In the first of three articles on artificial intelligence and the impact of transformative technologies, Ed van der Sande, Managing Partner Odgers Berndtson Amsterdam and Head of the Tech & Services Practice, interviewed Maarten de Rijke, a Professor of Machine Technology at the University of Amsterdam. How exactly does AI learn and what it can already do better than us?
Trending AI Articles:
1. Deep Learning Book Notes, Chapter 1
2. Deep Learning Book Notes, Chapter 2
3. Machines Demonstrate Self-Awareness
4. Visual Music & Machine Learning Workshop for Kids
Advances in machine learning are predicted to radically shake up the employment landscape in the coming decades. Technological innovations and automation have displaced workers in the past, but the so-called ‘fourth industrial revolution’ will see rapid breakthroughs in robotics, AI and related fields. All with a massive impact. PwC estimates that will displace around 7 million workers in the next 20 years in the UK alone. But it’s not all bad news. PwC also projects that the same technological advancements will create 7.2 million new jobs over the same time period, generating large productivity gains for society and increasing consumer spending power.
So what is AI and how does it work? Professor de Rijke dove deep into this transformative technology to demystify a topic that’s never far from the headlines.
How does AI learn?
Learning is fundamental to AI development. There are three kinds of learning that are of importance. Supervised learning consists of someone showing the machine how something should be done. Unsupervised learning, the second type, consists of pointing the machine to a data set and telling it, “go discover something and then return to me and tell me what you have found.”
The third type, reinforced learning, occurs when the machine is incentivized to try different solutions until it reaches the correct one. Machines can act and observe the effects of those actions to determine if its action was right or successful. If the action was unsuccessful, the machine can be taught to act differently in the future. All of this can be done better now because we have more computing power, more data and more algorithms.
Is that how search engines work?
Search engines have been around for over 50 years. At first, search functions were carried out manually, but since the 1980’s search engines have in large part relied on machine learning. Machine learning is the practice of using algorithms to analyse data, learn from it and then draw conclusions or make predictions about something in the world.
Importantly, a web search engine is not one single operation; it’s tens of different processes (geographical, encyclopaedic, product-related, visual, etc.) happening together. When a search query is processed, the technology needs to decide to which search engine that query should be sent.
This is where reinforced machine learning comes into play because the technology can improve its processes based on our responses to the search results. The search engine is not simply displaying the results; it is assessing their usefulness based on our behaviour. The AI component thinks, “when the machine displayed those results, the visitor left, so that didn’t work. I’ll have the machine try something else next time.” Over time, after much trial and error, AI can figure out the optimal combinations of search results to display based on the user’s query.
Is AI capable of ‘deep learning’?
Deep learning is driving the explosion in AI capability today. When there is no need to teach a machine what the characteristics of something are because it can teach itself through trial and error by training itself on massive amounts of data, that’s deep learning.
Deep learning is a set of calculation structures inspired by how we once thought the human brain operated. Since then, computer scientists have adjusted the process in lots of ways that are not necessarily anchored in biology anymore.
Can a machine produce, for example, a better video game than a human?
Definitely. When you look at the new video games coming out now, there is no dictating anymore what the storyline is. Only the conditions are created. Through reinforced learning — observing the actions and responses of human players — AI can effectively create a game that players want to keep playing.
How does AI utilise knowledge?
AI is much more than machine learning through trial and error. AI also envelops knowledge (certain and uncertain) and reasoning. Searching a database for information is utilising ‘certain knowledge’, but AI can also be used to generate uncertain knowledge such as estimating someone’s expertise on a particular topic.
In addition, AI can solve problems, learn, communicate, adapt to new situations and observe the effects of its actions. Think about Watson, the computer that won Jeopardy!, Deep Blue, the computer that defeated a world champion chess player, and Google DeepMind’s AlphaGo program, which defeated a South Korean master in the board game Go.
AI is not science fiction anymore — it is everywhere.
What can AI do better than humans?
AI can interpret large amounts of data and images more quickly and accurately than people can. For example, until recently, radiologists would assess brain scans to look for Alzheimer disease and other neurological disorders. Machines can now do this better than humans.
The growth of AI stems from a substantial growth in computing power over the last couple of years and an enormous growth in data. We no longer need to tell a machine what to do, we can tell it: “here is the data, go find out how to do it.”
As a result, all of the following are possible:
- Amazon Go is a brick and mortar store without the staff that can see into your shopping cart and determine how much money to deduct from your credit card.
- Netflix learns your taste and can give you personal recommendations pulled from a wide variety of movies and TV shows.
- TomTom can quickly make maps from a visual material.
- Tesla cars can recognize what is happening around them on the road.
- Baidu only needs to hear a few seconds of a voice to be able to recreate that voice perfectly.
What else can AI do that people have always done in the past?
Computers can listen in on meetings and offer relevant content from the archives that attendees will find useful. A computer can help journalists with the writing process by generating text and information as they type. It’s only a matter of time before computers will be able to provide live commentary on sports games.
Thank you, Maarten. It’s remarkable how quickly we have become reliant on speedy search engines, recommendation systems like the one employed by Netflix, and even virtual assistants like Amazon’s Alexa. All of these tools are having a massive impact on our productivity and our quality of life. Thank you for explaining how this technology is evolving and sharing some of the ways AI and machine learning are changing our world.
In the next two articles in this series, Ed van der Sande interviews Elfried Klarenbeek, co-founder of Harver and an independent digital innovation strategist, on exponential technologies, the power of innovation transformation and the role of leadership.