Artificial Intelligence (AI) - latest craze or disruptive technology?

Artificial Intelligence (AI) - latest craze or disruptive technology? 

It seems everywhere in the media people are talking about AI and how it's going to take over everything. They say once we hear something around 6 times from difference sources we start to believe it's true. 

What is AI?

The dictionary defines AI as:

"the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages."

From this point of view we have already been using many AI systems for years and they are already part of our daily lives. Speech recognition tools such as Alexa are finally starting to become useful in a limited set of circumstances and people have been successfully using Google Translate to obtain rough (read as literal) translations for a number of years.

AI Limitations

The interesting thing about all AI currently is that they are just algorithms that try to perform a very specific task well enough to be useful in some circumstances. For example you can ask Alexa for the time or to play some music but it's still incapable of answering the most basic questions if you stray even slightly outside the list of questions it has had been pre-programmed to answer.

What AI doesn't provide is true intelligence, it can't find a solution to a problem that it hasn't already been designed to solve or even solve the same problem is a slightly different way and that's a massive limitation. 

If you were to hire a receptionist at work you might ask him to greet visitors, check they have an appointment, get them to sign-in and then call the person they are visiting. You wouldn't be expected to have to tell your receptionist that if the bosses wife rushes into the office with their sick child and asks the reception to call her husband right away that you can go and get him rather than just turning her away as she doesn't have an appointment! Why would you tell them that, it's common sense and everyone would know that, but an AI wouldn't unless someone thought of that situation and programmed it as such!

Neural Networks

So what's this about Neural Networks, surely they are the brains in AI? Well, again not really. A Neural Network is a fancy term for an algorithm which comes up with a probability of the input being close to other inputs that it's already seen. For example a Neural Network can be configured to identify chairs, and if you show it many pictures of chairs it will find things looking like those pictures. But for example if you have shown it chairs just with leather seats and then you show it a chair with a wire mesh seat it might decide that's not a chair and then you would need to "teach" it this new type of chair. 

A Neural Network is not intelligent and can be easily fooled by stimulus which is clearly unrelated to what they "should" be looking for. This is one of the reasons why Neural Networks are used in combination with other more deterministic algorithms (e.g. facial recognition algorithms that detect eyes and nose) and not just by themselves.

This is fundamentally different from how a human would approach the same problem. When we first look at the chair we would think, hmm that looks a bit like a chair but the seat is different. Then we would think ah it's just a chair with a different seat design and then store that design away so that next time we see it we don't have to think about it next time. 

What does the future hold?

As yet nobody has invented a way to make a computer truly intelligent, to go beyond it's programming and be able to think for itself. The topic is currently so wrapped up in media hype that people are starting to that they can but they can't, at least not yet.

There's an interesting theory (AI singularity) that once we invent a computer more intelligent than ourselves that computer will in turn design a smarter one and so on until AI will eventually be infinitely smart. Depending on your point of view this then triggers either an Apocalypse or new some Utopia! In practice this seems highly unrealistic, our experience tells us that any development in technology tends to be gradual and has it's limitations in the real world (e.g. energy demands or other resources come into play). 

Why does it matter?

It's very easy to jump on the band-wagon and forget that technology has its limitations. If we hand over more and more responsibilities to flawed AI algorithms there are going to be serious implications in the real world. Autonomous vehicles are of course one of the obvious areas that could result in death or serious injuries as a result of a bad computer algorithm but there are many other areas where people's lives maybe impacted. Imagine being denied boarding of your flight or being denied hospital treatment because an algorithm says so.

Of course AI has the potential to provide loads of cool new benefits to our lives and help us in ways we haven't even thought of yet. We shouldn't forget that despite this they are flawed algorithms which have many limitations and that human decision making should always be able to override the machine.