Internet Policy in the Age of Machine Learning

Evolving complexity in Telecommunications and Internet Protocols

By Nishanth Sastry and Mischa Dohler

IEEE Internet Policy Newsletter, December 2018

The internet was built on a philosophy of simplicity^[1]. This set the base for the growth of the internet as a network of networks operated by mutually independent domains with potentially different underlying technologies that enter into a business relationship with each other to provide mutual interconnection to each other’s networks, and beyond.

However, in recent years, there has been a tremendous increase in complexity. For instance, much of today’s internet traffic consists of video streaming, which needs support for a good amount of bandwidth and a low variation or “jitter” in the bandwidth over time. This is difficult to guarantee over the so-called “Best Effort” internet. To get around this issue, massive amounts of network data and traffic measurements are collected to try and ensure the best possible consumer experience. Content Delivery Networks (CDNs) have sprung up to do this job and have been extremely successful, but the problem is by no means solved. Exciting new research is emerging that uses deep neural networks to help decide the best rate^[2] and best servers for each client.

This rise in complexity is reflected in other areas as well. For instance, middle boxes—specialized nodes such as Network Address Translators, Load Balancers, Firewalls and Intrusion Detection Systems—have seen a sharp rise in growth, to the point that they are on par with routers and switches in terms of numbers^[3]. Middleboxes such as load balancers change the last hops near servers; network address translators can make multiple clients appear as the same host; firewall rules can grant access on some ports for some protocols while at the same time preventing reachability over another port for another protocol. Thus, the increased use of middleboxes has made it difficult to reason about how paths are chosen through the internet. If a certain destination is not reachable, it becomes very difficult to debug and pinpoint the cause of the issue.

The Role of Machine Learning and AI in the Future Internet

Examples such as the above have given rise to the idea of using predictive and intelligent oversight approaches to deal with the rise in the complexity. This trend has been fostered by the growth of available data, as well as increased availability of telemetry and network analytics tools to process such data. However, beyond basic analytics such as performance dashboards, predictive approaches typically involve machine learning in one form or another.

One of the broadest applications of Artificial Intelligence (AI) into networks is seen in Clark et al.’s bold vision of the Knowledge Plane^[4]. The Knowledge Plane calls for a network that can reason about itself, and aims to develop a network architecture that is sufficiently self-aware to be able to identify why a problem occurs when one occurs. For instance, if a user is not able to connect to www.ieee.org, the network should be able to tell whether the problem is in the first hop (e.g., the user is not authenticated to a captive portal, or the broadband provider has cut off network connectivity because a bill was not paid), or whether the problem is within the network (e.g., a routing misconfiguration is causing packets to go in a loop). A holy grail for a network to have various self-properties, such as being self-configuring (choosing appropriate values for different system parameters), self-healing (fixing errors by itself), self-optimizing (adjusting system parameters automatically to achieve best performance), and self-protecting (guarding against viruses and other attacks without external intervention).

Impact on Internet Policy

The Knowledge Plane and other proposals for self-healing or cognitive networks have not yet been realized in their fullest forms. However, many companies are developing cognitive or intelligent solutions as features or products, leading to the adoption of the principles of the Knowledge Plane in various different forms. Crucially, such aspects are important differentiating factors and innovations for competing products. Thus, many developments are happening without full oversight of standards bodies and other neutral entities. Given the growing worries about AI taking control of many aspects of human life (prominent examples include Elon Musk’s warnings^[5], Stephen Hawking’s ambivalence^[6], etc.), it is timely to ask if specific policies on this topic are needed to ensure technology is put to its best use and provide society with a means to manage the complexity of providing universal interconnection.

We believe that in the context of the internet, three issues have to be settled in relation to the use of AI and machine learning: accountability, privacy, and universal access. Advanced machine learning techniques can be used to infer faults and ensure relationships are governed by strong Service Level Agreements (SLAs). Going forward, it may even be possible to build predictive models of potential failures, and then use these to fix such errors.

Ensuring accountability may require sharing information, as well as collecting additional data, which leads to the concerns about privacy. There may be potential misuses of data, and the effects of these should be considered in a holistic manner together with the potential benefits before using additional data. Similarly, upcoming legislation such as the General Data Protection Regulation (GDPR) in the EU has already started to look at this and will provide fundamental protections^[7] to individuals regarding their personal data.

Finally, the potential for building global AI-based monitoring mechanisms, even if it be for well-intentioned reasons such as ensuring accountability, also showcases the power imbalances inherent in this complex ecosystem of stakeholders. In many countries, the internet currently functions on an implicit assumption of equal access, and it is important that such principles are taken into account in building a new “AI-enhanced” internet.

References:

Internet Policy in the Age of Machine Learning
A longer and more detailed version of this article has been published as a white paper by the IEEE Internet Initiative at: https://internetinitiative.ieee.org/images/files/resources/white_papers/IEEEII_WP_InternetPolicy_Nov2018.pdf.

^[1] Clark, David. “The design philosophy of the DARPA Internet protocols.” ACM SIGCOMM Computer Communication Review18, no. 4 (1988): 106-114.

^[2] Mao, Hongzi, Ravi Netravali, and Mohammad Alizadeh. “Neural Adaptive Video Streaming with Pensieve.” In Proceedings of the Conference of the ACM Special Interest Group on Data Communication, pp. 197-210. ACM, 2017.

^[4]https://www.cs.cornell.edu/courses/cs5413/2014fa/lectures/26-datacenter-middleboxes.pdf.

^[5] Clark, David D., Craig Partridge, J. Christopher Ramming, and John T. Wroclawski. “A knowledge plane for the internet.” In Proceedings of the 2003 conference on Applications, technologies, architectures, and protocols for computer communications, pp. 3-10. ACM, 2003.

^[6] https://www.vanityfair.com/news/2017/03/elon-musk-billion-dollar-crusade-to-stop-ai-space-x.

^[6] Stephen Hawking has previously said AI could destroy humanity (cf: http://www.bbc.co.uk/news/technology-30290540) but more recently expressed an ambivalent attitude that AI could either be the best or worst thing for the human race (cf: https://www.theguardian.com/science/2016/oct/19/stephen-hawking-ai-best-or-worst-thing-for-humanity-cambridge).

^[7] https://ico.org.uk/media/for-organisations/data-protection-reform/overview-of-the-gdpr-1-13.pdf.

Nishanth Sastry

Nishanth Sastry is a Senior Lecturer at King's College London, UK. Previously, he spent over six years in the Industry (Cisco Systems, India and IBM Software Group, USA) and Industrial Research Labs (IBM TJ Watson Research Center). His honours include a Best Paper Award at SIGCOMM Mobile Edge Computing in 2017, a Best Paper Honorable Mention at WWW 2018, a Best Student Paper Award at the Computer Society of India Annual Convention, a Yunus Innovation Challenge Award at the Massachusetts Institute of Technology IDEAS Competition, a Benefactor's Scholarship from St. John's College, Cambridge, a Best Undergraduate Project Award from RV College of Engineering, a Cisco Achievement Program Award and several awards from IBM. He has been granted nine patents in the USA for work done at IBM. Nishanth is a frequent keynote speaker and he has been interviewed or his past work has been covered in print media outlets such as the Times, New York Times, New Scientist and Nature, as well as Television media such as BBC, Al Jazeera and Sky News. He is a member of the ACM and a senior member of IEEE.

Mischa Dohler

Mischa Dohler is full Professor in Wireless Communications at King’s College London, driving cross-disciplinary research and innovation in technology, sciences and arts. He is a Fellow of the IEEE, the Royal Academy of Engineering, the Royal Society of Arts (RSA), the Institution of Engineering and Technology (IET); and a Distinguished Member of Harvard Square Leaders Excellence. He is a serial entrepreneur; composer & pianist with 5 albums on Spotify/iTunes; and fluent in 6 languages. He acts as policy advisor on issues related to digital, skills and education. He has had ample coverage by national and international press and media.

Editor:

Syed Hassan Ahmed

Syed Hassan Ahmed (S'13, M'17, SM'18) is an Assistant Professor in the Department of Computer Science at Georgia Southern University, Statesboro Campus, USA. Previously, he was a Post-Doctoral Fellow in the Department of Electrical and Computer Engineering, University of Central Florida, Orlando, USA. He completed his Bachelors in Computer Science from Kohat University of Science & Technology (KUST), Pakistan and Master combined Ph.D. Degree from School of Computer Science and Engineering (SCSE), Kyungpook National University (KNU), Republic of Korea (South Korea). In summer 2015, he was also a visiting researcher at the Georgia Tech, Atlanta, USA. Collectively, he has authored/co-authored over 130 international publications including Journal articles, Conference Proceedings, Book Chapters, and 03 books. In 2016, his work on robust content retrieval in future vehicular networks lead him to win the Qualcomm Innovation Award at KNU, Korea. Dr. Hassan's research interests include Sensor and Ad hoc Networks, Cyber-Physical Systems, Vehicular Communications, and Future Internet.