI agents, says Peter van der Putten, head of the AI Lab and chief scientist at Pegasystems, a workflow automation and decision management provider.
He states that at one end of the spectrum, AI agents can read, integrate, and synthesize information to draw certain conclusions, but take no action. At the other end, AI agents take action based on the information they have synthesized.
"The true success of agents is not in the intelligence of the agents themselves, but in how they are embedded into real-world applications," he says.
However, he continues, most enterprises must experiment before seeing the value of AI agents.
"I'm sometimes even surprised at what these systems can do," van der Putten says. "The only way to understand that is through safe experimentation."
2. Better Reasoning Models
Another expectation around AI agents is that large language models (LLMs) will continue to serve as their brains. This means LLMs need to become stronger in reasoning to enable AI agents to better execute their tasks.
Aslam says chain-of-thought prompting has already demonstrated this.
The idea is that the model not only generates an answer to a question, but generates multiple answers and reasons through a series of steps to arrive at the final answer.
While this may be more costly, as enterprises need to run multiple inferences to generate the chain of thought, it also enhances the reasoning capabilities of the models, Aslam says.
He adds that this will be an area of deep exploration for the AI industry and academia in 2025.
"This way of injecting explainability into the models makes a lot of sense, and we'll see more work and research going into that direction, scaling up the computation when reasoning and having the models arrive at predictions in a systematic and reasoned way, rather than just generating content," he continues.
3. Specialized Task Agents
While more agent-based use cases may emerge by 2025, this will not eliminate the need for human intervention.
However, with the new levels of automation brought by AI agents, the fear of job displacement remains.
Some in the industry say that while AI agents will have a certain level of autonomy by 2025, they will not be fully autonomous. In other words, AI agents will take on a portion of an individual's work, but not the entire job. For example, an AI agent may help you find the contact information for a travel agency, but cannot complete the entire booking process.
"We'll see agents not fully taking over entire job roles, but taking on a portion of a person's responsibilities or a portion of a process, then working in conjunction with traditional automation systems, human collaboration, and other agents," says Mark Greene, senior vice president and general manager at UiPath.
These agents that take on partial responsibilities will be specialized and complete tasks in a singular way. This will make the AI agents more precise in their task execution, Greene says.
"The more defined the responsibility, the more measurable the impact," he says.
4. AI Agent Infrastructure
In addition to the rise of single-task AI agents, 2025 may also be a year for building the infrastructure for AI agents, says Olivier Blanchard, an analyst at Futurum Group.
To enable AI agents to communicate with other agents, and even collaborate with humans on tasks, a coordination layer is needed, Blanchard says.
"2025 won't be the year we see fully mature agent-based AI," he says. "2025 will be the year we build the infrastructure for it, the foundational framework."
He adds that the vendors that may help build this infrastructure could include chip manufacturers like Qualcomm, Intel, and AMD.
"Qualcomm's processors will be primarily used for agent-based AI on devices," Blanchard continues. Meanwhile, Nvidia's processors are currently more used for cloud-based collaboration with agent-based AI.
"NVIDIA's GPUs have already been widely used in training AI models, laying the foundation for future agent-based AI," he said. "In two years, agent-based AI will be a hybrid of cloud-based and device-based software, working in coordination with each other."
Currently, NVIDIA mainly collaborates with the cloud, while Qualcomm focuses primarily on the device side. On the other hand, device manufacturers like Apple and Samsung will participate in creating the coordination layer, allowing agent-based AI to work in coordination across platforms, devices, and applications, Blanchard said.
"We already have these foundations," Blanchard said. "What we're missing is a system that can 'do everything.'"
5. One way to move towards the coordination layer is multi-modal AI
While generative AI systems like ChatGPT have input-output functionality, they cannot represent human connections to other applications.
However, as multi-modal AI evolves and matures, enabling image input to generate video output, this will facilitate agent-based AI to work better.
"As the models become more intelligent, our agents will also become more intelligent," Coshow said.
Blanchard said that AI agents need a coordination layer that can work across platforms and devices. The coordination layer consists of links that can allow AI agents to switch from one platform or interface to another, or from one application to the next.
If Qualcomm establishes its own coordination layer, and AMD also establishes its own coordination layer, this will make the interoperability of agent-based AI a major challenge.
"If all chip manufacturers are using their own coordination layers, they may not be able to communicate well with each other," Blanchard said.
6. Challenges for agent-based AI in 2025
Similar to other AI technologies, agent-based AI will face a series of challenges in 2025. One of them is the data problem.
Because data is often scattered across different sources and processes, providing agent-based AI with the data needed to perform tasks may become very challenging, Greene said.
Another issue is the lack of knowledge about the design process for agent automation, Greene added.
For example, the industry needs to understand when humans should interact with agent-based AI, how to interact, and through which channels to communicate with agent-based AI, he said.
Another challenge is the trust issue, Sunders said.
"If the underlying technology still relies on generative AI and large language models, then those shortcomings will also be inherited by agent-based AI," he said.
Despite these obstacles, Sunders believes that 2025 will be an important year for agent-based AI.
"We'll figure out where agent-based AI makes sense, how to deploy them, how to earn trust, and then fully let go," he said. "The promise of it being fully autonomous, I think that will ultimately be realized; but whether it will be in 2025, I don't think so."
Link to the article: https://www.hellobtc.com/kp/du/12/5608.html
Source: https://www.techtarget.com/searchenterpriseai/feature/Next-year-will-be-the-year-of-AI-agents
