AI of the future!
AI (Artificial Intelligence) frequently makes the headline and this is one sign of its rapid growth. In this post we will highlight 5 points to keep in mind in any project related to AI.
The importance of meta-data
You all know meta-data, those data associated to the main information. In an email, that could be the sender and the receiver address, the time when the mail was sent, the header title. In short, it represents the context of the data. They are really important and make data processing easier and faster. For example, if you’re looking to analyse a picture, looking at metadata could help to deduct an information and not force the algorithm to rediscover these metadata again. How to trust a new data? If there is associated with it the place, the time and by whom it has been done it would really help. But even more important there should be a way to be sure that the meta-data are genuine and to achieve that there could be one more metadata that would serve as the certificate.
Do not underestimate the 2nd use of data
One thing that the industry teaches us is that: with some data collected for a specific use, it may be possible to reuse them again at their benefit to make money. So, whatever your projects are, just keep all the data. Memory is cheap and later you might discover value on them. Here is a simple example to illustrate this idea, let’s take a hospital that collect data about a patient to diagnose them, this will be the first use. As a second use, these data could be very valuable taken alongside those of many patients to help with automated diagnosis and in disease prediction.
Ai & Ethics
AI is the next revolution and like for everything powerful we need to be very careful about what we are doing. Try to keep it under control. This will be the responsibility of every player. With time AI becomes more and more capable, it is now possible to create a fake video which brings back the problem of genuine data. This will force us to be more critical about the information we get with a huge focus on their origin.
But do not be mistaken, people also have learned and discovered that in some case, it was possible to trick the AI. A good illustration of this point is in image analysing, where a noise that looks like random, can change the prediction of its content without human being able to tell any differences. You have to be aware that the result returned by an AI could be false, even though the AI does well on tests, especially if the input has been constructed to fool it.
There are the big player we all know, like Google, Facebook or Amazon. But these are not the only ones. These players have a lot of clients, which allow them to collect a massive amount of data. This data is used to improve their product or make more money. The issue is for the others, new or small competitors. They are at a disadvantage because they do not have those pieces of information. And even if their idea is better than what is on the market, they could just fail due to this inequality. There is some law that should be set, for example at the European level to tackle this problem and ease the transition between actors. The GDPR is already heading in this direction, but we should go further.
The power of people
As you can see thanks to the big players out there: knowledge is power, and the more knowledge you can amass, the stronger your power gets. But people can willingly help bring knowledge to good use. One example is the Duolingo application that help people learn new languages and in parallel use people translation inputs to find the best possible translation for book sentences. By combining people's knowledge, many things can be achieved, such as big data sets that will later be processed by AI. Many applications ask for people help in order to train their AI and to get a lot of useful data.
We should not be pessimistic about the future with AI, but be vigilant about what we are doing. If you liked this post, learned something or just know someone who has projects related to AI please share it!
By Arnaud Gellens (Master in Computer Science and Engineering, UCLouvain) and Romain Hubert (Master in Computer Science, UCLouvain).
Special thanks to Prof. dr. ir. Erik Mannens for his awesome presentation on which this post was inspired.