Curiosity is the source of all knowledge or so goes the saying. However, it is proving to be a useful sermon when it comes to AI as computer algorithms, when armed with curiosity can make AI much smarter than it already is. Recently, such a breakthrough was obtained by designing a model that could generate intrinsic curiosity in the algorithm so that even without a very strong feedback signal, it could work. Basically, AI will absorb as much as possible from the environment and learn all aspects of it and then try to know more and in depth about each aspect.
Such an invention could surely prove to be the decisive factor in machine learning as it can become much better. In reality, doing things out of curiosity leads to few rewards but a lot of knowledge. However, in case of AI, this knowledge is precisely what is required to make it smarter. Recently, there have been some breakthrough technologies surfacing from here and there. AIs can now tell which would be the best route to reach a goal by learning through reinforcements. With so much development of technology, it is now beyond simple coding and is all about creating modules to integrate within AI.
Bettering the previous
Reinforcement, however, can be cumbersome and tiresome as it depends on feedback a lot. You need to ensure that the system gets immediate feedback since reinforcement depends on it. However, researchers claim that the curiosity model works much better and it can detect behaviours and their subtlety in a more responsive manner. The learning process thus becomes more accommodative and AI can work more smoothly without getting bumped by dead ends. In fact, you need not tell the AI that it is doing it right or wrong. AI can figure that out by curiosity and enhance its skills much quicker.
This technology has been in the radar for some time now, only to be successfully implemented recently. In fact, neuroscientists are also finding it fascinating since AI is moving beyond inquisitive robots and showing near-human traits which can open up new possibilities in interdisciplinary fields in this regard. Currently, the research is being done to test how this new technology functions regarding unknown objects or awkward situations. This is precisely the innate nature of curiosity and human beings respond to it positively. However, AI’s response will determine if this curiosity model is going to revolutionize AI technology or not.
Learning to think like humans
Thinking and learning simultaneously is a complicated process that requires a lot of cognition. For AI to possess human-like cognitive capability, this is absolutely pivotal. Curiosity learning has pushed AI development one step towards that simultaneous possibility. Machines can now actually think and then apply according to situations without a feedback about what to do. However, this level of curiosity, it is safe to say, is much lesser than a child. Human inquisitiveness has no limits and can unearth much more knowledge than AI. Surely, that brings one to the question as to what this curiosity model can do and to what extent.
The answer is pretty simple. It will learn about environment and work faster to adapt to the surroundings. It is yet to produce a radically new knowledge system that the developers have no clue about. As of now, AI is only curious about what effects the actions will have on the environment and vice versa. It is only when environment as a whole will become an object of curiosity to AI can you then start talking about human-like AIs. Also, this method of knowing is a more indirect one since average human would want to know about objects directly instead of its own actions. So, there are plenty of differences still, but the innovation is exciting to say the least. Since it is at a nascent stage as of now, the scope for development is aplenty.