# What are limits of AI and deep learning?

I was testing AI (artificial intelligence) and deep learning for one project and during that time I was thinking where we can use AI and were are limits of this technology.

At first we must tell what AI is but don’t worry – we won’t talk about complicated mathematical formulas. We can say in general that the core of deep learning is statistics and calculating probability that some of known events would occur. We better explain that on example. You can imagine receptionist at a company seeing Mr. Jones come to work every morning. Every day Mr. Jones comes at 7 o’clock but on Tuesdays he comes at 8 o’clock because he takes his son to school. One day somebody ask the receptionist at 7:30: “Is Mr. Jones already here?” Receptionist says: “Oh no, it is Tuesday today, so he comes at 8 o’clock.” The AI is so clever that it is not confused when something unusual happens. For example when Mr. Jones one Tuesday comes into office at 7:00 the receptionist asks: “Why so early today, Mr. Jones?” and Mr. Jones reply: “My son is ill.” Next Tuesday if Mr. Jones came at 7:00 the receptionist might think that son is still ill.

This technology is very good when we have big amount of data with vast of parameters. At this point we can estimate future steps according to current situation and known history. We can also identify deviation from normal state. What it is good for? This technology has good result in estimating future progress of cancer patients. There was study where AI has beaten experienced MD in survival prognosis. Great result has this technology in enhancing photos. The AI thanks to statistics of billions of pictures knows what the picture should look like. When it doesn’t look well the AI tries to repair the picture according to known patterns.

Every tool is suitable only for some kind of work and with AI it is the same. For example if we need to find out a new unusual solution then we have to use our brains. Of course there is still chance to use AI if we have enough data from similar situations in other areas. Then we can use AI to find out analogous solution.