The recent surge in interest surrounding generative AI is primarily due to the explosive adoption of large language models (LLMs), such as GPT-4. These models have become a hotly debated topic among experts; some view them as the next paradigm shift in artificial intelligence (AI), while others focus on their downsides and limitations.
Despite being an intriguing phenomenon when studied individually, the true value that enterprises stand to gain from these models will only materialise once they are integrated into larger systems. With this in mind, we will explore various ways these models can support different levels of enterprise automation. We’ll begin with simpler methods and progress towards more advanced cases, explaining the main drivers behind each level of automation.
Connecting large language models to internal data
The first step towards automation with large language models is connecting them to internal data, usually achieved through an information retrieval approach based on semantic search. This process maps a user’s query into a semantic space where text segments with similar meanings are closer together. After mapping the query and running a semantic search, the relevant information is sent to the prompt of the language model and incorporated into its context when generating answers. Although this method doesn’t completely solve hallucination issues, it significantly improves the reliability of the content generated by LLMs.
While this design pattern brings order to unstructured textual data within enterprises, much more data exists in structured databases or other structures, such as knowledge graphs. With vast amounts of textual data also come signal-to-noise problems that cannot be overcome with just semantic search. Therefore, it’s crucial to not just rely on this simple design pattern but to also include approaches such as keyword filtering or queries to structured databases to obtain data to be included in the prompt.
This system, at its first level of automation, is principally driven by the proactive engagement of the user in their quest for information. In other words, the system relies heavily on the users’ initiative, who typically seek out information through a user interface, such as a chatbot.
The benefits include easier access to accurate information while saving time and effort – sometimes even unlocking previously inaccessible sources of information – which can lead to better coordination across an enterprise and more rapid updates in decision-making processes.
A problem commonly encountered by companies at this stage of automation is the familiarisation and productisation of the technology. Enterprises often struggle with understanding how to effectively integrate these models into their systems and how to create a user experience that encourages usage across the organisation.
The next frontier in enterprise automation is marked by the advent of ambient intelligence. This involves the creation of workspaces where technology doesn’t just assist but becomes an intrinsic part of the environment – responsive, adaptive and consistently working in harmony with human tasks and processes.
The key to this evolution are multimodal models: AI technologies capable of processing information from a multitude of sources, from spoken words and visual content to written documents. These models operate quietly in the background, processing and storing data from an array of inputs, thereby enriching the contextual understanding and responsiveness of the ambient intelligence environment.
The transformative potential of ambient intelligence is rooted in the dual role AI systems undertake as both archivists and curators. As archivists, AI models capture and digitally encode a myriad of information sources. Everything from meetings to brainstorming sessions and even personal notes are stored, ensuring that every piece of information is preserved, however trivial it may seem.
Conversely, as curators, these systems filter through the extensive data repository, eliminating the noise and highlighting pertinent information. Analysing user interactions and preferences, they offer context-aware recommendations and guide individuals toward useful resources. This intelligent curation revolutionises how we interact with our digital workspaces and enhances our capacity for informed decision-making.
The practical applications of ambient intelligence are as diverse as they are innovative. Consider the automation of taking meeting notes: the AI system transcribes speech in real time, ensuring all critical discussions and action items are accurately captured. This streamlines workflows and significantly boosts the efficiency of meetings. Similarly, applications may range from analysing and understanding screen captures, to converting handwritten notes to digital text, to creating personalised news digests.
Perhaps one of the most exciting aspects of ambient intelligence is the potential for creating personal knowledge bases. By continuously capturing and curating data, AI systems can ensure that valuable tacit knowledge – the unspoken and unwritten information individuals possess – is not lost. This results in a personalised and dynamic knowledge repository that evolves with the user, offering insights and information specifically pertinent to their needs and context.
However, the promising landscape of ambient intelligence also has its challenges, chief among them is trust. As AI systems are entrusted with a plethora of sensitive data, it’s critical to have robust data security and privacy measures in place. Users need the assurance that their information is secure and won’t be misused. Only with this trust can ambient intelligence fully realise its transformative potential.
In part 2, we will continue our exploration of enterprise automation with a look at intelligent agents and embodied intelligence.
Regional Head of Data, AdriaticBoris is a machine learning expert with over 15 years of experience across various industries. As Regional Head of Data, he is dedicated to growing our data capabilities in the Adriatic region. Throughout his career, Boris has co-founded and led a data science consultancy, established an AI department in an IT services company, and advised governments on AI strategy and standards. He is passionate about innovation and has a good track record of spotting key technology trends early on. Boris also enjoys sharing his insights through speaking engagements, advising start-ups, and mentoring. Recently, his main areas of interest include large language models and other generative AI technologies.
13 November 2023
Delving Deeper Into Generative AI: Unlocking Benefits and Opportunities
07 November 2023
Retrieval Augmented Generation: Combining LLMs, Task-chaining and Vector Databases
19 September 2023
The Rise of Vector Databases
27 July 2023
Large Language Models Automating the Enterprise – Part 2
01 June 2023
Challenges for Adopting AI Systems in Software Development