I am currently in the process of playing with an extremely interesting AI system called Josh, and it got me thinking about machine learning and what its really about. Right now Josh is just a version of Siri from what I understand. When I ask Josh from the weather he responds like Siri, they both go to an internet source for Josh its Wunderground and they determine my current location and give the weather for San Francisco. Its a simple request, but it also feels mechanic, theres no way of hiding that Siri and Josh are unintelligent, but behind them is a group of highly intelligent humans who have written hundreds of lines of code to make these machines respond to human request. Right now I am giving Josh a rule that if I request a song lyric he will get it for me from a website, and soon I will make it so he could return me a place from Foursquare. So maybe if I ask for Chinese food he can find some for me.
But yet to the average users these AI systems seem strange and cold, they are still not quite human enough for our taste. They are not quite smart enough. Our current AI systems can only take requests, and return to us data from a database or the internet. If you ask an AI system like Siri "what is love?", she would probably return a definition that she has in her database or one that she finds online. If you ask her "why does Fred not like me? I have called him eight times today", she might come up blank or just search that sentences in google and bring up a yahoo answers about clinginess. Siri lacks the ability to have an emotional response or even a critical thinking response, and to program her ability to be able to talk to you about why Fred does not love you because your clingy would take years, and thats just one question.
I think right now we are doing brute force AI's, we are hard coding everything they do. But the real question is it possible to hard code feelings and emotional responses? Maybe. Maybe Siri could start to recognize voice patterns and see when you are sad, and know how to mimic those patterns so she can seem sad as well. Can we trick ourselves is the real question. As we begin to understand the brain we have in our skulls, that sends signals to process data, very similar to our machine counterparts, I believe we will eventually be able to create a true machine that can learn and adapt.
But what is our end game for Josh and Siri? Do we want mechanical slaves or do we want something we can be friends with? Can Siri ever be my friend in the way a human being can be my friend. In theory I could have a Siri that has all my data, all my online profiles, my physical health, she could recognize my voice and speech patterns, and with all this data could she be able to "understand me" and be my "friend". She could probably give me advice, console me, fake love, but is this what we want? Are humans only a series of predictable algorithms made complicated by a brain? Creating machine slaves is easy, thats only the first step. But creating machines that can react to our feelings, feelings they will never have, thats the next step into dangerous territory. Will we have AI systems like in the movie "Her" that are able to "love" us and only want to learn from us, or will we have the terminator, or are the movies all wrong and we will go down a completely different path.
No comments:
Post a Comment