NORBERT BIEDRZYCKI

The emergence of LAMs: a new frontier in the development of artificial intelligence

Large Action Models

What are Large Action Models, how do they differ from Large Language Models, what capabilities and limits do they have, and what can they be used for?

Unlike traditional AI assistants, usually limited to answering questions or executing simple commands, Large Action Models (LAMs) are a new class of Artificial Intelligence models. Not only can they understand instructions in natural, spoken, written language, but they go a step further and can perform complex tasks independently.

While AI assistants powered by the Large Language Model (LLM) predict the next word in the text, the autonomous agent powered by LAM can predict the next action in a complex process. LAMs have an uncanny ability to capture our intentions and behaviors, allowing them to operate the machine’s user interfaces and perform actions on our behalf.

Imagine the great convenience of simply asking an AI assistant to book a ride, products or services online, booking a table at a restaurant, and even performing advanced, multi-step business processes, all without any hassle and even anticipating our next queries or actions.

The next step to GenAI

LAMs are trained on extensive data sets on human interactions, processes, and methods of operation. Thanks to this, they can learn how to use virtually any application or software or device, even those without native AI support. Interestingly, they can continuously learn from our opinions and preferences, adjusting their capabilities to seamlessly adapt to our unique needs and habits. As if they anticipated our next step.

But what distinguishes large action models from their predecessors? While LLMs excel at generating coherent and contextually relevant text by predicting the next word or token based on the input, their inference abilities are inherently limited to one-step and relatively simple pseudo-reasoning based on linguistic patterns. In addition, they have a limited ability to apply external knowledge beyond the text itself. Someone would say that this is a prediction based on statistical models.

LAMs represent a significant step forward in Artificial Intelligence built on the foundation of LLM. They improve these language models to become autonomous agents capable of understanding language, generating, complex reasoning, and taking autonomous actions. They represent a significant step towards General Artificial Intelligence (GenAI). This remarkable ability is made possible by breakthrough advances in neuro-symbolic programming. This enables autonomous agents to understand and interpret, for example, complex user interfaces of software systems.

The implications and applications are extensive and far-reaching

They can automate workflows across all types of software systems or use simple voice commands to automate complex, multi-step tasks. They can operate the user interface of a computer just like humans. By understanding buttons and text on web pages or apps, and performing actions based on our voice instructions, we can essentially ask the model to operate our computer for us, ushering in a new era of voice-controlled operating systems and a wave of screenless devices that will revolutionize the way we interact with computers.

Many business applications are now human-operated, requiring extensive training. Organizations can build an AI layer on top of their legacy IT systems, without replacing those old systems with new ones.

Potential applications. Sky is the limit

Global retail companies can deploy LAMs-based autonomous agents to monitor inventory levels across multiple warehouses and stores.

Agents can use LAMs to analyze historical sales data, customer behavioral patterns, trends and seasonality, and other relevant factors to accurately predict demand and automatically place orders with suppliers, maintaining optimal inventory levels while reducing excess inventory. In cybersecurity and threat detection, financial institutions can deploy autonomous agents to continuously monitor network traffic, user activity, and system logs.

These agents can use machine learning techniques to identify patterns and behaviors that deviate from normal operations, detect vulnerabilities, and respond to cyber threats in real time. They can detect suspicious activity, such as unauthorized access attempts. Organizations can leverage LAMs-based autonomous agents to provide intelligent customer service and support through virtual assistance. These virtual assistants can understand customer queries, access account information, and deliver tailored solutions to escalate complex issues to human agents, increasing customer satisfaction and reducing the burden on human support teams.

We are only at the beginning of the road

While large action models are now focused on understanding user interfaces on digital devices, this is just the beginning.

The true potential is revealed when these models are implemented in robotic devices. Imagine a world where robots powered by LAMs can operate all the machines and interfaces we have designed for humans. Robots could operate machines in factories, control home appliances, navigate electronic devices, and much more, revolutionizing the way we live and work.

LAMs offer organizations an unprecedented opportunity to streamline processes, streamline decision-making, and improve customer experience through the use of advanced technologies such as machine learning, natural language processing, and predictive analytics.

But the best is yet to come

The emergence of Large Action Models (LAM), a new frontier in the development of artificial intelligence, represents a breakthrough in autonomous AI agents capable of performing complex tasks. Unlike Large Language Models (LLMs), which focus on text generation, LAMs can predict and execute next steps in complex processes, which enables them to support a variety of user interfaces. They’re trained on big data about human interactions, so they can learn how to use different applications and adapt to user preferences. Potential applications include workflow automation, advanced inventory and safety monitoring, and intelligent customer service, and their future developments could revolutionize the operation of machinery and equipment.

Read next

  1. Data centers in orbit. The physics is starting to work. The spreadsheet isn’t, yet

    Space data centers are no longer a conference slide. We already have the first satellites, the first billion dollar valuations and the first delays. The delays tell us the most about this market. For the last two hundred years, industry has gone where the energy

  2. AI Was Supposed to Speed Up Developers. It Turns Out It Slows Them Down. Why Is No One Talking About This? 

    For the past two years, the tech industry has been buzzing with the promise of a big change. Artificial intelligence was expected to reduce software development cycles from weeks to days, or even hours. AI assistants promised to generate code faster than any human could.

  3. Robodebt: What are you going to do when an algorithm wipes your account clean?

    AI-based welfare systems are not flawless. While tech companies announce their cheerful visions of social change, the world’s poorest are being harmed by software errors leading to robodebt. I have previously written at length about the merits of artificial intelligence in business, environmental protection and