When I first saw the news about Cornell’s Creative Machines Lab and the new results provided by their chatbots, I found it quite interesting. AI has come a long way since 30 or 40 years ago when the first “intelligent” games appeared on the market. Nevertheless, despite the results, I still believe that there is a lot more work to be done before these chatbots can really be called clever.
Now, while the results are amazing and they definitely show that these people have put a lot of work into programming the bots to give intelligent replies and even form conversations, I think the main problem is that they still attempt to mimic the human mind without actually following its main characteristics.
The bots are basically programmed to choose from a large variety of possible replies what they calculate as being the correct one, according to a complex set of rules. The human mind simply doesn’t work like this and I believe the whole approach is far too linear. Creating a valid blueprint for a chatbot would require not only teaching it how to speak but making it capable of learning how to think as well. This is probably the greatest challenge that AI experts have ever faced and, while current results show some excellent progress, I think there is still much work to be done before results can really be impressive.
