Although artificial intelligence has been around for many decades, it has emerged as the next big thing only in recent years. Everyone is talking about it but in my view not everyone knows precisely what it is and what it is used for. I hear the term artificial intelligence used in many different contexts, anywhere from simple data analysis to full-fledged robots with the potential to conquer the Earth. Let’s look further into what is artificial intelligence, how it is applied, how mature it is and what are its promises and challenges.
In science fiction, artificial intelligence is often portrayed as computers or robots with human-like characteristics, sometimes also in human form, but not necessarily. A machine that thinks like a human can be a representative example of artificial intelligence. It reminds me of HAL, the computer in 2001: A Space Odyssey, a prime example of artificial intelligence that is willing to exert extreme measures to protect its own existence. But generally, artificial intelligence refers to machines that can learn, reason, and make their own decisions when faced with new situations, just like humans do.
In the current reality, we still have a long way to go until we get there. The artificial intelligence that is available today, such as Siri, Watson, image recognition, self-driving vehicles and so on is limited to a narrow field of expertise, for example playing chess or performing the repetitive tasks of a personal assistant. In all of these cases machines must make decisions by weighing consequences of any action they perform, taking into account external circumstances. These machines have the capability to learn from their own behavior. We do not yet have artificial intelligence that can do everything that a human being can do. However, machine learning is improving at a fast rate and is becoming better than ever.
Some examples of machine learning applications that are already in use today include:
- recommendation systems when we buy stuff, for example a book recommendation on Amazon or a movie recommendation on Netflix
- banks predict changes in the stock market
- insurance companies calculate policy premiums and predict claims fraud
- police forces identify suspects from grainy images and predict locations and time periods with a higher likelihood of crime occurrence
- courtrooms offer advice about whether to grant bail to criminal suspects
- image recognition helps doctors to spot disease or to calculate the probability of illness such as cancer
- linguists use artificial intelligence to help in translating or to decipher lost languages
- scientists can predict probability of earthquakes or calculate weather patterns
- self-driving cars use machine learning to navigate roads and adapt to traffic conditions
- companies make decisions in hiring and firing processes
- and many more…
Artificial intelligence and robots
When we talk about artificial intelligence, we can hardly avoid associating it with robots. But the two may not necessarily mean the same thing. On the one hand, a robot can be intelligent, but on the other hand you can have robots that have been programmed to perform a specific task (flipping burgers in a fast food restaurant, or assembling car parts in a manufacturing plant, to name just a few), without asking them to make intelligent decisions or learn from their own behavior.
The main difference between an algorithm that can be programmed by anyone with a sound knowledge of computer programming, and artificial intelligence is that artificial intelligence can solve problems and make choices based on previous experience and learning, similar to how humans interact with the world around them. A robot is simply a vehicle that contains the algorithm or the intelligence, in a manner resembling how the human brain guides the human body.
Natural language processing
As anyone who has learned a foreign language can attest, learning a language is a daunting endeavor. There are rules and grammar structures that can be learned, but there are exceptions to the rules and nuances in chosen expressions that can’t be simply memorized. A language is learned by constantly using it, being immersed in that language’s surroundings and getting it into the subconscious mind. That is why it is so difficult to program machines to do automatic translation from one language into another, because there is no simple algorithm to transfer the rules of one language to another.
As we know from Google translate, taking phrases or expressions and translating them accurately is a challenge, resulting in occasional misleading or comic results. Google translate is a perfect example of how experience improves results, as we have seen that translations are becoming more accurate over time due to the self-learning nature of the algorithm. When the algorithms reach a point when we can say that they can interpret every single input phrase correctly and translate it into another language, conveying the intended meaning, then we may look forward to real-time translation, so that there would be no need for anyone to ever learn a new foreign language again.
Taking it one step further, natural language processing is not limited only to translation. We expect future artificial intelligence to communicate in a manner that resembles natural human conversation. Should we expect artificial intelligence to understand our spoken language and speak back to us? We know that speech recognition already exists, but the question is, can a machine understand what we are saying and read between the lines so that it can talk back to us like a human being? Will artificial intelligence make jokes? Will it express emotion? Should it?
Risks
What if artificial intelligence will one day prevail? Scientists predict that true artificial intelligence, such that resembles human reason, is still decades away. While we may think that is too far in the future, it is never too late to think about safety to ensure that science fiction scenarios where artificial intelligence eradicates the human race will not happen. We can hear many thought leaders, for example Elon Musk and Bill Gates, warn about too much autonomy in artificial intelligence development already today.
We don’t even have to go far into the future to think about disaster scenarios. We already have examples of artificial intelligence going astray (incorrectly identifying a criminal suspect, for example) or artificial intelligence being just annoying. Many of us have had unpleasant experiences with recommendation systems that don’t always hit the mark, as they sometimes recommend stuff that we would never want. Can we expect that recommendation systems will improve over time and will eventually become more intelligent? Or will the artificial intelligence of the future be as awkward as some recommendation systems appear today, thus causing chaos instead of technological advances?