From the ethernet and laser printing to fiber optics and natural language processing, Xerox's Palo Alto Research Center (PARC) has been at the heart of some of tech's most important breakthroughs. This year, PARC is celebrating 50 years of innovation and is looking ahead to what the next 50 years hold.
Explainable AI? Digital packaging? Clean tech?
To get a sense of the scope of PARC's contribution over the last half-century, as well as insights into what technological breakthroughs we have to look forward to thanks to PARC's ongoing research and development, I connected with Naresh Shanker, CTO at Xerox.
GN: What's your personal favorite piece of technology out of PARC over the last 50 years?
Naresh Shanker: Ethernet was developed in the early 1970s as a local-area network (LAN) at PARC to enable connection between a series of computers, the world's first laser printer and the early internet, the ARPAnet.
Ethernet is probably my favorite, because of the massive impact, reach and ubiquitous nature of the technology. Just as Xerox did with the original copy machine, Ethernet once again redefined the work experience by changing the way people communicate.
While Ethernet was not the first network protocol, it is by far the most successful and has grown in lock step with the internet over the past 40 years. It is the tool that enabled interconnected offices and is still the technological foundation for most networking communications today.
GN: What technologies from PARC do you think haven't gotten their due or have been overlooked or overshadowed by more flashy tech?
Naresh Shanker: PARC's pioneering work in symbolic, grammar-based Natural Language Processing (NLP) started in the 1970s. By developing Lexical Functional Grammar, a computationally efficient model of grammar, Xerox/PARC's research allowed computers of that age to understand natural language text with high accuracy.
While statistical machine learning approaches have become more prevalent in recent years, we have not abandoned model-based AI, as it has become clear both approaches can be useful. While machine learning is useful in situations with large amounts of training data, there are many areas where there is a lack of training data that makes machine learning AI approaches fall short.
Xerox is now focusing on solving high-value problems where we can model the system/environment with knowledge from experts and then expand that knowledge through collaborative interactions with users.
This is the core premise behind our innovation focus area in AI workflow assistants for knowledge workers. Xerox is in the process of bringing technology to the market now that will assist knowledge workers in authoring business documents, the first of which will be a request for proposal (RFP) assistant. The RFP assistant leverages our years of expertise in document modeling and natural language processing to enable business development professionals to create draft proposals in just minutes. This is the first in what we expect to be a family of collaborative authoring tools designed to support knowledge workers.
Instead of replacing people, we believe that tools like AI can augment people and help them do more complex knowledge work more effectively.
GN: Why is PARC working on explainable AI? What's it all about and why is it going to be a game changer?
Naresh Shanker: Most modern AI algorithms are fairly blind in nature—they're like black boxes. You input the data and the AI learns and builds correlations, resulting in answers and recommendations without any insight into how the system arrived at those answers.
Our challenge is to ensure that AI truly understands and reasons in context. It's a huge priority as we continue to innovate in this space, and that's why Xerox/PARC is working with partners such as the Defense Advanced Research Projects Agency (DARPA) on explainable AI initiatives.
One project Xerox is working on with DARPA is The COGLE Project (Common Ground Learning and Explanation). The goal is to enable human-machine collaboration by creating a common language between drones and humans to accomplish tasks such as finding a lost hiker in the woods. The human and drone need to establish some common ground, or common language, to be able to work well together to get a result. These learned representations would then be shown to the human via COGLE's rich sense-making interface, enabling people to understand and predict the behavior of an autonomous system.
The overall goal is to make results more explainable and more transparent. As AI is increasingly used in situations that have a high degree of societal impact, such as sentencing decisions, job applications and medical diagnostics, we need to be sure that these algorithms are making the right decisions and that they are objective in their results. The way to do this is to make these systems transparent in a way that they can explain their work; including the assumptions they made, the different options they considered and eventually why they came up with the answer they provided.
By enabling this communication between humans and computers, we can increase trust in these AI systems, alleviating many of the ethical concerns that exist within the industry today and enabling collaboration between humans and computers to solve the most complex problems in the world.
GN: The 3D printing revolution is already upon us. What is liquid metal printing going to add to the field?
Naresh Shanker: By combining the power of Xerox's technology, bolstered by decades of experience in printing and materials science, manufacturers can make parts from start to finish in hours instead of days without sacrificing quality or strength. The promise of bringing 3D printing to manufacturing has long been a futurist vision for advancing the industry and Xerox is turning that vision into a reality.
The Xerox difference is our liquid metal technology that uses off-the-shelf alloys and eliminates many of the post-processing steps associated with metal powders, such as depowdering, debinding, and sintering, which are timely and costly steps. Pairing this technology with AI-based 3D software will radically transform the supply chain, allowing manufacturers to design and fabricate parts that meet their structural and cost requirements on the first try. Parts are denser, faster to make and cheaper compared to those made with metal powders. Our AI-based design software integrates with CAX packages seamlessly to boost productivity.
The biggest challenge is one that the entire industry is facing: full-scale buy-in to the technology from manufacturers. While some have been quick to see the advantages of 3D, others are taking a more cautioned approach. In order for 3D to take hold in true manufacturing, the industry needs to focus on the nuts and bolts, like design for additive, better integration into manufacturing workflows, standards and other areas beyond the print technology.
We are initially focused on the automotive and aerospace industries, but liquid metal printing is an entirely new process, and differentiation of the technology combined with Xerox's jetting know-how will open us up to even more markets going forward.