We live in an era where robotics is taking over many jobs, or, more simply, robotics automation is advancing rapidly. The rate is faster than any construction company could even imagine. What are we constructing? Manufacturing? Or producing? Robots? If so, what are the features of the robots? What kind of features would you want in your work?
Enough of generic robots with AGI! Now it’s time to certify robots and specialize a model. It can be understood as follows. Suppose you want to automate coding work in your organisation for beginner-level. Then you need to use AI for it; choose one of the major AI vendors now. But you code in C++, the vendor does not provide specialization on what your coding language is, it just says we can code any code, we are good at automation. Wait, there are major differences in programming language constructs even within a programming language, for example, in Linux-based C++, we have STL and pointers, Visual C++ has MFC classes for GUI, while Linux-based C++ is altogether different. And there are Java-based codes that don’t use pointers explicitly, while C++ codes depend on pointers and dynamic allocation directly, unlike the internal implementation in other languages. So, this is what we mean by specialization. Do not confuse AGI; specialize in your own company. The problem is learning new concepts post-training.
The best model for AI or robots is to specialize in AI within your own company’s code. With your own comments and applications. This is for coding in C++ or Java. What about other areas, such as web services? Well, one AI model doing it all, can it mess with coding constructs? No! or yes? So why not certified robots or certified AI models? At least they can’t go up the ladder? Can they? Maybe? But since robots mimic humans not just in looks but also in life, humans gain expertise and specialization in professions. So, is it time for robots to mimic humans in specialization? Benefits are yet to be seen; the impact is yet to be felt.
What would certification achieve? With certification AI robot or AI model would get to work on its specialization. Each person in the world has a specialization. Some are experts in civil engineering, some in manufacturing, and some are experts in coding. So why not a certification tag on your AI model or your AI robot?
If one model fits all, that’s good, but why not specialize with certifications on the models to build more trust in AI and improve reliability? How can that happen when AI is being trained on a chunk of all information? AI is a master of all, but can specialize in all areas as is demanded, this is by the nature of AI. The problem of forgetting in neural learning needs to be looked at, as LLMs are trained on chunks of data. Training from scratch on your own personal model for your company?
So, how do we specialize? More training data being fed into it, or a new learning AI? Or we have to accept that AI will always be the master of all. What happens with specialization? What are the benefits of specialization? Say a faculty member wants to teach robotics components at a university, and her specialization is operating systems. It would take her time to build expertise in robotics, whereas she is already an expert in operating systems. So, this is the same as robotics, but robots are masters of all. What mess can they make in teaching robotics? None. But imagine it as a daily routine of teaching robotics, then being a specialized person in a field matters.
Specialization matters for humans? What do you think about robots that are specialized or robots that are not specialized doing the work of a specialized person? Imagine a person with experience in networks with C++ and a robot with artificial general intelligence? Who would be the master, a human with experience in networking with C++, for the job of a network engineer? So, what about specialized intelligence? Is it not the time for specialized intelligence? How can specialized intelligence be achieved? Well, let’s take an example of coding only: a new LLM can be trained by, say, a gaming company that writes code in C++. The gaming company needs to train its code and English language constructs from scratch. The clear case logs, the check-in comments, the null pointer allocation bug that comes and is logged in, trained right on your own APIs? LLMs can find post training learning forgetting, but what about nascent specialized training? Would it work? We need specialized artificial intelligence now, more than before.
When we go to school. We are taught the basics first, then English, then science. We are given specialization much later in high school, when we are told to specialize in science, engineering, commerce, or the arts. So we have not lost way, we are on right track, this is how humans found their way to specializations. In the same way, we have reached pinnacles and the base of artificial general intelligence has been reached, now is the time for specialized artificial intelligence.
Would it be beneficial? Only time can tell!
Thank you for reading,
Regards,
….
Artificial Intelligence, Computer Science & Mathematics 