The short answer to this question is quite easy: it is practically impossible to predict the future of machine learning; one of the most dynamic, complex, and challenging fields mankind has ever created! The history of Machine Learning is tightly linked to the history of Artificial Intelligence and teaches us that its evolution was marked by ups and downs, periods of high interest and hype followed (usually in a rather unexpected way) by periods of “oblivion” – the so-called “AI winters”.
Future of Machine Learning
While its future is impossible to be accurately predicted, most of the trends and developments that have a serious impact on Machine Learning today can be identified and analyzed. One might even dare to assume these trends and developments will play a major role in shaping its future. For me, the best way to describe the future of Machine Learning is to use some of Sci-Fi cinematography’s most famous words: “Clouded this boy’s future is”. On one hand, it’s a fair bet to assume that the raw power of cloud computing will continue to play an increasingly important role in the development of the field. On the other hand, it’s also fair to state that there is some level of uncertainty linked to the directions of future developments. As it happened throughout the history of humanity with any powerful asset, Machine Learning has its own pitfalls, negative aspects, and even potential dangers.
As it happened throughout the history of humanity with any powerful asset, Machine Learning has its own pitfalls, negative aspects, and even potential dangers.
Exploring this dual development potential is at the core of my take on the future of Machine Learning. The promises and positive impact are the reason why this field is extremely important to each one of us, whether we understand the technology or not. Materializing these promises and benefiting from the positive impact cannot be done without acknowledging, understanding, and pro-actively trying to address the negative aspects. This part is the responsibility of all of us involved in advancing the field and putting it to work.
I am an Artificial Intelligence and Machine Learning enthusiast from tip to toes. I truly believe they will continue to change our lives for decades to come in ways that we cannot even imagine today. The Star Wars fan that I am is convinced that Machine Learning will be successful in avoiding to slip towards the dark side. All my technical talks about Machine Learning use this assumption as a foundation and contain countless examples of the great things we can do with it. Yet, in the broader discussions (this one included) you will hear me talk more about the issues, the challenges, and the dangers rather than about the good things. This is simply because I am convinced we need to keep a keen eye on them to make sure things do not get out of hands. History has countless examples of how quickly powerful stuff can become dangerous or even catastrophic.
How can Machine Learning help?
Quite often I am asked what is the most important thing Machine Learning does or will do for us in a reasonable future. While there are so many of them, my choice is always healthcare. The mathematician in me believes that cures for the toughest diseases we face will be either calculated or predicted. Machine Learning already plays a fundamental part in workloads like predicting response to therapy or decoding various connections that exist in the human body at the most fine-grained levels. An example I love to mention is Microsoft Research’s 2018 project named Machine Learning for Cancer Immunotherapy. It’s one of the countless examples of powerful tech giants working together with academia and healthcare providers to put Machine Learning to work in solving extremely difficult problems.
In the not too distant future, we will also talk about how our smart homes and connected cars understand and adapt to our behavior and habits.
Another area where Machine Learning is generating giant leaps forward is our individual interaction with the digital world we are living in. Better and finer-tuned personalization is already here. Think about your online shopping experiences or the recommendations various platforms provide. In the not too distant future, we will also talk about how our smart homes and connected cars understand and adapt to our behavior and habits. In the same context, Machine Learning will continue to drive huge improvements to the ways we search for and find information in virtually any field of knowledge. The other major field we should mention here is security where Machine Learning is used in the frontlines of battling against cyberattacks and a whole range of other types of security-related threats.
Overall, the potential is immense, and it’s being recognized by everybody. In a 2019 article, Forrester stated: “The future of Machine Learning is Unstoppable”. Gartner predicts “AI will be in almost every new software product by 2020”. And the list can go on and on. The reality is that Machine Learning is empowering business, enabling the creation of new services and even completely new markets. And, most probably, we’ve barely scratched the surface. The spectacular increase in computing power and the transition from software-based to hardware-based implementations are probably going to fuel this development even more. When Quantum Computing becomes a commercially viable thing, we’re going to need to completely reset our plans and expectations. But that’s a topic for another interesting discussion 😊
In a 2019 article, Forrester stated: “The future of Machine Learning is Unstoppable”. Gartner predicts “AI will be in almost every new software product by 2020”
Is Machine Learning Dangerous?
At the same time, we’re starting to be more and more aware of the dangers and pitfalls of this immensely powerful field that is Machine Learning. Yoshua Bengio, acknowledged as one of the three founders of deep learning and winner of the 2019 Turing award, says about it “the dangers of abuse are very real”. He is one of the drivers of the 2019 Montreal Declaration on Responsible AI, a collection of foundational principles that should govern the future development of Artificial Intelligence in general and Machine Learning in particular. Indirectly, the declaration is a comprehensive list of potential pitfalls and dangers we could be facing in the near future.
Indeed, the challenges we face in the field are monumental. They range from lack of skills to evaluate, build, and deploy machine learning models to very practical and immediate issues like increased inequality, security breaches, bias, or adversarial attacks. The list can be completed with military applications (think of killer drones), deep fakes (think of personalized text-to-voice or even videos), and intentional data poisoning or biasing (think of public interest vs. profits in healthcare). Another area of concern is the data itself which in many cases is riddled with problems of quality, quantity, homogeneousness, poisoning, and compliance (think limitations for privacy reasons).
From my point of view though, the biggest threat to Machine Learning is excessive hype, especially marketing hype which managed to put it on a dangerous pedestal. And when supersized and over-hyped expectations fall short of reality, the pendulum can quickly swing to the other extreme. We’ve already seen this at least two times in the short history of computer science and we call these events AI winters.
From my point of view though, the biggest threat to Machine Learning is excessive hype, especially marketing hype which managed to put it on a dangerous pedestal.
I hope this short introduction managed to spark your interest in continuing the discussion about the future of Machine Learning. If you want to hear more about this fascinating topic, make sure you don’t miss my session dedicated to it at the Microsoft Azure + AI Conference this November. Looking forward to seeing you in Las Vegas this autumn!