The future of AI: what’s the next step?
After tremendous success of generative AI, the World is waiting for the next big leap.
Another round in the development of AI brought all humanity into ecstasy. We have never been so inspired by the capabilities of new technologies, as with the arrival of the generative AI. From experience of communication with it, we get an incomparable influx of emotions and ideas, sometimes catching ourself thinking that far from every person we are able to get such mutual understanding as from communication with AI. How quickly did we come to this point and where will it lead to us? But first, let’s know more our companion.
According to the accepted periodization, we are now only one (OMG!) stage from Artificial General Intelligence (AGI), when artificial forms of consciousness will be able to self-learn and perform tasks subject to a person.
Today, AI algorithms easily resemble human intelligence in some tasks, they have their own logic and reasoning ability. They are trained to perform individual tasks, but so far their accumulated knowledge has not been transferred to a new task. Now let’s try to look into the near future and see what AI will be like in its next version.
We live in an age of great AI learning. Humans have become its natural guides to our world, as we show a child how our world works, we teach to distinguish a tiger from a dog, we feed it with all the knowledge and information that we have accumulated during the existence of mankind. Graduates of this school become successful startups and solutions that are ready to help their skin-creators in thier work. One of the main graduates was ChatGPT, which made a real revolution and pushed the boundaries of the technological possibilities of understanding the world. Based on it, a lot of projects have appeared, sharpened for a narrow task. And the pace of development is so fast that the future is not far off, where neural networks will communicate with each other.
AI already works on equal terms in a team, complements or replaces human labor. Until the moment when AI will do all the work in the world for a person, it is still very far away, but our Uncle Sem is already talking about this. The fourth industrial revolution gave society unprecedented tools for the development of a creative economy and production automation. With their help, we learned to disassemble the processes of creating value literally down to the atom. We managed to fully automate some of the processes and reduce the role of a person to the role of an operator.
As Yuval Noah Harari said: “We all have been hacked by AI”
However, in many areas we are moving so slowly that progress can hardly be seen. Somewhere authorization and approval of a person is needed, but more often technical progress itself, or rather the existing architecture, does not allow the implementation of new solutions. We still have a lot of work to do. With the advent of AI, we can not only show how to do this or that process, but also receive feedback in the course of the task and look for new, more profitable options for achieving the goal. A big impetus to this was digitalization, which made it possible to make a digital copy of each action, that is, we sort of translated what we were doing into the computer language and asked for help. But still, a person is (so far) more important than a machine: we still leave the most responsible tasks behind us.
Of course, there are examples of full automation. For example, autopilots. Several factors played a role here. First and foremost is the size of the industry. Big money is at stake. One working solution can replace a huge amount of labor and bring a lot of profit to its creators. Secondly, this is a rather visual image that has formed in our culture. Investors and society will simply not accept any half-measures with an operator on board.
Yes, there are dozens if not hundreds more examples, but let’s get back from the exception to the general cases. Why is the issue of full automation of processes and its implementation by AI still not resolved? Where is the voice from the refrigerator that will tell us that the milk is running low and the milkman is on his way? If this future is so clear to us, repeating itself in every second sky-fimovie, what stops us from implementing it?
As always, we will not hear a simple answer. It’s complicated.
First, there is no legal framework for regulating such systems. Who will be responsible in case of violations: operator, programmer or founder? Of course, the development companies themselves should push for its creation. By creating a precedent for using their solutions together with other counterparties, companies project possible boundaries of interaction and rules of the game that are beneficial to everyone.
Secondly, public mistrust. The user is very careful with his personal data, and even more so with access to money, digital and intangible assets, such as online or offline reputations with colleagues. Because of this, it’s difficult to imagine that a person will outsource communication with colleagues, money transactions and other intimate processes to a neural network.
Another important factor is AI cybersecurity. There is a reasonable fear that hackers will gain access to the user’s assets. Here you can solve the problem of multi-level authentication and splitting the system into separate modules, access to which will not be end-to-end and will be controlled by the user. Such a decentralized structure with separate access to different nodes has already become a new standard in the world and is actively replacing the old model with a single point of access to the entire network.
But the main reason lies in the evolution of IT development. Computers, first of all, developed due to the focus on the person as his personal tool. Accordingly, user interfaces were created for the convenience of human work. But the time has come for machines, we are frantically looking for what other processes can be automated and shift our work to AI. However, the problem is that machines do not understand how business processes and user interfaces work. Companies made their CJM and UX/UI based specifically on human needs: bright icons, large buttons, and so on. And pardon me, but how you wants AI learn these processes?
From this situation, there are two possible exits for companies. Either make your own AI-neural network and teach all the business processes of the company yourself, or wait for the appearance of tools for specific tasks (CRM, RPA, etc.). Large brands have already taken the first path: Samsung, Apple, JP Morgan and others have already announced their work on their own LLMs based on company data. It is important to understand that they were also pushed to such a decision by the high mistrust and fear of data leakage when implementing solutions from technology giants like Microsoft and Google. But there are also a large number of small companies and private entrepreneurs who will not be able to create their own solution. The only thing left for them is to teach the AI to understand it’s human flow. This is a huge job and we are in for an interesting time of adapting artificial systems to the real world.
Therefore, we see the next big step in the development of AI in the development of multi-embodiment tools that will allow users to connect neural networks with technologies from the real world to create full-fledged autonomous systems. This progress will bring closer the dream of artificial general intelligence (AGI) and its ability to autonomously build models to control a wide variety of agents in the virtual or physical world.
Like any super popular technology, AI will develop in all directions at once and it is impossible to single out one thing. One thing is clear: the amount of man-hours that will be spent on the study and development of this technology will be comparable to the greatest construction projects of mankind. Such fruitful creativity will definitely bear fruit and make our world a little better and fairer.
“The best investment is an investment in tools of our own production”
Benjamin Franklin
