They all suffer from the same problem - not being able to deal with anything beyond very basic and targeted language. As soon as the conversation needs to be more detailed or more nuanced, there is a problem, which either (in good scenarios) ends up with the service falling back on giving you a human to interact with, or in pure frustration and anger as the limited (or just poorly designed) bot isn’t capable of helping you. 

I think the question of how human a chatbot should be is contextual - there are some cases where a chatbot that is too human could cause issues, such as in healthcare apps, and others, where it doesn’t really matter as long as it is sophisticated enough to actually help you. 

The more human they are, the clearer it needs to be that it is a CHATBOT and not an actual person you are chatting with; some services still fail to make that obvious at times. 

I tend to observe a similar problem across all of these technologies which link to an inability to deal with complex queries issues. They focus on the basic stuff alone. Once the conversation reaches the point of requiring some level of human-level cognition or application of intuitions, the bot do quite awfully. In the best case you get directed to a human support and in the worst-case chat terminations that can increase frustrations for the lack of a capable help to your problems. 

 

I think it is hardly feasible to get chats bots to talk like humans. I think the problem may be linked to what I consider over-expectations by human. If we understand the limitations of chat bots -  what they can and cannot do, then it becomes easier to manage expectations of satisfying customer queries in the same way human support will do.