AI has reached a status of ascension today and shows all promises to be an intrinsic part of human lives in the coming years. Yet just like all mysteries in modern sciences, not all the problems faced today can be solved by AI. With so much going around in the world of AI, it is sometimes difficult to ascertain the fundamental challenges where AI principles have shown no signs of actual progress.
In this article, we will delve into some of the problems that AI is yet to comprehend completely. Although these do not represent all the problems, the intention is to highlight the ones that still remain unsolved.
1. Exhibiting Common Sense
One of the most prominent problems for AI is displaying common sense. It is very difficult to emulate this ability since there are too many factors at play, such as depth and credibility, to display common sense. The machine learning aspect of AI does not necessarily help to display this emotion in its entirety.
This can be attributed to the fact that common sense perceptions are individualistic and are established on the basis of these perceptions, which may span in millions and would keep growing. This is difficult for AI to emulate because at certain instances, they are not confined to the norms of human behaviour (individuals with mental illness) as well as the situation of ‘normal circumstances’ (philosophy, religion among other social factors). Another downside with common sense is it can be woefully wrong at times. For instance, judging political outcomes solely on common sense is baseless. All these criteria are too much for AI to handle, and even if it does, it is unlikely to make perfect sense.
2. Visual Aesthetics
There is a stellar difference between how human brain functions and how AI processes the same image when it comes to perceiving visual information. Although concepts such as Convolutional Neural Networks (CNN) are slightly based on the working of the human brain, other factors such as pre-processing and applying filters, are to be done manually if incorporated in an AI system.
On the other hand, the exact working of the brain regarding visual perception has not yet been fully understood, which makes it difficult to replicate in a perfect AI for vision-related projects. However, research has shown that visual perception is an active process that scours visual information constantly from our eyes (stimulus) and manipulates sensory signals accordingly to the stimulus, without any interruptions — even if there are various eye movements. This complex process of capturing visual information is just a tip of the iceberg when it becomes to brain activities surrounding vision. There are other factors such as subconscious actions among others, at play. AI systems and methods cannot just be developed based on theoretical knowledge and facts of the brain.
3. The Face Of Conscience
In 2017, Facebook had performed an AI experiment that ended up with a mysterious result. The AI bots developed by the company’s researchers, communicated in a secret language that scarred them for life. The experimental project had to be shut down immediately due to ethical and social concerns that might arise as a result. The reason for concern here was the doubt if the bots had developed conscience. To break it down, what if AI systems had developed their own moral sense just like humans? It may seem good, but at the same time is harrowing to imagine the possibilities it may lead to.
4. Lifespan Of AI
Another aspect to prospect is the concept of ‘life’ in AI. It may seem philosophical at first but how AI systems sustain for a long time on their own is still an unchartered territory in research. Interweaving the dimensions of life on AI systems needs a massive amount of insights from classical disciplines such as physics and biology. Bridging the gap between computers and particles of life is insurmountable.
5. The Intelligence In Artificial Intelligence
Even though AI systems are designed from the right information available from research, it is sometimes difficult to ascertain why they arrive at certain conclusions or why they behave in a specific way. Although there is a lot of research being done on this subject by organisations such as OpenAI and Deepmind, they are yet to flourish a gold standard to trace the path of intelligence in AI systems.
On A Concluding Note
AI’s existing problems are sometimes highly abstract in nature and may lead to different interpretations. Rooting technicalities needs diligence on both social and ethical aspects. In the end, it is not just about solving AI problems but about why they need to be solved.
Try deep learning using MATLAB