Adil Moujahid, Technical Manager at everis (an NTT DATA company) and PMO Manager of the NTT DATA AI Center of Excellence, talks to Gernot Kapteina, Founder of OYSTEC, about the importance of Artificial Intelligence (AI) in the present and future.
Adil, thank you for doing this interview about Artificial Intelligence. First of all, if we were not talking face to face right now, but over the phone, how would I make sure I am interviewing the real Adil and not some AI impersonating you?
Moujahid: Thank you, Gernot, for this invitation, it’s always a pleasure talking to you. Your question is actually very interesting and a great way to start our conversation about AI. I think that it’s very difficult to be 100% sure that you’re talking to a real person unless you see the person in question by yourself in real life. The best you can do is apply the Turing test to check if you’re talking to an unsophisticated AI. The Turing test was developed in the 1950s by Alan Turing to assess AI’s ability. The test in its original form is text-only. You should select a few questions and send the same to a human and the AI. The answers should be anonymous, and based on them, you should determine which is which. If you can’t reliably tell, then you can say that the AI has passed the test. Now going back to your original question, and let’s assume that you’re talking to an AI. If the AI doesn’t pass the test, then you can tell that you’re talking to an AI. If it does pass the test, then you won’t be able to tell if you’re talking to an AI or human.
Thank you for this interesting entry. Let us talk about you for now - can you tell our readers something about your professional background?
Moujahid: My name is Adil Moujahid. I was born and raised in Morocco until I finished high school at the age of 18 and decided to move to Japan. There, I joined Tokyo University of Foreign Studies, where I took a one-year Japanese language course aimed at international pre-undergraduate students who are planning to study at Japanese universities. After the language course, I went to Kyushu University in Fukuoka and studied electrical engineering and computer science. I stayed there for six years, during which I completed a Bachelor of Engineering in Electrical Engineering and Computer Science and a Master of Engineering in Information Science and Electrical Engineering. After receiving my master’s degree, I moved back to Tokyo and joined NTT DATA. For the first two years, I worked in the Global Business Sector, where we supported newly acquired companies in their post-merger integration process - e.g. expanding their solutions to new markets. I then moved to the Research and Development department, where I worked as a data scientist in various initiatives and projects in the healthcare, manufacturing and transportation sectors. In 2017 I moved to Barcelona to join the Artificial Intelligence team at everis which is an NTT DATA company and I am now PMO Manager of the NTT DATA AI Center of Excellence (CoE).
That sounds like a solid resume. But your AI expertise didn't just become apparent when you moved to Barcelona. I remember that back when you were living in Tokyo, you were working on AI projects. When did you first become interested in AI?
Moujahid: Absolutely right. My interest in AI began when I studied Machine Learning at the university. The topic of my master thesis was related to AI. I investigated the application of Machine Learning algorithms to reconstruct high resolution images from low resolution images. Also, through my professional career at NTT DATA and everis, I had the chance to work on different AI projects and initiatives that showed me the impact of AI and grew my interest in the topic.
In recent years, AI has become a major social trend. But not everyone understands AI in the same way. How would you define "Artificial Intelligence" from your professional point of view?
Moujahid: AI refers to machines or algorithms that demonstrate behaviors or capabilities associated with humans. In most cases, by “Artificial Intelligence”, we mean Machine Learning. Machine Learning is the branch of computer science that deals with algorithms that learn from data. Machine Learning algorithms are usually divided into three categories: First, there is “Supervised Learning“: In a supervised learning task, we use a dataset that contains both the input and the output to build an algorithm that can take a new input and makes a prediction about the output. Second, there is “Unsupervised Learning“: Unsupervised learning algorithms try to find patterns in unlabeled dataset (from the input only). And finally, there is “Reinforcement Learning“: Reinforcement learning algorithms are used in tasks like training robots to walk or training computer agents to play games like chess or Go. Over the last 10 years we have seen a huge increase in the popularity of a branch of Machine Learning called Deep Learning. Deep Learning refers to a class of artificial neural networks (ANNs) that consist of many processing layers. These algorithms have had many successes in the fields of computer vision, automatic speech recognition and natural language processing.
Can you describe in some detail how AI works by explaining your examples of AI applied to games like Chess and Go?
Figure: Artificial Intelligence now and then
Where does AI come from, and what are the reasons for its success?
Moujahid: AI as an academic discipline dates back to the 1950s. Since then the technology has gone through several hype cycles, some years with many promising advances and increasing enthusiasm for the technology and other years with less interest. The periods of reduced interest in AI are often referred to as AI winters. The recent rise of AI can be traced back to the mid-2000s with Geoffrey Hinton's groundbreaking work in Deep Learning. In addition to the algorithmic innovation of Deep Learning, the increase in computing capacity using GPUs and cloud resources, and the large datasets available to academia and enterprises are all factors that have contributed to the recent development of AI.
What are currently the best-known AI use case scenarios?
Understood. Now, think about System Integrator companies (SIs) that also implement AI functionalities for their customers. Is there anything that makes an AI-related project special?
So far, we talked mainly about the AI technology. Now, let me ask you: How important is Ethics in AI?
Moujahid: There are many ethical implications associated with AI. This issue has received a lot of attention in recent years with scandals such as Cambridge Analytica, accidents with self-driving cars, concerns about mass video surveillance and the displacement of jobs due to automation. Many countries and organizations are in the process of publishing their guidelines for dealing with the ethical implications of AI. For example, the European Commission has defined its Ethical Guidelines for Trusted AI with specific key requirements that AI systems must meet to be considered trustworthy. Such as human capacity and oversight, privacy and data management or accountability. To better understand why we need these requirements, let me give some concrete examples:
First, there is the case of a famous technology company that decided to develop an AI algorithm to sort job applications. They used a dataset from their HR system to train a Machine Learning algorithm that makes the final decision: hire or reject. The dataset contains historical resumes of applicants and the final decision made by the recruiters of the company. After testing the trained algorithm, the company noticed that the algorithm discriminates against a certain type of profiles for some specific jobs. When they analyzed the results, it turned out that the algorithm made similar decisions to some of the recruiters and that the algorithm had the bias of these recruiters. This shows the importance of the data used to train these algorithms. If the data is biased, then the algorithm is also biased.
If an AI-controlled car has to make ethical decisions in the event of an accident, which let's say in any case includes the injuring of people, how should it react in such a situation?
Moujahid: This is a very important question and it’s actually a variation of an important philosophical thought experiment called: The Trolley problem. I personally found a book by Michael J. Sandel called “Justice: What's the Right Thing to Do?” a great resource to understand and think about these types of problems and this question in particular. I’ll try to answer it using the ideas from the book.
Imagine our AI-controlled car is going 60 km/h and its brakes stopped working. In front of the car, there are five people and on the side, there is only one person. The car has two choices: Either continue going straight and for sure it will kill the five people. Or, turn the steering wheel and for sure it will kill the one person on the side. If you ask this question to a large group of people, you’ll get a majority taking the second option and using the reasoning of sacrificing the one person in order to save the five. This type of reasoning falls under a school of philosophy called: Utilitarianism. Utilitarianism is concerned by maximizing the happiness and well-being for the majority of people involved in a certain situation. The most famous utilitarian philosophers are Jeremy Bentham and John Stuart Mill. We can see in this example that by sacrificing the one person, we’re maximizing the well-being of all the people involved. Now, if we change the setup a little, we’ll start seeing the limits of utilitarianism. Let’s take the same example with the car and the five people, and instead of having one person on the side, we have the same person standing on the bridge on top of the road where we have the AI-controller car. Behind the person, you’re standing and you have two options: Either you do nothing and the five people die, or you push the person that is standing on the bridge, the person lands in front of the car and stops it from hitting the five people, but that person in question dies. If we apply the same utilitarian logic, sacrificing the person who’s standing on the bridge is the right moral action to take since it maximizes the well-being of the people involved by saving five people. But if you ask this question to most people, they won’t favor the action of throwing the person off the bridge and they’ll try to find different types of justifications.
Another school of philosophy that gives us another way of looking at the problem is Deontology with Immanuel Kant's theory being the most popular formulation of deontological ethics. A deontological approach to ethics judges the action under a series of rules, rather than based on the consequences of the action. Going back to our example, we could define a rule that states for example that no one can take any action to end another person’s life. If we take a deontological approach to the trolly problem using this rule, then we shouldn’t turn the steering wheel in the first case and we shouldn’t throw the person off the bridge. But if we have other rules, then of course the consequences will be different.
Now, going back to your original question, unfortunately there is no straightforward answer to it. We will see different countries deciding on different regulations and rules for AI-controlled cars. I think that these rules will vary based on the culture of the countries implementing them.
These types of thought experiments have been studied for centuries by philosophers and most of these experiments didn’t have any practical implications until now with the rise of AI and other advanced technologies. I think that it is very important to involve philosophers to help framing these questions in order to define the right regulations and laws that can govern these technologies.
An inspiring explanation based upon the given environment. Now, when we think about next-gen AI, how will AI continue to change our economy and society?
Moujahid: In general, it is very difficult to make predictions about technology. However, I think that we are still at a very early stage of AI and we will see some incredible advances and applications in the short to medium term. If we just take existing AI technology and we push the adoption in many areas of society, I could imagine in a few years having quite interesting areas of AI experiences. Let me name a few: What do you think about AI assistants that organize your day and execute different tasks like bookings on your behalf? Or a network of self-driving cars that you can use as your main means of transportation. This could dramatically increase mobility and reduce traffic congestion, isn’t it? These applications will become normal in everybody’s life. What about an AI healthcare assistant that has access to your healthcare information and your real time vital data collected from wearables? This AI assistant can then detect early signs of diseases and give you advice on how to improve your health. Consider also that all types of medical diagnosis in hospitals may be fully automated by AI then. Another example: Unattended and cashier free supermarkets and stores like Amazon Go will definitely become the norm and you will have personalized service in every store and restaurant you go to. We can also think about all of these examples even from a wider scale: Imagine that all cities and governments around the world implementing cameras and sensors that collect real time data about each citizen and using this data to either improve cities or for mass surveillance.
If we think for the long term, there are two areas of research that many AI specialists often refer to: The first is Artificial General Intelligence: The Machine Learning and Deep Learning algorithms that we’ve been discussing uses a dataset to learn how to execute one specific task, for example: Face Recognition. This type of AI is often referred to as “Narrow AI”. The goal of Artificial General Intelligence is to build algorithms that are capable of executing tasks that were not specifically trained for. This type of AI is often referred to as “Strong AI” and it’s closer to human intelligence. Then, think about Brain–machine interfaces: There are companies like Neuralink that are developing brain–machine interfaces with the goal of merging human brain functionalities with AI.
As far as the impact of AI on businesses and societies is concerned, as we have discussed, there are many positive aspects, but also some challenges that we need to address. I believe that most technologies (including AI) are essentially neutral, and it is the decisions we make as a society that determine the impact of these technologies. AI has the potential to contribute to the creation of more equal societies, where access to quality education, health care and other resources is guaranteed for all citizens. Nevertheless, we must ensure that we address the ethical implications of such technology.
If you were now able to talk in real time to an ancestor of yours who died a long time ago but is now represented by an AI bot in the cloud, what would you tell him or her?
Moujahid: I think that when we look at these types of applications, we need to separate Artificial Intelligence from Consciousness. If we consider Artificial Intelligence only, there already exist algorithms and solutions like GPT-3 that generate text and conversations that sound very realistic. If you combine these algorithms with chats or recordings of any person, you can build very convincing chatbots that sound like the person in question. Having said that, I think that having a conversation with an AI, even the most advanced ones wouldn’t be very fulfilling without consciousness. At this point in time, we’re very far from understanding how human consciousness works, let alone artificial consciousness.
That is awesome. In addition to these sources, is there a way for interested readers to engage directly with you?
Adil, thank you very much for this interview!