Chapter 8. The future of the intelligent web
- Summary and review
- Future applications of the intelligent web
- Social implications of the intelligent web
In this book, we’ve presented to you the current state of the intelligent web and walked you through the basics, teaching you what an intelligent algorithm is and how to evaluate one. Your eyes should now be open to the plethora of intelligent algorithms you’re interacting with on a daily basis! You’ve learned about design considerations and the key pitfalls you should try to avoid in your work.
The current philosophy of machine learning is that “more data beats a cleverer algorithm.”[1] As you’ve seen in the case of deep learning, it’s been possible to make great advances due to the changing nature of computing and the internet. So much information is available on the web that channeling and accessing it appropriately becomes of utmost importance. This is the focus of the appendix. Although it may seem removed from what many consider an intelligent algorithm, we can’t stress enough the importance to the enthusiast of having at least a passing interest in this area. Future practitioners in this space will absolutely need to know the magnitude, velocity, and availability of their real-time data.
Pedro Domingos, “A Few Useful Things to Know about Machine Learning,” Communications of the ACM 55, no. 10 (October 2012): 78–87.
We’ve covered what we consider some of the most pervasive topics in intelligent algorithms: extracting structure; providing recommendations; classification, click prediction, and deep learning; as well as selection and testing. These topics aren’t completely unrelated, and we’ve drawn parallels between concepts wherever possible. Think of these as design patterns—or blueprints for a particular solution. If in the future you’re faced with something that feels like a recommendation problem, refer to chapter 3. Or if you need to write a system to choose among several options, refresh your memory with chapter 7.
Remember, though, that this book is by no means comprehensive. Intelligent algorithms are a huge topic covering several related fields, and it would be foolish of us to attempt to cram all of this knowledge into these pages! Instead, we hope you put down this book with a flavor of how you might approach problems with an intelligent algorithm in mind and with a feel for what has been done before. The good news is that that this really is still a fledgling area. There are many application-specific problems to be solved and lots of well-researched techniques to do so. As a practitioner in the area, you’ll have a responsibility to connect the dots, be aware of the limitations, and find a path forward. Good luck!
8.1. Future applications of the intelligent web
We started this book with a an outline of how a real intelligent web application might work—the Google Now product—so it seems fitting that we draw a line from here into the future and highlight some of the application areas we feel will provide fertile ground for the creation of new algorithms. Some of the applications are on the cusp of mainstream, but others are some way off. We’ll leave it to you to decide what is science fact and what is science fiction.
8.1.1. The internet of things
The internet of things is an umbrella term for a new wave of computing whereby all devices are online and attached to the internet. It’s a realization of Marc Weiser’s vision of ubiquitous computing,[2] whereby computing power has shrunk to the level that it has disappeared into the background of our daily lives. The internet of things has largely remained conceptual to date because the path forward is strewn with issues around standardization of communication and security. Ultimately, you might argue that we have seen some progress in this area through consumer products in home automation,[3] but what about a fully connected house that does the shopping for you, makes appointments, interacts with your smartphone, starts dinner, and activates the washing machine? This appears to be some way off yet. Many intelligent algorithms would need to be designed to realize this vision, and thus there are many likely avenues of work to pursue.
Marc Weiser, “The Computer for the 21st Century,” Scientific American, September 1, 1991: 66–75.
Nest Thermostat, 01/01/2015, https://nest.com/thermostat/meet-nest-thermostat/.
8.1.2. Home healthcare
Extending the idea of the self-aware home, what if we went a step further and built a home that was interested in the health of its occupants? This might be especially useful for the elderly, the infirm, or people on home release from the hospital. Such a house could monitor general behavior and activity trends[4], [5] or provide periodic monitoring of vital signs of occupants. This would move us from a world in which medical practitioners see only a snapshot of your health to providing a more holistic view of your health status (perhaps available only on a private network). Ultimately, it could give many increased security and independence away from the hospital environment. In such a world, algorithms would need to be developed to understand and detect acute incidents or abnormal vital signs. Such algorithms would need to have a low false-negative rate and be aware of the privacy concerns of the monitored individuals.
Douglas McIlwraith, “Wearable and Ambient Sensor Fusion for the Characterisation of Human Motion,” International Conference on Intelligent Robots and Systems (IROS) (IEEE, 2010): 505–510.
Julien Pansiot, Danail Stoyanov, Douglas McIlwraith, Benny Lo, and Guang-Zhong Yang, “Ambient and Wearable Sensor Fusion for Activity Recognition in Healthcare Monitoring Systems,” IFMBE proc. of the 4th International Workshop on Wearable and Implantable Body Sensor Networks (IFMBE, 2007): 208–212.
8.1.3. The self-driving vehicle
As most readers will know, Google’s self-driving car has been in development for some time now.[6] In the future, you might imagine a totally distributed taxi network that seeks to algorithmically maximize the throughput of customers and minimize their journey times. What if we went a step further and imagined that all road vehicles were autonomously controlled and connected to the same network? Provided the right security measures were met, this would help maximize the flow of traffic, reduce traffic incidents, and increase safety—all through the use of intelligent algorithms. You might argue that we’re already moving toward such a world, because several insurers are providing reduced (or more appropriate) insurance premiums through the use of data-monitoring boxes in the insured cars.[7], [8]
Google Self Driving Car Project, https://www.google.com/selfdrivingcar/.
Adam Tanner, “Data Monitoring Saves Some People Money on Car Insurance, But Some Will Pay More,” Forbes, August 14, 2013, http://mng.bz/PWgf.
Leo Mirani, “Car Insurance Companies Want to Track Your Every Move—And You’re Going to Let Them,” Quartz, July 9, 2014, http://mng.bz/1elQ.
8.1.4. Personalized physical advertising
Several times in this book we’ve talked about the personalization of online advertising. What if this personalization were to jump off the screen and into real life? You may have already seen an illustration of this in the film Minority Report,[9] where personalized ads are presented physically as people interact with their environment. Farfetched, you think? Well, in 2013, in the city of London, smart refuse bins fitted with advertising screens were found to be using an open Wi-Fi network to count footfall—by counting the number of unique MAC addresses that had connected to its network. In this way, the advertiser could determine whether the same person had walked past multiple times and potentially tailor ads to them.[10] The practice was stopped almost immediately,[11] but it does provide an interesting proof of concept and food for (ethical) thought.
Andrew Orlowski, “Facebook Brings Creepy ‘Minority Report’-Style Ads One Step Closer,” The Register, November 09, 2015, http://mng.bz/ML01.
Siraj Datoo, “This Recycling Bin Is Following You,” Quartz, August 08, 2013, http://mng.bz/UUbt.
Joe Miller, “City of London Calls Halt to Smartphone Tracking Bins,” BBC News, August 12, 2013, http://mng.bz/k56c.
Solutions in this application space might migrate away from localized pods and onto the web, bridging the gap between the digital and the physical. Click prediction might then be based on other factors, such as your movement in the physical world or whether you had seen an ad on your morning stroll to work.
8.1.5. The semantic web
The World Wide Web contains a vast amount of information, but the way in which we interact with it is, for the most part, still rather primitive. We rely heavily on one of several search engines to make sense of our requests (keywords) and find us pages or documents that are relevant to the search. This model works but is far removed from the natural sorts of interactions that we enjoy with our family, friends, and colleagues.[12] Even simple relationships between objects (for example, a cat is a type of animal) may not be encoded, so keyword searches can’t use this implicit information that we as humans have available at the time of the search.
Although as we have seen, this is changing rapidly—consider Google Now and Apple’s Siri. www.apple.com/uk/ios/siri/.
It would be a much richer experience if we could ask questions of the web, and, through a series of relationships, an answer could be derived by deduction or a table of information returned or an action performed. This is the vision of the semantic web, a term coined by Tim Berners-Lee et al. in 2001.[13] Their vision of the semantic web relies on knowledge being encoded into web resources through a markup language and the existence of a set of ontologies to bind facts together and make them more easily manipulated by logic. Agents operate over the semantic web, extracting information and performing actions based on a set of deductions or proofs. One of the key benefits of such an approach is that the deductions reached are explainable by the agent in a language that is close to our own. Should a user disagree with the information returned by an agent, the deductive rules and ontologies used can be returned so the user can inspect the logical steps in deduction.
Tim Berners-Lee, James Hendler, and Ora Lassila, “The Semantic Web,” Scientific American, May 1, 2001: 34–43.
We might be some way from the fluid and natural user experience with the web as described in the opening example of the aforementioned paper, but much research toward this is finding application in industry. What we can be sure of is that any attempt to provide richer semantics to the vast amount of information available on the web can only be a good thing for designers of intelligent algorithms, allowing us access to knowledge as opposed to just data.
8.2. Social implications of the intelligent web
In the previous sections, we’ve provided several potential areas for the development of intelligent applications. Whether our visions will come to pass is up for debate, but one thing is for sure: the technology exists or is being created to make these visions a reality. The major question is one of adoption and legislation. Just because something can be done, that doesn’t mean it should be. There are many social implications to consider before leaping headfirst into such a future.
The underlying concerns regarding most intelligent algorithms seem to be focused on privacy and security. Users have the right to expect a degree of privacy with respect to their actions both online and offline. Similarly, users have a reasonable expectation that data held about them is secure and safe from malicious intent. Thus, designers of intelligent applications should respect these requirements and neglect them at their peril! Ultimately, we believe the question boils down to utility. Users are willing to give away some of their rights to control their data if they’re provided with some measurable benefit that outweighs the risk and are reasonably assured that it’s in safe hands.
Consider the adoption of mobile phones as a case in point. These devices allow your movements to be tracked precisely and in real time across the globe, with this data held by a company that, on paper, has no interest in the specifics of your movements. In return, you’re provided with the ability to communicate with similarly connected users at the touch of a button. Here we see a trade-off in the extreme: detailed movement information for access to a killer application. We believe the adoption was so swift because of the rewards associated with instant real-time communication. We predict that we’ll witness even more extreme examples of detailed and personal information being given freely, provided the intelligent application that it powers helps in the lives of the users it serves.
We believe that we are now at a crossroad for the intelligent web. Users are more informed than ever before and demand to know how their data is being used. It’s our job to provide meaningful advances to their experience in line with their privacy and security expectations. Thus, developers of intelligent algorithms hold much power and are obliged to wield it responsibly.