By Gregory Pings
Two things you must know about artificial intelligence (AI):
“AI Investment Is Healthy and Increasing: Investments in AI pilots and proofs of concept are beginning to yield results, fueling double-digit budget growth for AI.
“If You Approach AI as Traditional Tech, You Will Fail: The opportunity cost of AI aligns with its return on investment (ROI) when technology organizations use it to redesign business models and operations, not to optimize individual tasks.”[i]
No longer should you wonder why the artificial intelligence hype has reached fever pitch. That said, “hype” is the operative word. For this reason, leaders from all sectors are forgiven if they can’t distinguish between what is real and what is fiction. Failure to join the AI fray, however, will likely be unforgivable.
That’s why Xerox and Forrester invited Silicon Valley leaders to a lunch and networking event at PARC, the famed Xerox research lab.
The event focused on real-world applications of artificial intelligence. Hosts Lawrence Lee (Xerox Vice President of Incubation and Strategy) and Brandon Purcell (Principal Analyst at Forrester) highlighted emerging trends and questions from this nascent field, and a look at how AI can impact Xerox’s (and our clients’) business model.
The artificial intelligence revolution is underway: AI experts discuss applications, opportunities and trust at PARC summit.
We're putting artificial intelligence to work for the print industry right now. What are you waiting for?
— Xerox (@Xerox) July 4, 2018
Artificial intelligence examples
Lawrence Lee dialed the way-back machine 30 years to PARC’s early work in AI at a time when memory was expensive and computers were slow. Everything about computing was expensive. Despite these limitations, PARC researchers created pioneering technologies in natural language processing. They created symbolic representations of grammar as well as the computational algorithms that could parse language into semantic representations, and also reverse them to generate natural language text.
More recent examples are found in machine automation. PARC and Xerox researchers developed a model-based approach for an AI planner that controls the software for Xerox reconfigurable production printing presses. These presses can run with a number of different modules, and the planner understands how they fit together and creates a process plan to complete the print job using the modules in a given press. The planner, in essence, “programs” the control software for a particular print job so that print runs are much more reliable.
Resource: artificial intelligence
AI and Human-Machine Collaboration: PARC’s work in artificial intelligence and human-machine collaboration is at the forefront of a new, transformative era of machine intelligence.
— Xerox (@Xerox) July 31, 2018
“Many times,” Lawrence pointed out, “a human programmer cannot anticipate all the ways in which a machine will operate in the field, and problems occur when a machine’s programming can’t handle an unforeseen operation.”
In both cases, PARC developed a hybrid approach to AI that combines model-based reasoning and machine learning. This allows PARC to create robust AI systems even when training data is sparse, such as often is the case with highly engineered physical systems and unstructured enterprise applications.
For example, existing enterprise workflow automation solutions work well on repetitive processes that are predictable, but the next frontier will be in intelligent assistants that can help with more complex and variable knowledge work. Along these lines, Xerox is working on AI technologies that can help automate collaborative content authoring workflows based on its deep expertise in document understanding and natural language processing.
“A document,” Lawrence explained by way of example, “is defined by its intent, structure, and content. We can combine document models with training examples, and for specific document types, our system will be able to generate a first draft automatically.” From that point, humans review and edit the draft to add the creativity and judgment that is difficult for an AI system to understand.
To be sure, this is the type of automation that will disrupt some sectors, but Brandon says Forrester’s research points to the fact AI will not cause massive unemployment. “The term ‘augmented intelligence’ has taken root,” he said, “and the nature of people’s jobs will change.”
“I think about this type of automation in terms of human-machine collaboration where machines help people do their jobs better,” Lawrence responded. “This is the Xerox legacy of making technology so easy, that everyone can use it.”
“Augmented intelligence” describes the powerful combination of artificial and human intelligence. Such systems result in a modern centaur, so called after the half-human, half-horse creatures of Greek mythology. These human-machine teams, take advantage of each member’s strength: the computer’s speed and the human’s knowledge and perception.
Can you trust artificial intelligence?
No question, introducing AI into your organization is a formidable challenge. Given the primacy of accurate models, up-to-date information, and correcting for bias, Brandon pointed out that trust is critical. If a machine advises a certain action, the human must understand how the machine is modeled, as well as the predictors it uses.
“It’s a huge issue for doctors making a medical diagnosis where the consequences could be quite dire,” he explained. “Even the marketers I talk to won’t use a black box algorithm if they don’t at least understand the predictors.”
Lawrence agreed with Brandon’s assessment: “If humans cannot understand why a system makes a decision, or if the machine cannot demonstrate that it understands the broader context and explain why it made a specific recommendation, then AI will not realize its full potential.”
Trust is the crux of “Explainable AI (XAI),” a research initiative funded by the Defense Advanced Research Projects Agency (DARPA). PARC was selected as a participant in this program, working with collaborative partners to develop a framework to increase the transparency of machine learning through explanatory models.
The notion of collaboration and trust among humans and machines is the key. How can these teams of modern centaurs work in an enterprise? How do you create systems that figure out how to break down tasks among humans and machines? If you can answer these questions, then you have a leg up on AI.
— Xerox (@Xerox) July 6, 2018
Resource: Innovation from Xerox
Xerox innovation at work – The future is underway with big, juicy ideas that solve today’s business challenges with an eye toward tomorrow.
Xerox Connect reports on innovation — News and ideas from or about executives and employees in the Xerox Research, Development and Engineering group.
What’s next for Xerox and artificial intelligence?
PARC’s work on the DARPA XAI program offers a good model for what could be next in Xerox’s future. By leveraging government research funding to solve important technical challenges in the world, PARC gains highly valuable insights that can be applied to solve similar problems for Xerox’s customers, such as creating intelligent assistants that collaborate with knowledge workers and turn information into action.
For example, PARC scientists are working on conversational interfaces that will offer humans both a higher degree of transparency and a more natural conversational interaction. Whether it’s better chatbot experiences or voice interactions, the solution must go beyond single-turn inquiries. The machine must be able to understand questions from people who don’t know how to use the structured vocabulary that the device expects, or who don’t have the expertise to express a perfect query. The system can’t simply say “I don’t understand,” or give you a list of search results for possible answers. It should engage in a conversation and create an evolving shared understanding between the human and machine to deliver the right information, or complete a collaborative task together.
As another example, The U.S. Department of Energy is funding research at PARC on printed sensors and electronics. Inexpensive sensors within buildings can provide high resolution thermal models that could optimize heating and air conditioning systems to improve efficiency. Lawrence pointed out that we’re only seeing the beginnings of what will be possible in terms of new low-cost sensors in the Internet of Things, which will generate new forms of data that will further drive the capabilities of artificial intelligence.
For Xerox, research in artificial intelligence and, more importantly, how it will deliver concrete business value, elevates conversations with our customers beyond print, and help them embrace the era of intelligent work.
— Xerox (@Xerox) July 6, 2018