There are many articles written about the impact of artificial intelligence in the labor market. We are heading towards a fully automated world. And now that the end of the mobile-first era is approaching, we all need to prepare ourselves for an AI-first society. But what is AI?
The word “Artificial Intelligence” actually speaks for itself. It is an artificial form of the human intelligence. AI tries to simulate that human intelligence with algorithms.
We, human beings, take small and bigger decisions during the entire day. The combination of intentional and unintentional behaviours simply helps us to achieve a goal. We accumulate our intentional behaviours in the course of time, mainly while taking decisions based on experience.
That is the core of “learning”: you act in some kind of way and (intentionally or unintentionally) you evaluate the result. Did that action or decision bring you closer to your goal? In that case, we learn automatically that it was a good action. Was the result less desirable? Than we keep in mind that we will avoid taking this action or decision to achieve a goal in the future.
We learn how to bike, to calculate, to send e-mails, to use a new App, how to conduct socially, etc. People learn their skills relatively automatically. During these learning processes, we are not aware of the millions of small decision processes being taken.
That is the big difference between computers and human beings. Computers have always been very good in repetitive calculations based on detailed instructions and that in a relatively stable environment. That stable environment is impossible in real life. In contrast to computers, people live in a world that changes constantly and we gain intelligence without detailed instructions.
The development of Artificial Intelligence is a subject scientists have been studying for decades. Although AI has been a subject of research for quite some time, the topic is currently more important than ever. On the one hand, the reason for this is that people accumulate more data than ever before. On the other hand, we more and more learn for what purpose data can be used. Furthermore, as a baby will acquire knowledge by growing up and by collecting experiences, that is how AI becomes more intelligent, namely when there is more data to learn from.
Presently, a lot of software consists of different features. Developers have determined for every feature how they have to deal with certain kinds of input. Action A will lead to result B. This kind of software does not become more intelligent by being used more frequently. This (older) software does not adapt itself to your personal habits. That is called “regular computer processing”.
Furthermore, Machine Learning is a type of Artificial Intelligence. Software that gains more intelligence by processing more data has learning capabilities and falls under the category of “Machine Learning”. These algorithms do not have to be explicitly programmed by people.
Google’s services are clear examples of Machine Learning. Millions of people have been using Google Search and Google Translate worldwide for years. When someone types a search term in Google’s search bar, that word is saved. If you write typo’s, Google suggests the correct spelling. That does not happen because a developer has learned Google Search what the correct spelling of the search term is, but because many before you typed and searched the same word. Consequently, the system decides that there is a bigger chance that you tried to find the correct word than the search term with the typo.
Google Translate is another example of Machine Learning and is the translate tool par excellence for many of you. By using this tool, Google suggests a translation of your written text. If you are not happy with the translation, you can adapt it. Consequently, these adaptions will be memorised by algorithms. The next time someone wants to translate the same piece of text, the algorithm takes your adaptions into account. In that way, you can imagine that Google Translate becomes very intelligent at a high rate. Every user is important and contributes to the data collected. In this case, we talk about user-generated data.
Waze gains knowledge in a similar way. Waze-users are drivers. These drivers can indicate if a traffic jam was caused by an accident, stationary vehicles along the side of the road, etc. While giving these data, we all help Waze to become more intelligent and to gain more insight into our daily road use. It becomes even more interesting when Waze is to divide traffic flows between different routes to decrease daily traffic jams. Based on these predictions, Waze-algorithms can suggest actions to make our life easier and more pleasant. That is where AI becomes really valuable.
In these examples the user, in this case the human being, generates data to make the system more intelligent. Nonetheless, data can be collected in many other ways.
Let us for example look at how an elevator works. Every time an elevator is used, the system saves different parameters such as speed, weight, the amount of stops, temperature, the number of errors and so on. Based on these data, a lift producer will be able to predict when maintenance is required before the elevator really breaks down. By generating more data, the algorithm is optimized and can in that way make more correct predictions. As a lift producer you are now able to provide a more reliable service.
However sometimes it is not possible to collect such quantitative data.
Take for example diagnoses based on symptoms or the determination of the best treatment. Experts, which are in this case doctors, have been making diagnoses based on a combination of symptoms. These data make it possible for algorithms to make a diagnosis and to suggest a treatment. The accuracy of these data and the amount of data determine the success of such Artificial Intelligence. That is why it is crucial for the data to be generated by users such as doctors, scientists and professors.
We process a lot of data on a daily basis to make the correct the decision or to take the right actions. Every developer in the domain of Artificial Intelligence takes on the mission to simulate human processing algorithms. Up until now, many steps have been taken to artificially simulate certain aspects of the human brain.
Nonetheless, it will still take some time to develop an algorithm that can process the same amount of data as the human brain and that can make meaningful predictions. Many AI-tools are very good in one defined task.
In some cases, these algorithms are even better than the human ability to process. Why? Because algorithms process enormous amounts of data in a very objective way. People, conversely, use often shortcuts in their decision-making process. Mnemonics, memory aids, induction and deduction are well-known examples of shortcuts or “heuristics”. They can help us find a solution or make a decision more rapidly. These shortcuts are necessary as we humans are often forced to take decisions based on incomplete information.
Wikipedia describes the difference between heuristics and algorithms as follows:
Heuristics are informal, intuitive and speculative solution strategies, which people develop to tackle certain problems. Unlike algorithms, which work anytime and anywhere, heuristics are specific strategies that we learn to use in specific situations and that do not always guarantee a solution. Heuristics provide general guidance on possible solutions and save us a lot of time and effort by limiting the solutions to those we are most likely to apply. Every task has heuristic possibilities, and the more experience one has with a task, the better one is able to develop good heuristics.
In 1997 Deep Blue won from IBM from Garry Kasparov, the then world champion. This is an example of a computer that can reach more accurate decisions than humans. The processing speed depends on, among other things, the calculation speed of the computer, something that has evolved enormously over the years.
Next to that, the algorithm’s structure is another condition to approach human intelligence. The more parameters are included in the algorithm, the more the process approaches the human ability to process. But consequently, the more complex the algorithm becomes.
Artificial Intelligence is therefore “smart” within a certain domain. Yet, people are still responsible for the parameters used as an algorithm’s input.
In God we trust and all others bring data.
Algorithms need to be able to recognize certain patterns in data. For the Waze-algorithm to organise daily traffic, it has to know where daily traffic jams occur. The knowledge of daily traffic jam patterns is generated through data collected over time.
A patient will gain more trust in his or her diagnose made by a computer that can make predictions with an accuracy that approaches perfection. In that case, we talk about perfection only when the computer makes as much or less mistakes than specialists.
The accuracy of AI is determined by tests with test data. Test data are in that case a combination of symptoms of which you are 100% certain that the diagnosis is correct. If an enormous amount of test data is processed by the algorithm and if the prediction is time and again identical to the actual result, than we can conclude that the system is as smart as the specialist within these parameters.
Are you planning on developing a smart solution that can, for example, solve the car park problem in your city? Make sure you have access to sufficient and adequate data within that context. Collect, for example, data about the use of existing parking facilities, parking habits, the number of inhabitants, data from parking meters, etc. Only with this access algorithms can make accurate predictions to manage existing car parks as efficient as possible. The better your predictions are, the more people that will use your solution. And next to that, the more data you will collect. In that way, your predictions will be more on point to attract more customers and the circle goes on.
Do you intend to invest in a smart tool? Keep in mind that sharing your data will make the tool more intelligent and precise. In that way, the tool will be more of value for your organisation. However, it is impossible to develop in a vacuum. By sharing data, others will consequently benefit from the increased accuracy. Furthermore, your organization will also take advantage of the data shared by others. We could compare AI to love. Moreover, AI can be explained as Joey once said in “Friends”: ‘It is a love based on giving and receiving as well as having and sharing.’
I would like to end by advising you to start collecting data NOW and to build strong algorithms. Artificial Intelligence and Machine Learning confirm the “first mover advantage”.
I go back to a previous example to illustrate this. At the end of the nineties, Google developed a fairly primitive, but genius algorithm to improve on search predictions. That algorithm had an enormous success. Google made all information accessible to and usable for everyone. More and more people started to use Google and in that way Google collected more data. Consequently, the accuracy of the predictions increased. Other big companies, such as Microsoft, were just standing still passively. It soon became for these companies too late to catch up with Google. Furthermore, it became impossible to equal Google’s impressive algorithm as it collected an enormous amount of data.
1. The users who generate data: customers from more than 100 companies from various sectors (banks, insurers, retail, B2B, healthcare, etc.)
2. The data: feedback from all these customers about their experience with a particular organization: "Customer feedback about their Customer Experience"
3. The algorithms that process the data are of course our biggest secret. What I can tell you is that we process the customer feedback into insights. This way, every company that works with Hello Customer can immediately get started to improve the Customer Experience even further!