Artificial Intelligence – opportunity or threat?
Artificial Intelligence is today one of the top buzzwords in the tech world and it’s here to stay. But what is AI exactly? Gonçalo Martins Ribeiro founder of dDataMarket.com which enables organisations to handle Data Privacy explains
Back in the early years of AI, the fathers of this scientific field described it as “any task performed by a machine that, if a human carried out the same activity, we would say the human had to apply intelligence to accomplish it”.
This is a bit of a broad definition, but we can look at AI as systems or machines that are able to reproduce human behaviour, such as learning or interacting.
The AI concept has been known by the general public for a while and one only has to mention cult movies in which popular AI concepts are mirrored: Terminator, Ex Machina or Westworld.
Due to the fact that it is closely linked to popular culture and the legacy of science fiction, the term AI has caused some misconceptions, with unrealistic features and improbable expectations of how it will change the future of humanity.
Despite controversial media coverage, focusing on ethical and philosophical aspects of AI usage, AI has been around for decades but has only started having worldwide personal and daily applications over the past year. The controversy usually arises from two arguments:
● AI is going to replace us: Machines have been replacing humans since the Industrial Revolution, two hundred years ago. It is to be expected that humans would stop doing simple or repetitive tasks as technology evolves, and that is what is happening today. From mass production, to digitalization and now Artificial Intelligence, is it predictable to leverage our natural intelligence and have robots doing the work we don’t want to do. However, over the past decades, we have been taught that our purpose in the world is to be successful at our jobs and feel realized by accomplishing professional objectives. Now, we fear that AI takes that purpose from us. Having a robot doing our work for us should be an opportunity to start doing what makes us happy.
● AI is going to kill us: I’ve been participating in conferences and debates and there is always someone that ask the following question: “When a self-driving car crashes and it has to choose someone to kill, either it’s passenger or a pedestrian. Who is it going to choose?” For a start, this is not even a Machine Learning problem, because algorithms do not think or choose. If we want to tackle this question from an AI point-of-view, we should consider the following: AI responds to data and calculates probabilities; Data comes from the car and nearby sensors (assuming it is a self-driving car in a Smart City environment). In the last moments of the crash, the car will calculate probabilities, according to the data it receives, in order to avoid the crash. Every decision the car makes will be the highest probability to avoid the crash. That said, in two almost similar crashes, either the passenger or the pedestrian can die, because the car will be guided by the probability of not killing anyone. It is not about the algorithm. It is the data and the sensors. In this area, we should focus on building high-quality sensors for our cars and feed the algorithms with high volume quality data.
Having these in mind, we can now look at AI from a business perspective in order to understand its value.
Business challenges won’t change whether we use AI or not. Organisations exist to generate value for their shareholders and they will use whatever generates more revenue. It happens that AI is a solution for several use cases, across all industries and even within individual organisations.
According to Tractica, the global AI software market is expected to experience massive growth in the coming years, with revenues increasing from around $9.5Bn (€1.6Bn) in 2018 to an expected $118.6Bn (€103Bn) by 2025.
McKinsey estimates the AI related market has the potential to create between $3.5T (€3.07T) and $5.8T (€5.09Bn) in value annually across nine business functions in 19 industries. That’s a lot of money.
The business changes and challenges for companies could be:
1. Increased Revenue
2. Cost Reduction
3. Increased Efficiency
4. Legal Requirements
The challenges above can be addressed by developing a new product, changing an existing one, adding automation to a process and so on.
There are no specific challenges that AI fits best, nor one that could not apply to it at all. But there are some aspects that influence whether we should use it or not: tasks and processes that have to be done manually or are too complex to be executed by a human. AI will be used as the solution to a particular business challenge if:
1. It is difficult to directly code a solution
2. Difficult to scale a code based solution
3. There is Personal output
4. Functions change over time
Sample cases of such challenges might be classification problems: imagine pictures have to be classified; for example a cat picture or a dog picture. It would be easy for a human to do it for a small amount of images, but if we need it for millions of pictures, it would need a lot of time and it would be repetitive and boring. An AI approach would make sense in this scenario. The term itself, is composed by subfields, being Robotics the field we more often relate to. AI also comprises Computer Vision, Natural Language Processing (NLP), Machine Learning (ML) and others, some of these being more advanced than others.
Machine Learning is one of the most popular applications of AI, in which computers perform via cognition (very similar to the human brain). A more specific definition can be “Systems that improve their performance in a given task with more and more experience or data”. An ML solution could deliver the “Cat vs Dog” classification results in seconds.
Despite this short processing time, the learning process is quite different from a human: while we just need to see one cat or dog to learn what type of animal it is, an AI approach will need to analyse thousands of pictures of cats and dogs in order to properly classify new pictures with a decent accuracy.
Many organisations rely on classification problems, such as a CRM system to predict customer behaviour or recommendation engines. Industrial and delivery companies are focusing on predictive maintenance, which will allow them to save millions by executing maintenance at the right time, avoiding waste.
Effort is also being made in the field of Natural Language Processing, with the creation of AI-based chatbots for automated sales or customer support – think the Portuguese company Unbabel.
Technology is proving its value and the possibilities are endless. Organisations should start thinking in data-centric approaches instead or a normal software approaches and new possibilities will arise.
Some examples of other real world AI applications are:
1.Virtual Personal Assistants: Siri, Google Assistant and Alexa are some of the popular examples. They assist the caller in finding information via a voice command. All you need to do is interact with them, by asking something like: “What is my schedule for today?” or “What are the flights from Lisbon to London”. Building the response, the assistant searches out for the information, recalls your related questions, or uses other phone apps, to collect the information requested. Machine learning is an important part of these software applications as they collect and refine the information on the basis of your previous involvement with them. Later, this set of data is utilized to show results that are tailored to your preferences.
2.Social Media: From personalised advertising to face recognition in pictures, Social Media has many AI applications.
3.Online Customer Support: Most websites offer the possibility of chatting with someone from customer support. However, not every website has a live agent to answer your questions. In most of the cases, you talk to a chatbot. These bots tend to extract information from the website and present it to the customer. This is an example of Natural Language Processing techniques.
4.Recommendations: This one is more common because it has been around for a while and it has been evolving with Machine Learning. When you shop online you receive recommendations, suggestions and later on marketing emails based on your preferences. ML refines the shopping experience based on your behaviour in the website or app, past purchases, liked items, brand preferences, etc.
5. Fraud Detection: Machine Learning is being used to compare transactions and to distinguish between legitimate or illegitimate. This can prevent frauds or detect money laundering. Paypal is using this technique to compare transactions taking place between the buyers and sellers.
There are many real world problems that can be solved with AI. For a specific business problem, if it fits one of the mentioned categories, we should start addressing it as an AI or Machine Learning problem, which will lead to further questions:
● How much data is sufficient for building successful ML models?
● How to deal with data quality issues?
At this stage, organisations need to be prepared to deal will data, both in quantity and quality. It seems common sense that implementing AI has only 2 steps: collect data + train model; but in reality it is much more complex than that.
There are plenty of data preparation actions prior to model building and most Data Scientists consider that they spend around 70% to 80% of their time working on the data itself than building the AI model.
Furthermore, Data Scientists and Machine Learning Engineers consider that the number one factor in their failure when implementing AI is the lack of high quality data, not to mention that data privacy concerns most organisations, because of the risk of data leakage and the impossibility of sharing data with third parties (such as consultancy firms or others working on the solution implementation) because of legal impediments.
At dDataMarket we enable organisations to use AI, by providing them with the tools and the data to train Machine Learning models efficiently. We have an ecosystem of AI-as-a-Service (AIaaS) and Data-as-a-Service (DaaS) tools that gathers data from multiple data sources, from open to proprietary and from our partnerships. We enable the creation of a 360º view, allowing organisations to easily leverage on high quality ready-to-use datasets to build AI based solutions with a world perception more close to our own.
We strongly believe that the conscious use of data must start with the data curators and, as a DaaS provider, we designed our tools and even the resulting datasets, in a secure way, such that the use of private data is no longer an issue. Through the application of AI techniques, we are able to not only anonymize the data but also to create new synthetic data similar to the original one, that besides guaranteeing users privacy, it’s not proprietary.
Data privacy will be, if it is not already, the new buzzwords of the tech scene. The tech industry will be responsible for ensuring that personal or sensitive information is private because regulation only won’t be enough.
Text: Gonçalo Martins Ribeiro