The AI World Conference & Expo is packed few days with news emanating from the Expo floor, the plenary sessions, a hackathon and tracks. There’s more good stuff than a writer can possibly fit into post-event coverage. Our Reporters’ Notebook comprises some of the bits and pieces that we collected over the three days in Boston. –John Desmond, Editor of AI Trends; Benjamin Ross, Senior Editorial Assistant; Allison Proffitt, Editorial Director
In an address to attendees of AI World 2019 in Boston recently, Paul F. Nemitz, principal Advisor, Directorate General and Justice and Consumers, European Commission, issued a warning about privacy.
In a talk entitled, “Democracy, Ethics and the Rule of Law in the Age of AI,” Nemitz provided the European view of privacy, calling the GDPR (General Data Protection Regulation, in effect May 2018) “the most sophisticated system for protecting personal data.”
He encouraged the audience to consider when data is personal data. For example, “If the data is from a handheld device, or a car, or is only generated by one person, then the data is probably personal data,” he suggested, adding, “Machine data that does not move with the person, is probably not personal.”
The fine for non-compliance with GDPR in Europe can be four percent of a company’s revenue, a level that has received some attention. In the US, the Federal Trade Commission in July fined Facebook $5 billion for privacy violations. Nemitz commented, “In America if you promise something to the public and you don’t follow through, you get in trouble with the FTC. America is leading in the enforcement of privacy rules on a big scale. What’s happening with smaller companies in the US is less clear.”
Some states, including California, have passed their own privacy laws modeled on the GDPR, but not going as far. “The spirit that the Internet should not be regulated is finished,” Nemitz said.
He suggested that “people are being manipulated on the basis of profiles built about them.” Whether that is good for democracy is an open question. “If you win because you are the best manipulator, I would say no,” Nemitz said.
Now the Europeans are working on shaping the age of AI. “We are looking at proposals to regulate AI to comply with fundamental values,” Nemitz said, then offering a preview of what’s in store.
For example, “We need to know when we are dealing with AI.” If you get a message saying a politician is great, you need to know where it came from. “There must be a disclaimer that says, ‘I am a machine.’ “ for example, he said, just as in well-practiced journalism, paid content is labelled as such.
The Europeans are likely to grade AI programs according to risk. A program helping with important decisions around a person’s health, is likely to carry a higher risk, thus the impact assessment might be tougher. “Three to five categories of risk are likely, of increasing relevance to individuals and society.” Self-assessment will not be enough.
It is likely to take up to two years to put these new AI regulations in effect. He encouraged those in the audience to participate. “It’s better for democracy,” Nemitz said.
Learn more at European Commission, Justice and Consumers.
Exelon Corp. Uses Smart Drones to Inspect Utility Poles
Exelon, a Fortune 100 energy company based in Chicago, has formed the Aerolabs subsidiary to perform inspections of critical infrastructure assets using drone employing sophisticated computer vision. Shyam Krishnaswamy, Director of Innovation and Strategy for Exelon, outlined how the subsidiary is delivering on the promise of AI in the energy sector in a talk at the recent AI World 2019 in Boston.
“Digital transformation is putting the right information into the hands of the right people at the right time,” he said. The use of intelligent drones is transforming the way asset inspection is undertaken by the utility. Using the example of wooden utility poles, Krishnasway showed how the images of the top of the pole provide information on the condition of the pole, that an inspection from the ground cannot reveal.
“We can see if there is damage that we need to send a crew to check,” he said. “We get more clarity; we get high quality data that we analyze to get insights.” Initially, the firm used subject matter experts to inspect the images, but then they decided to train a model to analyze the images for defects and anomalies. Over a million images have not been “ingested” by the system.
Defects such as pole rot, leaking transformers, cracked insulators and broken crossarms can now be detected automatically. The model is now 87 percent accurate, he said. Work continues to improve the rate and try to eliminate false positives.
Inspections are expensive to conduct, so the system is paying off. “Now we are able to quickly identify where to send crews,” Krishnaswamy said. The development road map includes putting more compute power closer to the poles, maybe in the trucks, so that edge analytics can be applied, and to develop the ability to query the data.
Learn more at AeroLabs.
Protiviti Moving into Advanced Analytics Using AI
Protiviti, a global consulting firm known for providing services for internal audit, risk and compliance, is moving into the advanced analytics market incorporating AI techniques, under the leadership of Madhumita Bhattacharyya, recently brought into the company as Practice Leader, Advanced Analytics and Technology Consulting.
Bhattacharyya has years of experience with analytics from working with IBM, Cognizant and HP.
In an interview with AI Trends at the AI World Conference & Expo, Bhattacharyya outlined the three initiatives she has undertaken since coming to Protiviti in January 2019. They are: AI applied to internal audits, AI in the cloud and AI in machine learning.
For auditing, “You need to align the algorithm with the business objective, and make sure the data set is right,” said Bhattacharyya. The internal auditors may not be as well-versed in AI modeling, so Protiviti brings in a team that includes a subject matter expert to collaborate.
For the cloud, Protiviti views Microsoft Azure as its analytics platform. AI and machine learning is well-positioned to detect fraudulent activity and keep up with rapid change. “Business requirements that were once static are now dynamic; AI with machine learning can help with that,” she said.
The three initiatives are in an early rollout stage with several early adopting client companies, she said. They include a law firm with 15 years of documents, an insurance company that uses a drone to take photos of roofs after extreme weather events, and a Hong Kong supplier working to put different invoice formats into a standard form. The essential business model is to support client projects, and in some cases to have an AI expert working at a client site, so the client does not have to necessarily make their own hire.
Bhattacharyya observed that the ability today to incorporate structured and unstructured data, and apply the needed processing power, has created new opportunities to apply advanced analytics. “We used to have to take samples of data and execute on those. Now we can expand to the entire population of data. That’s what has changed in the past three to five years,” she said.
Learn more at Protiviti.
PARC Putting Digital Twins to Work by Combining AI and IIoT
Ajay Raghavan, Strategic Execution Director for PARC, the Xerox company, in a talk entitled, “Can Digital Twins Unleash the Potential of AI?” at AI World 2019 in Boston recently, defined the term “digital twins” and outlined why it’s making sense now.
Manufacturing plants in the US are “showing their age,” with many at least 10 to 20 years old, and their operations and management teams being asked to do more with less. In a manufacturing plant, a single elbow joint failure can lead to an explosion that could result potentially in loss of life and millions in damages. This happened at the Humber plant in the UK in 2001, and in the Texas City Refinery explosion in 2005, which killed 151 workers, injured 180 others and severely damaged the refinery.
“Unplanned downtime can be extremely expensive,” Raghavan said.
Industry 4.0 involves instrumenting legacy assets with sensors, connecting them, then using the data to monitor in real time, thus creating the “digital twin.” he said. “We have an opportunity to get better insights into what’s happening with the system.”
The digital twins are a hybrid of Industrial Internet of Things (IIoT) and AI modeling. PARC has packaged a set of services for this called MOXI System Analytics. PARC describes the resulting system as self-adaptive, enabling a transition to predictive, condition-based maintenance.
Early experience is showing a high success rate with predictive maintenance, preventative maintenance, self-adaptation and self-coordinated assets being put into place for clients.
Panasonic is an early adopter, using the system for robot fault detection at a manufacturing plant in Japan. “Industrial robots are critical to the automotive industry. A failure can mean the shutdown of a line,” Raghavan said. Using the digital twin approach, PARC developed vibration sensors with a deeper understanding, segmenting the signals into finer grains, enabling better predictions.
IHI, a diversified manufacturing of products in Japan, is working on “fusing multi-modal data streams,” which are challenging to consolidate, to create an adaptive real time planning and control system.
The volumes of data that can be produced by IIoT sensors is beyond what humans are able to comprehend unassisted. PARK is working on data stream visualization to help. “We’re deploying in factories,” Raghavan said. “Industry 4.0 is emerging as an acute need, and digital twins are key to achieving self-aware factories; they can unleash AI for Industry 4.0.”
Learn more about MOXI System Analytics.
CVS Health’s Deep Learning Model for NLP
An important component to AI’s—especially natural language processes (NLPs)—adoption in any space is its ability to understand basic language functions. Sadid Hasan, Senior Director of Artificial Intelligence at CVS Health, discussed the challenges that come when applying AI to a data-rich field like healthcare.
“With so many data sources, how can we provide solutions for the clinicians so that they can understand those data and use this data in a meaningful way?” Hasan said. One use case is using deep learning to identify disease names within clinical texts.
“If you have a system that can recognize disease names from the clinical text, you can use this model for patient profiling and clinical trial matching,” said Hasan. “This is a fundamental task.”
The process begins with the fundamentals, Hasan says. He relates it to a small child being taught the difference between an orange and an apple by their parents. “Similarly, our AI must learn the difference between disease name and text.”
As a human, Hasan says we are fairly capable of recognizing disease names, even if we are medically trained. For AI, it’s a much more challenging task.
“For one, there are lots of acronyms in healthcare,” said Hasan. “They’re also non-standard because healthcare providers have their own acronyms, as well as their own document structure, and you have to deidentify or anonymize the text. This induces more noise into the data.”
Hasan and his team proposed a deep learning (DL) model in which they would induce clinical and domain knowledge for the purpose of generating better results. “This means not just giving the level data, but also telling the model what the clinical terms are, adding layer by layer so the model won’t forget.”
There’s a patient engagement angle to the model as well, Hasan says. Oftentimes patients are unable to understand healthcare information due to their lack of understanding medical terms. Applying DL to clinical texts for the purpose of simplification can be a huge benefit on any healthcare system, though defining a subjective term like “simple” presents its own challenges.
“What’s simple for you might not be simple for me, and vice versa,” Hasan said. “We proposed some [DL] models for this task… beginning with paraphrasing phrase-level text.”
Because there was no dataset for this kind of task, Hasan and his colleagues had to build their own using published sources.
Hasan says initial results are promising, though additional research is required.
AI World 2019 Covered in RT Insights, TechTarget, CMSWire
The AI Trends team wasn’t the only one taking notes. Here is some of the other news coverage to come from the event.
IoT-extended AI brings three main benefits to manufacturers: supply chain optimization, smart manufacturing, and product or service innovation. Joe McKendrick, RT Insights, summarizes a panel on the impending convergence of IoT and AI. Condition monitoring or asset tracking is likely a good place to start.
McKendrick also moderated a session on leveraging edge data. The panel discussed retrieving data from older industrial systems and managing data storage.
At TechTarget, Kassidy Kelley predicted the beginning of a potential chatbot revolution. She summarizes talks from across the event discussing what conversational AI will look like, and what we need to get there.
Some of the prevailing themes at the conference—as distilled by Dom Nicastro from CMS Wire—included the recognition that AI in the enterprise is still a nascent venture, and that AI won’t produce tangible outcomes without a good foundation of usable data.
At Enterprise AI Doug Black outlined Constant Contact’s attempts at AI-based customer support to manage growing call volumes and high turnover of call center workers. Conversational AI has driven down Constant Contact costs and improved customer support.
“Quantum computing is rapidly becoming reality, and the time for businesses to start planning for a quantum future is now, particularly when it comes to how quantum computing and machine learning will work together,” says Ed Burns at Search CIO. Burns covered the quantum computing panel, outlining out quantum computing will enable AI and what we need to get there.