- How is multimodal AI revolutionizing how machines understand and interact with us?
- Are AI agents set to take over complex service roles once dominated by humans?
- Can synthetic data be the game-changer for faster and cheaper AI development?
- Is your workforce ready to embrace the creative opportunities brought by AI?
- Could AI tools be the key to unlocking major savings and efficiency in project management?
Interview with Marcel Kasprzak, Managing Director of NeuroSYS
In our conversation with NeuroSYS, Marcel discusses AI trends 2025 from Gartner’s insights authored by Pieter J. den Hamer, Vice President in Gartner Research. Key topics include AI Agents, Multimodal AI, Simulation & Synthetic Data, AI project management, AI literacy, and regulations. Marcel shares his observations, examples from our projects, and predictions about where AI is headed.
NeuroSYS: Multimodal AI is probably the biggest trend, especially considering the recent ChatGPT 4o model that has gained multimodal capabilities. What is special about multimodal AI?
Marcel Kasprzak: In Multimodal AI, the most important thing is to combine different types of data, which expands the range of communication methods between humans and AI. Multimodal AI uses data from different sources, not only text, but also images, sound and video. Thanks to this, it better understands the context of the query. The shift towards multimodality aims to get closer to human-like communication, where the context of the message is influenced not only by its content, but also by facial expressions, voice and gestures. Multimodal AI therefore tries to imitate this multi-aspect nature of human perception and understanding.
The effect of the multimodality of artificial intelligence is to generate more precise and therefore more reliable results. For example, a speech recognition system will do its job better if it uses video analysis from which it reads lip movements. By combining different types of data, multimodal AI can provide more comprehensive analyses and recommendations. For example, by providing AI with an X-ray image and test results in text form for analysis, we can receive a better, i.e. more accurate diagnosis from the tool.
AI agents, also known as AI assistants, are another important trend mentioned by the authors of Gartner’s insights. Who are AI agents?
AI agents are simply software that performs programmed tasks, such as interacting with the environment, making decisions, and automating various processes.
Meta’s launch of an AI voice assistant service, which closely mimics human interaction, hints at a potential shift from human consultants to AI agents. However, despite this technological advancement, many customers still prefer speaking with a live representative when contacting support. So, what is the actual trend?
AI agents are set to replace humans in some service sectors, as we can already see with the widespread use of bots in customer service, call centers, and technical support. Previously, Robotic Process Automation (RPA) allowed machines to perform simple, repetitive tasks independently, like issuing similar invoices. However, RPA falls short when tasks require human interaction, and advanced AI solutions are now stepping in to fill that gap.
To illustrate the difference between RPA and AI agents, consider the analogy of cars. RPA is like a smart car that alerts you to nearby vehicles and provides speed limit warnings, but ultimately leaves the decision-making to you. In contrast, AI agents are akin to autonomous cars that drive themselves, managing everything on your behalf. The key distinction is that while RPA handles repetitive tasks, AI agents go further by taking control and managing situations that previously required human intervention. They combine the routine efficiency of RPA with the ability to handle complex, unpredictable scenarios.
Read also: The benefits of implementing large language models in customer service
Among the most important AI trends for 2025, we find data simulation and synthesis. What is its application potential?
Simulation and synthetic data is intended to solve many data challenges in AI. To create an effectively functioning AI model, you need to collect huge amounts of data for it. However, this is expensive and time-consuming. Therefore, everyone wants to speed up and financially optimize this stage. One of the ideas is to simulate and synthesize data, which is much better than operating on “real” but limited resources. There is no point in building models on poor quality data, because it can make them completely ineffective.
How do data simulation and synthesis differ in the context of creating AI models?
Simulation involves generating artificial data, while synthesis utilizes existing data. This approach refers to ready-made solutions that can be applied to specific problems. AI engineers often purchase pre-existing data sets, enabling them to shorten the time and reduce the costs associated with acquiring data for AI projects. The focus is on identifying patterns that facilitate data acquisition in various, more efficient ways. Consequently, we will increasingly see AI engineers simultaneously using real, generated data alongside AI-ready data collected from diverse sources.
Do you see the potential in using artificially generated data in most AI projects?
Yes, artificial data can be generated in most projects. In the near future, people involved in AI will therefore decide how they will obtain the data on which they will train models, what quality this data will be, what they will do with this data and possibly the adoption of data simulation, i.e. generating it or obtaining it from outside.
AI engineering and AI project management is another burning issue that arises in the context of the development practices of this technology. How to properly manage AI projects?
I recommend integrating artificial intelligence projects within the same management framework as traditional IT projects, rather than treating AI as a separate category. AI initiatives should follow the same protocols for risk control, quality assurance, budgeting, and task management. This alignment is particularly crucial because, at some point, the AI model must be deployed and integrated with existing software systems. Even during a proof of concept (PoC), data collection efforts often lay the groundwork for future model development. Therefore, the entire process of data collection and AI model design should adhere to the same rigorous development procedures as any IT project, including quality control, code review, validation, data integration, testing, and ultimately, activation and deployment.
Read also: AI Workshop For Business Advantages
Do you think AI tools will improve project analysis and draw more effective conclusions from successful and unsuccessful initiatives?
Lewis Platt, former CEO of Hewlett-Packard, once said, “If HP knew what HP knows, we would be three times more productive.” I think AI will eventually be used to transcribe, transcribe, and codify all the knowledge from internal team meetings. So that something can be learned from it. Using AI will allow you to create a powerful tool for, for example, project analysis.
Let’s say your company has many projects and all the project-related meetings were recorded. AI-powered tools can provide deep analysis of meetings, helping to identify factors that contributed to the success of projects as well as those that led to failures. Additionally, these tools could potentially offer real-time alerts during meetings if discussions begin to veer off course. However, concerns about privacy and the potential impact on human behavior arise, as individuals may alter their behavior knowing their every word is recorded and could be scrutinized later. This raises the important question of how to leverage such tools without stifling human creativity.
Now let’s discuss the next AI trend for 2025: AI regulation. Meta announces that it will transfer the development processes of its AI models to the US and threatens not to make them available in Europe for fear of EU regulations. Won’t this hamper the development of AI?
Large tech companies have repeatedly threatened to exit Europe in response to regulatory pressures. Will this also apply to newly developed AI solutions? It’s uncertain, but one must question whether these companies can afford to partially or fully boycott a market as significant as the EU, which encompasses nearly 500 million people.
I don’t believe this will have a decisive impact on the European technology market. American companies typically introduce their technological solutions to Europe with a delay compared to what is available domestically. The reality is that the global technology leaders are the U.S., followed closely by China, leaving Europe in a position of catching up. Paradoxically, such a boycott could actually benefit Europe, as it would compel us to intensify our efforts and focus on developing our own homegrown solutions.
Which AI regulations do you think are the priority?
Definitely those related to protecting our privacy, because we don’t really want to operate on models trained from a Chinese perspective of assessing reality, right? In addition, I would see the need for greater transparency when it comes to algorithms using AI to assess our health status, which then calculates health insurance premiums. This could lead to socially unfair situations. In the future, there may be a situation where AI will be used to analyze research results in such a way that it can predict the probability of contracting a disease or even calculate average life expectancy and use this to calculate the amount of health insurance premiums.
And what about “safeguards” protecting data? Can we equip AI systems with them?
We are already developing such systems at NeuroSYS, and they operate in an on-premises model. In on-premises AI systems, data is stored on internal computers and servers, not in an external cloud. This allows companies to have better control over them and better protect confidential information. Such systems are chosen especially by companies that must comply with restrictive confidentiality laws and industry regulations, especially companies from the financial, pharmaceutical and healthcare sectors.
In addition, in on-premises AI systems, companies can better regulate access to data and control how it is used. This is possible thanks to better security measures, such as advanced encryption and secure physical storage. This ensures that AI compliance is maintained at the highest levels.
Let’s address AI trends 2025 and their impact on the global labor market. With the recurring concern of AI potentially replacing jobs, is this a real threat?
Some of the current mass layoffs are due to the crisis, and the crisis forces improvements, of which artificial intelligence is an element. AI is therefore an effect rather than a cause. The process itself cannot be stopped, although the issue of AI’s impact on the labor market is broad.
I would add that there is indeed a risk of simpler jobs being replaced by AI algorithms. Many countries, such as Poland, have traditionally excelled in providing a workforce for repetitive tasks, like manual invoice entry or payroll processing. In this context, it is worth considering whether replacing these jobs with AI solutions might actually offer more benefits than drawbacks. By transitioning these workers to more creative roles—tasks that machines are not yet capable of performing—greater job satisfaction could be achieved, while also adding more value to the economy. A relevant example is the replacement of cashiers with self-checkout systems, illustrating just one of many possibilities.
One potential way to mitigate the effects of robotization could be the implementation of a guaranteed income. In the future, work might become a privilege, with countries imposing additional taxes on providers of robots and AI solutions. The revenue generated from these taxes could be used to support citizens whose jobs have been displaced by automation. If individuals wish to increase their income beyond this guaranteed support, they would need to reskill and enhance their capabilities to adapt to the evolving job market.
How do you assess the way organizations are preparing employees for AI trends 2025?
The AI revolution is driving major changes, so it is all about managing these changes. Many companies struggle to prepare employees for these new realities, but more are now recognizing the need to support their teams in adapting to automation. In response, some companies like NeuroSYS offer AI consulting services to help employers ready their workforce for these technological shifts.
I would also like to emphasize that the basis for managing this change, both in the case of enterprises and smaller companies, must be technical education, because employees must understand how automation affects their work. It is also necessary in this process to build employee awareness of the changes that AI brings. Change managers should also develop methods to eliminate unjustified fears related to the AI revolution, but also educate themselves and their subordinates about the impact of AI on the entire organization.
And what about office work? Which areas do you think are ready for automation and AI?
I think it’s natural that call center services will move towards automation and AI solutions. Maintaining a call center is very expensive for corporations, as we know from the example of one of our clients. After we implemented a learning management system (LMS) in this company, the client’s profits jumped to the level of a 6-figure figure. And this is basically because the LMS allowed the client to really control the training of subcontractors working in different places around the world.
Read also: Revolutionizing Customer Service with AI: Efficiency, Speed, and Satisfaction
How is this control carried out?
The client now has control over what their externally hired consultants learn, how they learn, and how well they absorb the information. Because the client is paying for this onboarding out of their own pocket, ineffective training can end up costing them a lot of money. Our solution allowed the client not to pay a subcontractor for training a consultant who failed the final implementation exam. This situation highlights that even introducing such simple control mechanisms or engaging bots in call centers generates huge savings for companies, even tens of millions of dollars per year.
Moreover, for the same client, we are currently working on a mechanism from the field of AI in education that will automatically generate tests and exams based on course material. This way, training trainers will not have to do it manually. Interestingly, we conducted the first test on a group of several tens of thousands of users and as many as 70% of them failed the exam generated by AI.
Maybe this exam was too hard?
I can turn this question around and ask if the original one wasn’t too easy. I will admit, however, that the exam generated by AI used binary grading. In multiple-choice questions, it was enough to skip one answer for the system to award 0 points for the entire question, which could have distorted the result. Nevertheless, there is a strong suspicion that in traditional exams, instructors lowered the level of difficulty so that as many people as possible would pass, because this meant money for the subcontractor. Therefore, exams or employee training generated by AI are also trending and may accelerate the process of replacing people with algorithms, because they can translate into huge financial savings for companies.
Ready to transform your business with AI? Take advantage of a free 1-hour consultation with NeuroSYS and discover how our experts can help you overcome data challenges and unlock high-impact AI opportunities:
