WHAT IS AI ANYWAY?

Let's debunk and demystify some of the misconceptions and false beliefs about Artificial Intelligence. (AI)

We have all seen these sci-fi movies about futuristic droids that are super-intelligent: they swiftly take over the world and make all humans their slaves. 

The current state of AI could not be further from the truth. Let's talk about what it actually is.

How does AI really work?

The first thing to understand about AI is that it is composed of nothing more than algorithms. On its own, an algorithm is no smarter than a power drill. It is good at doing one thing and it does that one thing very, very well. The magic of AI happens when you chain a lot of algorithms together in a clever way, allowing a string of several different algorithms to perform what none of them could have done on their own. When you think about AI,  you need to think about it as a chain of algorithms that follow step-by-step instructions. 

When you have a chain of algorithms joined together, you easily get this illusion that  what you're receiving is the outcome of an intelligent behavior. After all, that’s why it's called artificial intelligence. It is artificial – man-made. It is intelligent only in the sense that it tries to mimic intelligent man-like behavior for the decision-making process within some pre-specified domain.  

Let's look at an example. A smartphone is an AI device that all of us carry in our pockets. .

If you ask the personal assistant on your smartphone – say, Alexa or Siri –  for the best place for sushi around you, what is actually happening?

I. First, there will be an algorithm that will take the sound waves of your voice recording and translate them into a digital signal. 

II. Next, another algorithm will translate this digital signal into words that are sent to a search engine. 

III. The search engine will look at reviews of sushi restaurants and sort them according to the GPS location of your phone. 

IV. The results from the search engine will be taken by another algorithm that can create a response in a sentence that humans can understand.

V. And the final algorithm will reply to you and trigger the speakers on your phone to make a sound wave for you to hear.

Most people don’t recognise the individual steps. Instead, they'll think  “oh wow, there's some mystical supercomputer communicating with me.” In reality, the whole procedure is just a clever way of tying together the results of simple algorithms. Each algorithm does its own specific thing and when you join them together, the result often convinces us that the device has an intelligence of its own. 

So, the first thing to remember about AI: it is nothing but a chain of algorithms.

Second, there are two distinguishing features of AI that separates it from traditional programming. 

AI deals in probabilities, not certainties. 

In trading algorithms for example, there is a probability of hitting a predefined price target and the confidence level tells us how confident the algorithm is about this projection. AI does not know for certain what's going to happen. It can only analyze past data and then give you the projections for how likely is the outcome – given the inputs and programming of the algorithm itself.

The second feature that distinguishes AI from traditional algorithms is that AI is programmed to learn the instructions from data itself. In traditional programming, developers must be very precise in coding the exact rules for the algorithms to follow a predefined path: “If this happens, then do that.”

What AI does is that it actually takes the data itself and learns what to do, deriving a pattern from the underlying relations. You can give a lot of animal images to AI and ask the AI to group the images in different clusters with the idea that it would place images of cats in one pile, images of dogs in another pile, and so on. Pattern-recognition on a computationally high level. 

For trading algorithms for example, you can give a lot of financial data to the AI and ask it to recognise patterns and rules that are profitable. You don't need to specify how and when to trade the market: The algorithm can learn these instructions from data. You just give the algorithms a lot of data, an end goal, and let it crunch the numbers to find what works and what doesn’t work. 

Okay, so now that you know what AI is, let's look at the chain of our trading algorithms and how a software’s brain works behind the scenes. 

At the base you have the inputs layer:the data feed. DATA is the bloodstream of the algorithm. We continuously receive data from the markets for all the symbols, all the assets that we track, their price information and their different indicator values. This information is continuously pumped into the system, 24 hours a day. We receive new data and we put it up into the next layers, which is the computational layers, where all the calculations happen 

In layer 2, we start to label all this data t and perform pattern recognition. In layer 2 we give the raw data from layer 1 some structure. We label this data and give it attributes. The attributes can include all sorts of qualities: what pattern or instrument is it, , what time frame, what is the market direction, what is the overall market trend and so on. This data labeling allows us to assign specific market patterns into separate clusters.

Beginning in layer 3, we can start to analyze the patterns that we labeled in layer 2. We can analyze them in various ways by looking for how to best trade them. Where to place the Targets, where to place Stops. To achieve profitability, which way is the most optimal way to trade these patterns?  Based on how they perform, we assign weights to these patterns. 

In  layer 4, we take all the weights from the patterns and calculate the expectancy for these types of patterns. When you know the potential reward for the patterns and you know how likely you are to get this reward, this allows you to make an informed decision whether or not to place a trade for any particular pattern. 

And finally, once all the calculations in layers 2,3 and 4 are done, we arrive at the output layer. This takes the form of  a text message, sent to the app. So when a new pattern has gone through all these layers – starting from the raw live market feed, through the computation layers of pattern recognition, testing and expectancy algorithms – the output message you receive in the app and telegram will contain a chart and information about this new pattern.

The output message will read:

“Hey trader, I detected a potential trading opportunity on EURUSD Daily chart. The expectancy for this pattern is 0.50 (hit rate% is 50%, Reward/risk ratio is 2, Confidence level is 90%). Do you want to place this trade with 1% risk?” 

As a trader, you can see this pattern on a live chart and make a decision based on this information: by simply tapping YES, you directly place this trade on your trading account. 

For when you next receive a trading signal, I hope that this explanation helps you understand what happens behind-the-scenes in the trading algorithms.