Firstly, it is always good to clarify exactly what we are talking about… Artificial Intelligence (AI) techniques are techniques that essentially allow a machine to learn how to do a specific task, without being programmed with the explicit instructions. To most people, this definition sounds rather eery, but in reality virtually all things that we don’t understand appear scary initially. Some of the technological achievements that we are experiencing right now seem unrealistic, and some may even say, almost beyond comprehensibility. One point that I must also make at the start is that technological development can sometimes be somewhat of an illusion. Developments that we may appear very close to (like General Artificial Intelligence) we are actually still very far away from, and things that seem far away may be closer than we think. Effectively, what I am saying is that in reality we don’t really know when a development will take place, until it does – so don’t put a time horizon on whatever I say.
To keep it relatively simple, the algorithm is programmed to try and recognise relationships within a set of provided data. Over time, the algorithm improves and improves (‘learns’) until it gets very good at the task that it has been designed to learn how to do. Most people are effectively worried about the idea that artificial intelligence can surpass human intelligence, which is known as ‘super intelligence’. Many headlines that are written are extremely sensationalist, and one example of this was the reaction to the Google’s AI beating the world’s best ‘Go’ player at the game in 2017. The AI had been specifically programmed to win games and was given the rules, it would then play a vast number of games and learn from each situation. In each game it plays, the algorithm calculates the best way to react to each situation, and registers what is and isn’t successful. In some cases, this allows the AI to create moves that humans have never played before.
Now I can see why that might sound scary… but does that make artificial intelligence smarter than humans? Of course not! Many children can acquire skills extremely quickly just by watching someone solve a problem a couple of times or sometimes even just once. On the other hand, AI needs huge amounts of data to be able to learn. This learning process is only regarding a very specific problem, whereas it is the versatility of humans that makes us extremely unique.
One thing that is worrying though, is how people may use AI. In the modern era, people are more concerned about the performance of the algorithm, rather than the intricate workings of the ‘solution’ that the algorithm has come up with. In essence, they can be seen as black boxes (shown below), where data is given to the algorithm and an output is produced, yet no one really understands what is actually happening inside of the box. Humans are always more focused on performance more than understanding. We need to ensure that the ‘solution’ that algorithms come up with – is the socially desirable solution.
For example, the algorithm might identify relationships based on historical data, and we need to ensure that relationships that are identified from say 20 years ago, are still representative of ideologies today. One example of this could be in identifying the best candidates for a job. If the job in question was computer science, which was previously very male-dominated, the algorithm may favour the male candidates in future decisions (which is gender discrimination). Hopefully that explains the dangers behind not understanding what the algorithm decides to do…
Another thing that is somewhat scary is the idea of this control problem which is referenced in Nick Bostrom’s book – Superintelligence: Paths, Dangers and Strategies. This control problem refers to the idea that you give the AI some desired outcome, and the example that he uses is making paper clips. The AI must then make as many paper clips as possible with the available resources. The scary part is that the AI could look at the resources it is being provided with, and then believe that it isn’t explicitly limited to those resources. Subsequently, it may try to turn the whole planet into paper clips. Now while you may laugh and say, ‘why can’t they turn it off?’, or ‘why can’t they program it not to do that?’, the reality is that it simply isn’t that easy.
AI has also penetrated many of our lives without us even really realising it. If you have an Amazon Echo with Alexa or a Google device with Google Assistant then you are already dealing with an AI. These voice assistants are constantly learning how to react to new situations, and will only continue to improve. Are you scared of your Google Home? (Well maybe some people are…)
Regarding this topic, there is really no need to worry that much about it in my opinion. If people are sufficiently worried about such a negative outcome, then it increases the probability of avoiding that outcome – as you actively ensure that it is avoided. Furthermore, far smarter people than you and I are already working on ensuring things like this don’t happen. For example, Elon Musk and a number of other wealthy philanthropists pledged US$1 billion to help create OpenAI, which basically works to ensuring that the journey towards general artificial intelligence, is a safe one.
Worrying about AI is a waste of your energy. Unless you are someone that is actively working on artificial intelligence, or working anywhere near that field – then there is nothing to worry about. Worrying about AI is the equivalent of constantly worrying about an asteroid obliterating earth, or a super-volcano exploding and destroying large sections of the planet. Most people are worried about AI as they think of things like Terminator – where AI essentially goes rogue, when maybe instead they should think about something more like C3PO from Star Wars!
If you enjoyed this article, or believe that the article could be improved in any way – just leave a comment below. I’m also open to any questions that you want to ask – and as a non-AI expert, I will do my best to give you a detailed answer…