Not too long ago I made a post that simplified what AI is for people who are not so acquainted with tech. If you missed it, it’s here (also a locked post). This is a step further in the discourse.
There is too much hype about the future with a fully developed AI. It is said that AI will transform work as we know it. I agree with that wholeheartedly! What I disagree with is the way they say it will make people redundant. If we designed a system to make our lives easier, better, more efficient and more productive, how could we possibly be on the losing end of it? I’ll write a discourse about the future of work later, but today I want to begin with something that remains a mystery in the development of AI. You won’t believe it — jokes! AI still can’t make jokes by itself.
There was an incident last year on Good Morning Britain where a robot made a joke (see it here). But you can tell it was premeditated. And of course, they had to stay in a particular line of questions for the system to keep flowing. But seriously, why can’t AI make a joke? I believe the answer is just as simple; because it is AI! From the post tagged earlier, I explained in a very simple manner AI, machine learning, deep learning, big data. The system learns to think from data. It is based on the trends and patterns from the data that the AI makes decisions. The decisions have to be logical and defined. AI is just basically an efficient decision making machine. It is can be taught to think but it cannot create thought beyond the patterns and trends from the data it receives.
The big problem of stuff like comedy and jokes is how to explain it to the system. It is not a logical end result. Telling the system a bunch of jokes and telling the system to find the patterns is a dead end. Many times what constitute jokes to us are things from our experience that we can relate to in a particular way. If we cannot teach the system the experience of laughter, then to make a joke is a mirage. Also, audiences matter when it comes to jokes. What’s funny to a group of people is most likely not as funny to another group that has a different life experience. The AI cannot get the joke if it has not experienced the life. And it cannot experience the life because it is a computer! For example, a robot cannot get tired; it can either be able to do something or be unable to do it. Charging the AI system cannot be comparable to the human action known as eating. There are different types of food with different tastes and forms; there is just one charging port. This is more than just feeling, it is knowledge that comes from experience. AI can be taught to mimic the crying gesture, may even be taught the right words to say in those situations, but the fact still remains that it is a response from data and not real experience.
Back to the jokes, a model that shows promise where at least we can be able to laugh at AI created jokes is if we allowed it to learn the hard way. Accumulate data from many comedy shows (of a particular community) and let it keep doing stand-up comedy until it gets it. The end result will be set at the laughter of people. But then to another community, it will have to start all over again. Another good model is to allow the AI start out with one or two jokes (which won’t be funny) and then spin off with logical facts why it (i.e. robots) can’t make jokes (in a way of talking to itself). That pattern could work and work quite consistently. But an AI that can do a human joke well, is virtually impossible.
What about trolling? Or sarcasm? That would be nice too. But how will the AI understand the limits? Plus, funny people are not funny all the time and most tines when they are not funny, they are annoying. Who wants an annoying AI? Every now and then there will just be court cases against AI systems. What if it was sarcasm and the human took it serious and it caused damage? Then it ruins the purpose of creating AI in the first place. The purpose of creating the AI systems is for optimization and efficiency. If they are liable to the same issues as humans, then they are not necessary. An AI that can make a joke is like an AI that can make mistakes; not just conscious mistakes or errors but unconscious mistakes that are not errors. That defeats the purpose of AI. Incidentally it’s one of the things that makes humans standout. And really that is where art comes from. AI was built to be perfect, to have one that can do jokes is to purposely build a flawed one.
There can be robots or AI systems specifically built for jokes, but they will be pretty useless for everything else. This is because it takes intelligent imperfection to tell a fresh random joke.
Would you buy an art work designed by a robot? It’s the thrill of having a peak into the mind of the artist that makes the art work desirable. The AI doesn’t have it’s own mind. Any art work it designs is based on a pattern or trend observed in people (or other art works). Some may buy it, but the fact remains that it was designed for you to like it (not of the AI’s own accord).
AI has its place, but there are just some lines it cannot cross. This is not just because the system cannot learn it, rather it is because the system has to be anti-intelligent to achieve it. Anti-intelligence will take AI away from its standout edge. And if an AI can make a choice whether to be intelligent or stupid, then the world is in trouble; what humans have not achieved in thousands of years will be achieved in days! (The total destruction of the planet).
In conclusion, water can flow uphill but there has to be a pipe and a pump. Why are we trying to sabotage ourselves by giving AI the capacity to be dumb and stupid? AI can make us laugh unintentionally, but if it becomes an intentional thing (that they can control), then the purpose of AI has already been defeated. Will AI ever be able to make a joke (like we do)? Nope.