Reposted from The Huffington Post
In Stuart Russell and Peter Norvig’s seminal book, Artificial Intelligence: a Modern Approach, the two authors wind up their work with a chapter looking at the future of artificial intelligence (AI). Their book is still the text of choice for teaching AI at many universities and so, I thought, reviewing the predictions they made 20 years ago could help guide us to make better predictions now, for the next 20 years of AI.
Upon rereading the book, though, their predictions from 1995 felt surprisingly salient and topical. There must be something wrong! How could a chapter on the future of AI written 20 years ago seem so similar to the many articles predicting the future of AI being written today? Have we not made any progress worth a mention in the past two decades?
Certainly, things are different today. There is much more enthusiasm and investment going into AI today than at any other time in the past. Maybe Russell and Norvig’s predictions were premature, but they are more imminent today than ever before.
So I can’t help but wonder whether most predictions being made today will still be topical in 2035. Even as recently as 2007, David Orrell failed to mention AI even once in his book The Future of Everything. Predicting the future of technology is indeed a dangerous game!
But let’s take a crack at it anyway, starting with the easier to predict and more immediate and incremental improvements, and then looking further out.
In The Near Term (In the Next 5 Years)
Prediction 1 – Improved speech, voice, image and video recognition will change the way we interact with our devices
Over the next few years, we will continue to see vast improvements in the quality and fidelity of speech, voice, image, and video recognition, and our ability to classify results will improve significantly. Cheap and omnipresent sensors and cameras will provide ever-increasing streams of data for processing in real-time. This real-time requirement, paired with cheap and available processing power and storage, will make it much more cost-effective and efficient to process the data at the point of collection and, eventually, to learn and act upon the data locally. We will see these systems widely adopted in industrial automation systems, factory operations, security systems, agriculture, traffic and transportation, and many other domains.
Prediction 2 – Personal assistants will become more personal and more context-aware
Personal assistants will become more acceptable to us as they become more personalized to our needs and more able to understand the context of our requests, which in turn will enable them to broker an ever wider range of capabilities. Whether conversationally-driven AI assistants end up completely displacing more traditional GUI interfaces to our daily activities remains a question in my mind; I don’t know if these improvements alone are enough. After all, natural language is (by definition) natural for all of us, but, as humans, we did pick this language skill up much later than the ability to point and shoot. Having said this, even for GUI interfaces I believe that intelligent context-reactive visual interfaces will take over, seamlessly augmenting conversational interactions.
Beyond current command-and-control style personal assistant systems, improvements in conversational systems will be the catalyst to finally bring robots into general use as household items.
Prediction 3 – More and more systems will run autonomously…to a point.
Autonomous systems are already more commonplace than most people think – think about how surprised you are when you actually get a live person on the phone when calling a support line. Each time you fly much of the journey is run by machine, not the pilot. Self-driving cars and autonomous drones seem inevitable, and we’ll see autonomous decision-making systems move well beyond transportation. But our fears about giving machines too much power will limit the use of autonomy, at least another 10 years. So I am not so sure that all future transportation will become driverless. It is only natural that humans will want to retain control.
Prediction 4 – AI will become much more widely adopted by enterprises
AI now feels like computing did 30 years ago. The advent of PCs created an explosion in the utility value of computers to classes of businesses beyond the data processing companies, and more companies started establishing computer divisions. Those divisions are now long gone, with computing having become a ubiquitous part of our work, not to mention our everyday lives. AI is being, and will continue to be, quietly adopted by enterprises, allowing them to extract knowledge from all the data that is being generated – and not just the structured data.
AI will continue to move towards taking on decision-making tasks. Automated fleet management, inventory management, and candidate resume screening are but a few examples. However, to conclude that AI will displace humans in all decision-making, is like concluding, millions of years ago when humans started using tools, that the day will come soon that they will no longer have a need for limbs.
Prediction 5 – The positive impact AI research can have on humanity will start to be felt across many walks of life – much of it behind the scenes.
Each stride forward in core AI research is opening up our abilities to solve new classes and scales of problems, which in turn enables the acceleration of research in almost every scientific domain, to the betterment of humanity.
Companies like Cloud Pharmaceuticals and Berg are using AI to achieve breakthroughs in drug discovery, with the ability to discover and bring highly targeted therapies to market in record time. Still other companies are using AI to invent new and better methods for disease prediction and treatment, to make cleaner and more efficient chemical engineering systems, to super-optimize fleet management and support ticket analysis, and to guide better investing, content optimization and home energy optimization.
While of course we’ll continue to see consumer-facing “AIs” more and more often over the next 5 years from Siri to Amy, Jibo, and beyond, the broadest impact AI will have to our lives – in the nearer-term at least – will be through its role as a force multiplier for almost every other field of human endeavor.
In The Medium Term (5 to 20 years out)
Prediction 6 – General purpose learning systems will drive the future of education and entertainment.
Progress in general-purpose learning systems – AIs that can learn on their own, whether on a supervised or even unsupervised basis – will lead to successful deployments in a range of specialized application areas. AI will grow beyond its role as curator and analyzer of content and become much more important in generating and augmenting content in the first place. These types of systems could be used in education: imagine a teacher that is learning alongside the student.
This same kind of technology has the potential to generate characters and plots for movies as well as immersive interactive virtual-reality games.
Prediction 7 – The rise of personal hypothesis systems will improve the quality, and possibly the length, of our lives
The use of AI to solve specific healthcare problems, small and large, is already well underway. Fast forward and we will start to see hyper-personalized hypothesis generation systems, which will operate on our background data like our genomics, paired with measurements from our wearables and other biological monitors, to provide each of us and our doctors a highly accurate lens – and a crystal ball – offering valuable insights into environmental and behavioral impacts on our health.
AI will also be used to interpret human brain activity in a way that can decipher intent, enabling augmentation to overcome physical challenges and new methods of communication for and with disabled patients.
Prediction 8 – The Intelligent Internet will become more prominent
There’s much talk about the coming creation of an intelligent Internet. The first steps towards such a networked intelligent system have already been taken. Siri, for example, is not an app that you use in isolation. Every time you use Siri, you are connecting to a single centralized AI system working on your behalf to provide you with the services your request. With AI moving to control more devices and sources of content, collaboration amongst these semi-autonomous AI agents will drive great benefit.
AI will impact designers and programmers too, automating much of the processes involved, mapping their desires, explicitly communicated or even implied, to achieve creations that fulfill those requirements. In parallel, this will result in increased satisfaction from the people that interact with these AI-automated designs/programs, creating surprise and delight by continuously morphing the design or program as the system takes into account learnings from interactions with other users.
Prediction 9 – Breakthroughs in distributed compute will allow AI to realize its full potential
The new uses of AI will demand processing speeds well in excess of what a single CPU can provide. AI is already being distributed over multiple machines within data centers and across data centers, with the algorithms capable of running in parallel over many hundreds of Graphical Processing Units (GPUs) on each computer. The ability to make use of this scale of compute capacity is nascent, however, and more breakthroughs in computing are needed to harness the full potential of the available processing power. New approaches, such as quantum computing extensions, are being researched today and show promise.
Prediction 10 – Like any successful technology, AI will be used for purposes other than good
The technology press has frequent outbreaks of dialogue about killer robots, and whether military uses for AI will be the saving or the end of humanity. AI will be neither savior nor destroyer, but like every major advance before it, ill-intended uses of AI are certainly possible. AI is no more or less dangerous than any other one of humanity’s inventions, and so far, the verdict on human technology has been pretty positive.
So what will we think of the list above in 20 years? I’m betting that we will still be hard at work on many of the things on the list, and still not quite there on many of them. In 20 years time, just as we have in the past 20 years, we will have made much progress on all fronts, increasingly using AI in our everyday life and work, and, for the most part, taking its ubiquity for granted.