Even though artificial intelligence entered our lives quite recently, it has already managed to settle in for good, making impact on both private and business routines of ours. We have become recipients of the content AI generates, but at the same time also initiators, or perhaps even architects, of its creations. Whether we’re looking for information, advice or perhaps we just want to create a fun picture, online chats based on large language models (LLMs) work perfectly for those purposes. No wonder then, that we also use artificial intelligence in our professional lives to make our work more efficient. But is this really a good solution?
Popular AI chatbots operate based on data available on the Internet. They easily answer questions and solve tasks related to general knowledge. However, when it comes to companies or other organisations, efficiency of artificial intelligence tools requires data and documents that in most cases should not be found online. This is why many employers do not allow their employees to use popular AI chats to assist with their professional duties. Is it, however, an effective way do deal with this situation?
In practice, imposing bans on external tools doesn’t work, as they only shift the problem to the so-called shadow AI: employees use private accounts on private devices, and the organization loses any control over the data flow – addresses the question Maciej Łabędzki, the head of PCSS AI team working at the Data Engineering and Analytical Platforms Department of the Poznań Supercomputing and Networking Center. So, how can that problem be solved? Our expert replies – Attempts to block the access to artificial intelligence is not the answer in such cases. The solution is providing an internal, secure AI assistant – a business implementation that is guaranteed by PCSS artificial intelligence specialists.
We do not make another chatbot. What we create is a trustworthy AI agent which knows documents, procedures and needs of a given organization and operates under this organization’s full supervision – tells us Marcin Wolski, AI projects manager at the Data Engineering and Analytical Platforms Department of PCSS. So, we are talking about safety or perhaps even security of a company here. As one of the the biggest risk of using common LLM-based tools in business is the possibility of uncontrolled disclosure of confidential information. Pressure to complete assigned tasks as soon as possible may easily cloud employees’ minds; as a result they don’t necessarily always consider the risk of sending personal data, business information, internal documents, or other content covered by trade secrecy to an external tool. Once again we give the floor to Maciej Łabędzki – Any leak of data is irreversible and usually invisible: if an employee is using a private account, the organization has no logs and doesn’t even know that anything like that has happened.
The solution proposed by PCSS eliminates such risk, because it ensures the flow of information within a closed, controlled environment, which means that no data leaves the ecosystem of the organization it belongs to. That is possible thanks to the use of local AI models. In this case the access to information can be restricted by granting users appropriate permissions, while the administrator has complete control over both the sources and the use of internal knowledge. To sum it up, creating a personal AI assistant or agent – as it’s commonly called – a company can take full advantage of artificial intelligence capabilities while retaining total control over its data.
A crucial safety-related aspect of artificial intelligence is the very server on which an AI agent runs. In PCSS’ case, the servers are placed in two data centers located in the capital of Wielkopolska – namely the city of Poznań. The company knows then where the information and documents it shares actually go. Moreover, its management can be certain that the data in question won’t leave the tight, secure ecosystem that meets the strict requirements of the AI Act. The Poznań Center provides also optimal conditions for artificial intelligence projects, which require enormous computing power. PCSS offers a unique and mighty combination of both HPC infrastructure and quantum computers, capable of handling such demands without much effort. And should anyone still have doubts, let the ultimate reference be the fact that PCSS is also the home of PIAST AI — Poland’s first artificial intelligence factory.
So, let’s now go through some basics that this service include. It all starts with an assessment to establish all the tasks the AI agent will handle within the company. The next step is data analysis, since that is the foundation of the agent’s work. Next comes the selection and implementation of language models, followed by integration with documents and databases, along with testing the system as a whole. Finally, the agent is trained, so it can fully meet the needs of the organization it is going to work for.
The longest, yet most important part of this work is designing the knowledge base: choosing sources, deciding which documents are valid and which are outdated, organizing them, and assigning access permissions – explains Maciej Łabędzki – During the implementation, it usually comes out that some of the organization’s knowledge isn’t documented anywhere and exists only in the heads of some employees. Our AI agent reveals those gaps very quickly, as they are reflected in questions it is unable to answer.
PCSS also provides smoothly functioning connections to document workflow, helpdesk, HR department, the system of searching for information within given source, data access management, training of users, as well as software maintenance, security monitoring, and technical support. Going further, individually tailored solutions can be developed to match a specific, more complex needs of an organisation.
Popular online AI tools can serve as fairly capable and even efficient universal assistants for personal use – but a PCSS agent becomes something different: a trusted specialist operating within an organisation. It may help a civil servant to find the right administrative procedure, support a company employee in locating an instruction manual or document template, or automate any helpdesk’s responses to recurring questions. Interestingly, in further stages of development, the solution could also flag various issues to auxiliary systems or support the execution of various processes within a company. It’s all about: Your model, your data, and your complete control.
Cover illustration was prepared using artificial intelligence tools.