The AI is still underestimated

More and more AI features will be available in software, but if a company really wants to use them to its advantage it can't wait around forever. Connecting AI solutions and services to corporate systems requires special knowledge, therefore it is worth to assign an expert partner with its implementation, says János Horváth Varga, Head of the Customer Relationship and AI Solutions Competence Centre at Magyar Telekom.

◼︎ The big opportunity is in connecting several models

– Thanks to ChatGPT artificial intelligence is now flowing not only from the media but also from the tap. When there is so much hype, one always suspects that it is a bit overrated, isn't it?

– Many experts share the same doubts. You think, well, well, we now understand and generate not just text but also images, sound and video, but what does that mean in terms of the bigger picture? I don't think so, I think that AI is still under-appreciated and we are much further away than most people think. It is no coincidence that more and more people are raising their voices about the need to stop for a moment.

It is certainly not only professionals who are interested in the subject. I see that almost all of our large enterprise customers are monitoring the topic, learning about the technology, looking for their place in corporate life.

– ChatGPT has really shown hundreds of millions of people what artificial intelligence is all about. But how does this become a technology that actually makes a positive difference to everyday work?

– The next big milestone will be when AI capabilities are built into office software. One example is Microsoft 365 Copilot, an AI assistant that is integrated into the Office software. These are now based on GPT-4 and will bring the benefits of MI to an even wider audience. What is it about? I don't have to write long texts - I just summarize the main ideas in a few points, the software writes it for me, I just have to review it. Or I can enter the text, describe the images I want, and Copilot will create the presentation. After a Teams meeting you don't need to write a memo, because the software creates the recorded audio, sends the tasks discussed to the people involved and can even create calendar entries.

– Tools to improve efficiency, like auto-complete, suggested answers, things like that, are already in the software, but in my experience they are not really used. Why will this be different in the case of AI-supported features?

– Generative artificial intelligence is on a different level than assistive technologies that offer a few pre-made sentences as possible answers. There will definitely be a rollout period, when people are still nervous about it, trying to see what it's good for, how good it is. Let’s the the example of a software that transcribes live speech. They have been around for years, but they were discouraged because it took more time to correct the text on the machine than to type it yourself. Now the accuracy has reached the point where it is not a question whether we should use them.

I expect the same for devices like Copilot. There will certainly be some annoying bugs at first, but quality improves at an exponential rate. ChatGPT has been available to the public for less than five months and is already visibly better; Copilot is building on the next version and OpenAI is already working on GPT-5. I think the improvement will be lightning fast, and the breakthrough will happen in a matter of seconds. This will be the first big leap. Users will also have a tool for cognitive and creative work that will be a huge help in almost all areas of intellectual work.

– What could be the next big leap?

– Artificial intelligence, or at least machine learning, has been a feature of many companies for some time. Image recognition is used in manufacturing to detect defective products, in the financial sector for credit rating or in stock trading. It is another question to what extent these systems have been integrated into company processes - there is certainly still plenty of potential.

Generative AI, which is now coming to the forefront, is also very exciting because it allows us to put the human-machine interface on a new basis. It's a completely different quality when I can give commands and tasks to the machine in natural language, either orally or in writing. For example, I don't create SQL commands, but describe what data I want to analyze, what I want to know, and the system runs the search in the background and gives me the answer in the format I provide, either in text or graphs.

But I also see huge potential in connecting and collaborating different AI models. The language model interprets the human command, starts the robotic process, and the language model interprets the result back to me.

– What opportunities does the average Hungarian company have to equip its systems and processes with artificial intelligence capabilities?

– On the one hand, it will be able to rely on big manufacturers, as they will all eventually incorporate AI into their products. Smaller companies can already use a wide range of standardized functionality from the cloud. The big question is what happens to software developed by small developer companies and in-house developed applications? They will also need to be educated, because users will get used to such features and will look for them in all software. However, before we jump into AI integration, it is worth to look around, plan the process and define the business benefits we expect to gain from its implementation. New skills will be needed in software development and business operations. We need to learn how developers call the APIs of major providers, how to use them and what to do with them. It will require a different mindset than classical software development. The key will be whether the developer can ask the right questions of the models, process the answers and integrate them into business processes.

– What about companies that have little functionality available in the cloud?

– For large companies that do not want to give access to their  confidential data to international service providers, we can also offer small language models running in their own environment, together with the necessary system integration services. It is very important that the language model, whether it is GPT-4 or the smaller Hungarian model, provides the toolkit and processes the texts. The models are getting more accurate, but all they do is understand what we tell them and generate a good quality output. However, the value of the company, the knowledge that everyone has to put into it, is between the two. We cannot expect this from OpenAI, as they are not familiar with our corporate operations. This requires system integration skills that can link the language model to enterprise systems and data - skills that will be extremely valuable in the near future.