It will help you to perform various actions
After the popularity of “ChatGPT”, many people have become accustomed to asking for advice not only on Google search, but also on various large language models (eng. Large Language Models, LLM) based chatbots. However, they have one big limitation – they can give advice, but in most cases you will have to do it yourself. The developers of AI agents are preparing to change this, promising that their artificial intelligence will be able to do more than just chatter.
“In the world of technology, AI agents are one of the hottest topics of the last six months. If I can now ask DI to plan a three-day trip in one or another city according to my wishes, then the DI agent could choose the flight tickets and order them himself. We can already see this in early beta versions of AI agents. You make a request, lift your hands from the keyboard and watch on the screen how programs are opened and closed on the computer and orders are made,” says L. Keraitis.
According to the expert, at the moment it is possible to observe only the beginnings of similar automations in customer service, data analysis or other areas, as well as in programming. However, there is still no universal AI agent capable of performing ordinary human tasks.
“At the same time, evolution and a revolution are expected in this field. The evolution can be observed in the text tools. Gmail already offers to finish your sentences for you. I tried another app from a small startup that writes for you in any computer program. It works, but definitely not good enough for everyday use. There are ChatGPT apps (eng. GPT’s) capable of generating presentation slides. They really do it primitively, but when asked to correct them according to the notes, they retort – correct yourself. So there is still room for improvement”, L. Keraitis observes.
It would also work on phones
According to the expert, a couple of years ago, most AI tools worked in one format – text, images or audio. Recently, we have seen more and more multimodality – the ability of the same tool to edit text, generate an image, and understand voice. This is very important when designing AI agents.
Another upcoming challenge is to allow AI agents to understand different apps on phones or computers, to control them independently. “Ideally, a phone with an AI agent would be able to independently “walk” between my apps, understand that I want to do calculations on a calculator, and put plans in a calendar,” assures L. Keraitis.
Most of today’s chatbots operate in the cloud – they need to connect to their servers online in order to find the best answer. In addition, generating the answers requires huge resources of computing power. Therefore, efforts are being made to fit these models into smaller and smaller devices, even personal computers or phones. AI agents operating on this principle could more efficiently employ the user’s personal data – analyzing it inside the phone poses fewer security challenges.
AI agents are being developed, but we will have to wait
While Siri or another assistant can call a taxi or find a book to buy, an AI agent could compare several online bookstores, find the best option, buy the book, find the nearest post office and direct the order to it.
According to L. Keraitis, such a vision of the future raises questions about online security and responsibility – what will happen if an AI agent “makes money” on behalf of the user? For example, if a person downloads a malicious agent and attaches a payment card to it? After all, is it likely that everyone will have their own AI doppelgängers to arrange meetings with each other, send and read letters, maybe even represent people on dates or in bureaucratic matters?
“Given that the technological implementation of AI agents is becoming easier to imagine, and technology companies understand the advantage they would gain by being the first to offer such a product, we can only wait for the results of this competition,” says L. Keraitis.
