AI took center stage in recently-announced updates to the Alexa virtual voice assistant, and in the charges this week from the European Commission that Amazon is breaking EU competition rules.
During Amazon’s Alexa Live event held in July, the company announced a major update to Alexa’s developer toolkit that brings AI improvements. Since launching in 2014, Amazon’s voice assistant has shipped hundreds of millions of units, which are targeted by a sizable developer community offering voice apps, called Skills, that extend the Alexa default feature set. Just as the Android and iOS large selections of third party applications differentiate those operating systems, so Skill plays an important role in Amazon’s growth strategy for Alexa, according to a recent account in siliconAngle.
Amazon added deep learning models for natural language understanding that the company said will enable Skills to recognize users’ voice commands with 15% higher accuracy on average. Current Skills users can use the new technology without any modifications, according to Amazon.
Amazon also enhanced the voice assistant platform for more specific uses that are emerging as Alexa is added to more devices, including smartphones, wearables and smart displays. A new tool, Apps for Alexa, allows developers of mobile apps to enable customer control in a hands-free way, such as with the Echo Buds wireless earbuds. Another tool enables developers to allow purchases such as food delivery orders on Alexa-powered smart screens, such as the Echo Show smart display.
Developers of Skills for the Echo Bud are getting a new capability called “skill resumption,” which allows Skills to automatically “resume” at opportune times. For example, if a consumer uses Echo Buds to hail an Uber car, Uber’s Alexa skill can automatically notify them when their ride arrives without requiring a manual invocation.
Skills have momentum; Amazon announced that customer engagement with Alexa Skills nearly doubled over the past year.
AZ1 Edge Processor Can Perform On-Device Processing, a Privacy Win
Alexa is also moving to the edge with its own chip in smart home edge devices. The Echo devices are using the company’s AZ1 Neural Edge processor, which consumes 20x less power, 85% less memory and features double the speech processing power as predecessors, according to an account from ZDNet.
The AZ1 in concert with Amazon’s AI advances is aimed at making the Echo more aware of its surroundings. Dave Limp, senior vice president of devices and services at Amazon, stated that the new Echo devices are designed to make “moments count.”The new versions of Alexa will be able to learn from humans by asking follow-up questions when Alexa has a gap in its understanding, according to Rohit Prasad, VP and head scientist for Alexa AI at Amazon, in a presentation on new Alexa features at the virtual event. New versions will also use deep learning space parsers to understand gaps and extract new concepts, will show more natural conversation, and will engage a follow–up mode when interacting with humans.
Alexa can use visual and acoustic cues to determine the best action to take. “This natural turn-taking allows people to interact with Alexa at their own pace,” Prasad stated.
The new AI foundation technology for Alexa’s ability to interpret context and adjust how to speak to you, has been in development for years at Amazon, Prasad said.
The AZ1 edge processor is making Alexa faster. “The processor on the device is key with a fast-paced conversation,” stated Prasad. “The neural accelerator on the device makes decisions much faster.”
Alexa for Business, rolled out over a year ago, has been adding features via AWS. Skill Blueprints were launched in April 2018 as a way to allow anyone to create skills and publish them to the Skills Stores with a 2019 update.
Prasad did not outline the roadmap for Alexa for Business, but did say Echo’s new capabilities would apply to office settings as well as to yet-to-be-determined use cases. “There’s the potential to be able to teach Alexa anything in principle,” Prasad stated.
The AZ1 processor, built with Taiwanese semiconductor company MediaTek, will speed Alexa’s response to queries and commands by hundreds of milliseconds per response, according to an account in The Verge. That allows for on-device neural speech recognition.
Amazon’s preexisting products without the AZ1 send both the audio and its corresponding interaction to the cloud to be processed and back. Only the Echo and Echo Show 10 currently have the on-device memory needed to support Amazon’s new all-neural speech models. Given that the data is stored and deleted locally, the edge computing is seen as a privacy win.
European Commission Charging Amazon with Unfair Competition
All this smart processing is getting Amazon into trouble in Europe, with the European Commission this week charging the company with gaining an illegal advantage in the European marketplace. This was based on the use by Amazon of sales data of independent retailers selling through its site, data not available to other companies in the European market, and which Amazon uses to sell more of its most profitable products.
Margrethe Vestager, the commission’s executive vice-president, stated that the commission’s preliminary conclusion was that Amazon used “big data” to illegally distort competition in France and Germany, the biggest online retail markets in Europe, according to an account in The Guardian. The investigators will examine whether Amazon set rules on its platform to benefit its own offers and those of independent retailers who use Amazon’s logistics and delivery services.
“We do not take issue with the success of Amazon or its size. Our concern is very specific business contacts which appear to distort genuine competition,” Vestager stated. The EU team has since July analyzed a data sample of more than 18 million transactions on more than 100 million products.
The commission determined that real time business data relating to independent retailers on the site was being fed into an algorithm used by Amazon’s own retail business. “It is based on these algorithms that Amazon decides what new products to launch, the price of each individual offer, the management of inventories and the choice of the best supplier for a product,” Vestager stated. “We therefore come to the preliminary conclusion that the use of this data allows Amazon to focus on the sale of the best-selling products, and this marginalizes third party sellers and caps their ability to grow.”
Amazon faces a possible fine of up to 10% of its annual worldwide revenue. That could amount to as much as $28 billion, based on its 2019 earnings.
In a statement Amazon said it disagreed with the findings. “There are more than 150,000 European businesses selling through our stores that generate tens of billions of euros in revenues annually,” the company stated.
The Internet of Medical Things (IoMT) market is expanding rapidly, with over 500,000 medical technologies currently available, from blood pressure and glucose monitors to MRI scanners. AI poised to contribute analysis crucial to innovations such as smart hospitals.
Today’s internet-connected devices aim to improve efficiencies, lower care costs and drive better outcomes in healthcare, according to a recent account in HealthTech Magazine. Devices in the IoMT domain extend to wearable external medical devices such as skin patches and insulin pumps; implanted medical devices such as pacemakers and cardioverter defibrillators; and stationary devices such as for home monitoring and connecting imaging machines.
Projections for IoMT market size were aggressive before the COVID-19 pandemic hit, with Deloitte sizing the market at $158.1 billion by 2022, with the connected medical device segment expected to take up to $52.2 billion of that by 2022.
Now the estimates are growing. The global IoMT market was valued at $44.5 billion in 2018 and is expected to grow to $254.2 billion in 2026, according to AllTheResearch. The smart wearable device segment of IoMT, inclusive of smartwatches and sensor-laden smart shirts, made up for the largest share of the global market in 2018, at roughly 27 percent, the report found.
This area of IoMT is poised for even further growth as artificial intelligence is integrated into connected devices and can prove capable of real-time, remote measurement and analysis of patient data.
Fitbit Trackers Found to Help Patients with Heart Disease
Evidence is coming in on the effectiveness of IoMT for health care. A study conducted by researchers from Cedars-Sinai Medical Center and UCLA found that Fitbit activity trackers were able to more accurately evaluate patients with ischemic heart disease by recording their heart rate and accelerometer data simultaneously. Some 88% of healthcare providers were found in a survey last year of 100 health IT leaders by Spyglass Consulting Group, to be investing in remote patient monitoring (RPM) equipment. This is especially true for patients whose conditions are considered unstable and at risk for hospital admission.
Cost avoidance was the primary investment driver for RPM solutions, which are hoping to achieve reduced hospital readmissions, emergency department visits, and overall healthcare utilization, the study stated.
Wearable activity trackers have also proven to be a more reliable measure of physical activity and assessing five-year risk than traditional methods, according to a study by Johns Hopkins Medicine, as reported in mHealthIntelligence.
Adult participants between 50 and 85 years old wore an accelerator device at the hip for seven consecutive days to gather information on their physical activity. Individual data came from responses to demographic, socioeconomic, and health-related survey questions, along with medical records and clinical laboratory test results.
IoMT Devices Seen as Helping to Control Health Care Costs
Medical cost reductions of $300 billion are being estimated by Goldman Sachs, through remote patient monitoring and increased oversight of medication use. Startup activity is picking up. Proteus Discover, for example, has focused its smart pill capabilities on measuring the effectiveness of medication treatment; and HQ’s CorTemp is using its smart pills to monitor patients’ internal health and transmit wireless data such as core temperatures, which can be critical in life or death situations.
AI systems are seen as able to reduce therapeutic and therapeutic errors in human clinical practice, according to an account in IDST. Developing IoMT strategies that match sophisticated sensors with AI-backed analytics will be critical for developing smart hospitals of the future. “Sensors, AI and big data analytics are vital technologies for IoMT as they provide multiple benefits to patients and facilities alike,” stated Varun Babu, senior research analyst with Frost & Sullivan TechVision Research, which studies emerging technology for IT.
The rise of AI and its alliance with IoT is one of the critical aspects of the digital transformation in modern healthcare, according to an account in IoTforAll. The central pairing is likely to result in speeding up the complicated procedures and data functionalities that are otherwise tedious and time-consuming. AI along with sensor technologies from IoT can lead to better decision-making. Advances in connectivity through AI are expected to promote an understanding of therapy and enable preventive care that promises a better future.
The impact of AI on personal healthcare is attracting wide comment. “AI is transforming every industry in which it is implemented, with its impact upon the healthcare sector already saving lives and improving medical diagnoses,” stated Dr. Ian Roberts, Director of Therapeutic Technology at Healx, a biotechnology company based in Cambridge, England, in an account in BBH (Building Better Healthcare). “The transformative effect of AI is set to switch healthcare on its head, as the technology leads to a shift from reactive treatments targeting populations to proactive prevention tailored to the individual patient.”
In the future, AI-generated healthcare recommendations are seen as extending to include personalized treatment plans. “Currently we are in the infancy of AI in healthcare, and each company drives forward another piece of the puzzle and once fully integrated the future of medicine will be forever transformed,” Dr. Roberts stated.
However, the increasingly-connected environment of IoMT is seen as bringing new risks as cyber criminals seek to exploit device and network vulnerabilities to wreak havoc. A recent global survey by Extreme Networks, a network infrastructure provider, found that one in five healthcare IT professionals are unsure if every medical device on their network has all the latest software patches installed — creating a porous security infrastructure that could potentially be bypassed.
“2020 will be the year when healthcare organizations of all sizes will need to realize that they are easy pickings for cyber criminals, and put a robust, reliable and resilient network security infrastructure in place to protect themselves adequately,” stated Bob Zemke, director of healthcare solutions for Extreme.
Data science is seen as leading to more precise analytics. “In 2020, we can expect to see better patient outcomes fueled largely by the growing prevalence of data science and analytics,” stated lan Jacobson, chief data and analytic officer at Alteryx, a software company providing advanced analytics tools. “Much of the data that is required to solve some really-key challenges already exists in the public domain, and in the next year we expect more and more healthcare organizations will implement tools that help to assess this rich information as well as gain actionable insight.” The tools are seen as being effective in monitoring proper use of prescription drugs.
A “chemputer” is a robotic method of producing drug molecules that uses downloadable blueprints to synthesize organic chemicals via programming. Originated in the University of Glasgow lab of chemist Lee Cronin, the method has produced several blueprints available on the GitHub software repository, including blueprints for Remdesivir, the FDA-approved drug for antiviral treatment of COVID-19.
Cronin, who designed the “bird’s nest” of tubing, pumps, and flasks that make up the chemputer, spent years thinking of a way researchers could distribute and produce molecules as easily as they email and print PDFs, according to a recent account from CNBC.
“If we have a standard way of discovering molecules, making molecules, and then manufacturing them, suddenly nothing goes out of print,” Cronin stated. “It’s like an ebook reader for chemistry.”
Beyond creating the chemputer, Cronin’s team recently took a second major step towards digitizing chemistry with an accessible way to program the machine. The software enables academic papers to be made into ‘chemputer-executable’ programs that researchers can edit without learning to code, the scientists announced in a recent edition of Science. The University of Glasgow team is one of dozens spread across academia and industry racing to bring chemistry into the digital age, a development that could lead to safer drugs, more efficient solar panels, and a disruptive new industry.
Cronin’s team hopes their work will enable a “Spotify for chemistry” — an online repository of downloadable recipes for molecules that could enable more efficient international scientific collaboration, including helping developing countries more easily access medications.
“The majority of chemistry hasn’t changed from the way we’ve been doing it for the last 200 years. It’s a very manual, artisan–driven process,” stated Nathan Collins, the chief strategy officer of SRI Biosciences, a division of SRI International. “There are billions of dollars of opportunity there.” He added, “This is still a very new science; it’s started to really explode in the last 18 months.”
The Glasgow team’s software includes the SynthReadertool, whichscans a chemical recipe in peer-reviewed literature — like the six-step process for cooking up Remdesivir — and uses natural language processing to pick out verbs such as “add,” “stir,” or “heat;” modifiers like “dropwise;” and other details like durations and temperatures. The system translates those instructions into XDL, which directs the chemputer to execute mechanical actions with its heaters and test tubes.
The group reported extracting 12 demonstration recipes from the chemical literature, which the chemputer carried out with an efficiency similar to that of human chemists.
Cronin founded a company called Chemify to sell the chemistry robots and software. In May of 2019, the group installed a prototype at the pharmaceutical company GlaxoSmithKline.
“The chemputeras a concept and the work [Cronin]’s done is really quite transformational,” stated Kim Branson, the global head of artificial intelligence and machine learning at GSK. The company is exploring various automation technologies to help it make a wide array of chemicals more efficiently. Cronin’s work may let GSK “teleport expertise” around the company, he stated.
Researchers at SRI are pursuing their SynFynsynthetic-chemistry system to expedite discovery of selective molecules.Collins recently published related research,Fully Automated Chemical Synthesis: Toward the Universal Synthesizer. AutoSyn,“makes milligram-to-gram-scale amounts of virtually any drug-like small molecule in a matter of hours,” he said in a recent account inThe Health Care Blog.
He sees the combination of AI and automation as an opportunity to improve the pharma R&D process. “Progress in AI offers the exciting possibility of pairing it with cutting-edge lab automation, essentially automating the entire R&D process from molecular design to synthesis and testing — greatly expediting the drug development process,” Dr. Collins stated.
SRI is pursuing partnerships to help accelerate the digitized drug discovery. A recent example is a collaboration with Exscientia, a clinical state AI drug discovery company, to work on integration of Exscientia’s Centaur Chemist AI platform to the SynFini synthetic chemistry system, described recently in a press release from SRI.
Exscientia applies AI technologies to design small molecule compounds that have reached the clinic. Molecules generated by Exscientia’s platform are highly optimized to satisfy the multiple pharmacology criteria required to enter a compound into the clinic in record time. Centaur Chemist is said to transform drug discovery into a formalized set of moves while also allowing the system to learn strategy from human experts.
Andrew Hopkins, CEO of Exscientia stated, ”The opportunity to apply AI drug design through our Centaur Chemist system with SynFiniautomated chemistry offers an exciting opportunity to accelerate drug discovery timelines through scientific innovation and automation.”
SRI also announced a partnership earlier this year with Iktos, a company specializing in using AI for novel drug design, to use Iktos’ generative modeling technology will be combined with SRI’s SynFini platform, according to a press release from Iktos. The goal is to accelerate the identification of drug candidates to treat multiple viruses, including influenza and COVID-19.
The Iktos AI technology is based on deep generative models, which help design virtual novel molecules that have all the desirable characteristics of a novel drug candidate, addressing challenges including simultaneous validation of multiple bioactive attributes and drug-like criteria for clinical testing.
By Dawn Fitzgerald, theAI Executive Leadership Insider
Part Three of Four Part Series: “AI Holistic Adoption for Manufacturing and Operations” is a four-part series which focuses on the executive leadership perspective including key execution topics required for the enterprise digital transformation journey and AI Holistic Adoption for manufacturing and operations organizations. Planned topics include: Value, Program, Data and Ethics. Here we address our third topic: Data.
The Executive Leadership Perspective
For the executive leader who is taking their enterprise on a journey of Digital Transformation and AI Holistic Adoption, we started this series with the foundation of Valueand then moved to the framework of the Program. Although these are the fundamental building blocks required for success, the results of any enterprise’s analytics, do, in the end, rely on the Data.
The executive leader has the responsibility to ensure that they and their team are dedicated to mastering data fluency and data excellence in the enterprise. The facets of Data Management are vast with the standard areas of focus including data discovery, collection, preparation, categorization and protection. Strategies for achieving maturity in these areas are well-established in most industries, and yet many industries still struggle. These standard areas of focus in Data Management are indeed necessary but are not sufficient for the needed AI Holistic Adoption.
To incorporate AI Holistic Adoption, a value focus must be employed where we create Value Analytics (VAs) as output from our enterprise Analytics Program. To support this program, we must expand our enterprise Data Management definition to include a Data Optimality metric, a Data Evolution Roadmap and a Data Value Efficiency metric.
The Data Optimality metric tells us how close the Value Analytics (VA) Baseline Dataset is to ‘optimal’. The Data Evolution Roadmap captures the milestones for the evolution of our Baseline Dataset for each Value Analytics release and the corresponding goals for harvesting data. The Data Value Efficiency metric simply measures how much value we achieve from harvested data. The combination of these is a powerful tool set for the executive leader to ensure the data provides the highest value to enterprise analytics at the lowest cost to the organization.
The Data Optimality Metric Definition
The Data Optimality metric tells us how close the Value Analytics (VA) Baseline Dataset is to the Data Scientist-defined ‘optimal’. The Baseline Dataset is a key component to any Value Analytic. The Baseline Dataset captures the data used for the VA as it relates to a specific development release. This link to a release is a critical distinction. By tying the Baseline Dataset to the VA design release, we recognize a snapshot of the training data associated with a specific release. We recognize that it may not be optimal so may change during the lifetime of the VA, and we plan for its change on a Data Evolution Roadmap.
To achieve enterprise AI Holistic Adoption the executive leader must ensure the foundation of Value which anchors the effort. They must also incorporate the nature of a technical development effort. Specifically, they must account for the go-to-market demands that drive risk management decisions regarding minimal viable product (MVP) in Agile or SAFe (Scaled Agile Framework)methodologies. By the very nature of development, the MVP-driven organization will plan early deliverables with incremental improvements over time. This will apply to the Baseline Dataset as well and thus, the Data Optimality Metric is created. It is used for visibility of the state of our Baseline Dataset, used to communicate expectations of its impact on the VA and used to drive the evolution of the data.
Data Optimality Metric Example
To illustrate the power of the Data Optimality metric, consider the Data Scientist who has defined an equipment predictive maintenance algorithm and has a corresponding Baseline Dataset definition. They will have defined the optimal dataset that they want which includes the IoT measurements (for example: temp, pressure and vibration), the duration of time they would like the Data collected over (for example: 6 months), the population size (for example: data collected from 10 Data Centers covering four key climate zone geographies) and a guaranteed data quality level (for example less than 10% data gaps). Since there is a low probability of this optimal Baseline Dataset availability aligning with the market-driven release timeline demands, the Data Scientist may be forced to compromise their initial Baseline Dataset by taking fewer IoT parameters (for example: only temp and pressure but no vibration), having shorter collection duration (for example: 3 months vs 6), having a smaller population size (for example: only 3 Data Centers vs 10) or accepting a lower quality level guarantee. The Data Scientist may also create simulated data for some or all of the data gaps.
The Data Scientist will then assign a Data Optimality metric to the current release Baseline Dataset (for example: current available data achieves 60% of the optimal dataset criteria). They will also state the lower Data Optimality metrics potential impact on the Value Analytic (for example: customers can expect only a 30-day prediction vs 90-day prediction pre-failure).
The executive leader can then make a business decision to go forward with this Data Optimality metric or wait the extra time necessary to harvest improved data to achieve a higher Data Optimality metric and corresponding VA improvement. To conclude this scenario example, input from the marketing team may indicate that a Q2 release of the VA with the current Data Optimality metric is acceptable due to first mover advantage and significant value, compared to competitive offers, delivered to the customer.
They may also specify that the higher Data Optimality metric must be achieved by Q4 in order to remain competitive. The Data Optimality metric enables defined incremental improvements to the Baseline Dataset over time which transcend to the ongoing VA improvement lifecycle.
The visibility provided by the Data Optimality metric is especially valuable with leading edge Value Analytic capabilities where first mover advantage in the market can lead to a substantial market penetration foothold for the business. The metric drives cost saving by bringing the decision point of release impacting information down to the local business, where the knowledge of the business is the highest. This simultaneously gives visibility to future data management actions through the enterprise and should be captured in the Data Evolution Roadmap.
The Data Evolution Roadmap
Driven by Data Optimality metric inputs, the Data Evolution Roadmap captures the milestones for the evolution of our Baseline Dataset for each Value Analytics release and the corresponding goals for harvesting future required data. The Data Evolution Roadmap establishes an enterprise framework that provides visibility, alignment, clarity and flexibility for local business decisions. It also challenges the business to define the Data Optimality metric and track Baseline Dataset improvements.
The power of the Data Evolution Roadmap enables the local businesses’ Agile development methodologies, gives cross-functional visibility of data management actions and delivers Data Management cost saving to the enterprise. Incremental improvements of the Data Optimality metric for a specific Value Analytic can be timed on the Data Evolution Roadmap based on demand. Early market traction data can be incorporated to update the business decision thus generating higher confidence in the data management expenditures and potential cost savings if deemed no longer necessary.
To achieve AI Holistic Adoption, the Data Evolution Roadmap must align directly to the Value Analytics Roadmap. Data management tasks must align and be traceable through both roadmaps to a higher end value. Successful execution of this requires rapid, tightly coupled agile development teams that span the key enterprise stakeholders such as IoT development, Data management, Data Science, platform development and marketing/sales functions. This demand-pull approach to Data Management aligns well with Agile development practices and combats the seemingly overwhelming challenges of exponential data repository growth and corresponding data management costs.
Data Repository Growth
The growth of the data repository should parallel the growth and maturity of the Analytics Program to ensure data excellence and avoid dark data obsolescence. The cost of technical debt must be acknowledged and measured.
Many companies make the mistake of a volume goal of collecting IoT data without a defined data evolution strategy aligned with the Analytics Program grounded in value. This leads to the data swamp, a stalling of the realization of Value from the AI solutions and an overall low Data Value Efficiency score as defined below.
A tighter alignment of the Data Management tasks with the Value Analytics also provides opportunity for more value-based incremental improvements of the enterprises’ tagging strategy.Tagging data with both technical and business metadata is critical but seldom donecorrectly first pass and certainly not without a Value focus, which requires a cross-functional team of a data architect, data scientist, subject-matter expert and marketing that anchor the value. The mechanism to continuously improve your data tagging methodology must be close to the value goals of the Analytics Program.
The Data Value Efficiency
Once the Data Optimality metric and Data Evolution Roadmap are established, a Digital Value Efficiency (DVE) metric can be measured. The Data Value Efficiency (DVE), a measurement attached to data elements, is simply the measure of how much value we achieve from harvested data. The DVE tracks the use of the data by its inclusion in different VA Baseline Datasets over time.
In most industries using AI, this metric would be considered very low. IDC research defines that currently, “80% of time is spent on data discovery, preparation, and protection, and only 20% of time is spent on actual analytics and getting to insight.”To achieve high DVE, a larger portion of our data harvested must translate into higher value actionable insights.
Since the executive leader’s responsibility is to ensure that the organization is efficient with the data management, they must focus their organization on shifting the percentage of time invested from data discovery, collection and preparation to a higher amount of time used in training models and insight generation. The DVE metric gives visibility to progress toward this goal.
The Data Evolution Roadmap pivots the enterprise focus from one of maximum data collection, and corresponding cost, to one of minimized data collection driven by the Value Analytics roadmap. Over time, this will improve the DVE metric and overall data excellence of the enterprise.
Dawn Fitzgerald is VP of Engineering and Technical Operations at Homesite, an American Family Insurance company, where she is focused on Digital Transformation. Prior to this role, Dawn was a Digital Transformation & Analytics executive at Schneider Electric for 11 years.She is also currently the Chair of the Advisory Board for MIT’s Machine Intelligence for Manufacturing and Operations program. All opinions in this article are solely her own and are not reflective of any organization.
from AI Trends https://ift.tt/38CCnZb
via A.I .Kung Fu
Understanding and Advising on Cyber and Physical Risks to the Nation’s Critical Infrastructure
Brian R. Gattoni is the Chief Technology Officer for the Cybersecurity and Infrastructure Security Agency (CISA) of the Department of Homeland Security. CISA is the nation’s risk advisor, working with partners to defend against today’s threats and collaborating to build a secure and resilient infrastructure for the future. Gattoni sets the technical vision and strategic alignment of CISA data and mission services. Previously, he was the Chief of Mission Engineering & Technology, developing analytic techniques and new approaches to increase the value of DHS cyber mission capabilities. Prior to joining DHS in 2010, Gattoni served in various positions at the Defense Information Systems Agency and the United States Army Test & Evaluation Command. He holds a Master of Science Degree in Cyber Systems & Operations from the Naval Postgraduate School in Monterey, California, and is a Certified Information Systems Security Professional (CISSP).
AI Trends: What is the technical vision for CISA to manage risk to federal networks and critical infrastructure?
Brian Gattoni: Our technology vision is built in support of our overall strategy. We are thenation’s risk advisor. It’s our job to stay abreast of incoming threats and opportunities for general risk to the nation. Our efforts are to understand and advise on cyber and physical risks to the nation’s critical infrastructure.
It’s all about bringing in the data, understanding what decisions need to be made and can be made from the data, and what insights are useful to our stakeholders. The potential of AI and machine learning is to expand on operational insights with additional data sets to make better use of the information we have.
What are the most prominent threats?
The sources of threats we frequently discuss are the adversarial actions of nation-state actors and those aligned with nation-state actors and their interests, in disrupting national critical functions here in the U.S. Just in the past month,we’ve seen increased activity from elements supporting what we refer to in the government as Hidden Cobra [malicious cyber activity by the North Korean government]. We’ve issued joint alerts with our partners overseas and the FBI and the DoD, highlighting activity associated with Chinese actors. On CISA.gov people can find CISA Insights, which are documents that provide background information on particular cyber threats and the vulnerabilities they exploit, as well as a ready-made set of mitigation activities that non-federal partners can implement.
What role does AI play in the plan?
Artificial intelligence has a great role to play in the support of the decisions we make as an agency. Fundamentally, AI is going to allow us to apply our decision processes to a scale of data that humans just cannot keep up with. And that’s especially prevalent in the cyber mission. We remain cognizant of how we make decisions in the first place and target artificial intelligence and machine learning algorithms that augment and support that decision-making process. We’ll be able to use AI to provide operational insights at a greater scale or across a greater breadth of our mission space.
How far along are you in the implementation of AI at the CISA?
Implementing AI is not as simple as putting ina new business intelligence tool or putting in a new email capability. Really augmenting your current operations with artificial intelligence is a mix of the culture change, for humans to understand how the AI is supposed to augment their operations. It is a technology change, to make sure you have the scalable compute and the right tools in place to do the math you’re talking about implementing. And it’s a process change.We want to deliver artificial intelligence algorithms that augment our operators’ decisions as a support mechanism.
Where we are in the implementation is closer to understanding those three things.We’re working with partners in federally funded research and development centers, national labs and the department’s own Science and Technology Data Analytics Tech Center to develop capability in this area. We’ve developed an analytics meta-processwhich helps us systemize the way we take in dataand puts us in a position to apply artificial intelligence to expand our use of that data.
Do you have any interesting examples of how AI is being applied in CISA and the federal government today? Or what you are working toward, if that’s more appropriate.
I have a recent use case. We’ve been working with some partners over the past couple of months to apply AI to a humanitarian assistance and disaster relief type of mission. So, within CISA, we also have responsibilities for critical infrastructure.During hurricane season, we always have a role to play in helping advise what the potential impacts are to critical infrastructure sites in the affected path of a hurricane.
We prepared to conduct an experiment leveraging AI algorithms and overhead imagery to figure out if we could analyze the data from a National Oceanic and Atmospheric Administration flight over the affected area. We compared that imagery with the base imagery from Google Earth or ArcGIS and used AI to identify any affected critical infrastructure. We could see the extent to which certain assets, such as oil refineries, were physically flooded. We could make an assessment as to whether they hit a threshold of damage that would warrant additional scrutiny, or we didn’t have to apply resources because their resilience was intact, and their functions could continue.
That is a nice use case, a simple example of letting a computer do the comparisons and make a recommendation to our human operators.We found that it was very good at telling us which critical infrastructure sites did not need any additional intervention. To use a needle in a haystack analogy, one of the useful things AI can help us do is blow hay off the stack in pursuit of the needle. And that’s a win also. The experiment was very promising in that sense.
How does CISA work with private industry, and do you have any examples of that?
We have an entire division dedicated to stakeholder engagement. Private industry owns over 80% of the critical infrastructure in the nation. So CISA sits at the intersection of the private sector and the government to share information, to ensure we have resilience in place for both the government entities and the private entities, in the pursuit of resilience for those national critical functions.Overthe past year we’ve defined a set of 55 functions that are critical for the nation.
When we work with private industry in those areas we try to share the best insights and make decisions to ensure those function areas will continue unabated in the face of a physical or cyber threat.
Cloud computing is growing rapidly. We see different strategies, including using multiple vendors of the public cloud, and a mix of private and public cloud in a hybrid strategy. What do you see is the best approach for the federal government?
In my experience the best approach istoprovide guidance to the CIO’s and CISO’sacross the federal government and allow them the flexibility to make risk-based determinations on their own computing infrastructure asopposed to a one-size-fits-all approach.
We issue a series of use cases that describe—at a very high level—a reference architecture about a type of cloud implementation and where security controls should be implemented, and where telemetry and instrumentation should be applied. You have departments and agencies that have a very forward-facing public citizen services portfolio, which means access to information, is one of their primary responsibilities. Public clouds and ease of access are most appropriate for those.And then there are agencies with more sensitive missions. Those have critical high value data assets that need to be protected in a specific way. Giving each the guidance they need to handle all of their use cases is what we’re focused on here.
I wanted to talk a little bit about job roles. How are you defining the job roles around AI in CISA, as in data scientists, data engineers, and other important job titles and new job titles?
I could spend the remainder of our time on this concept of job roles for artificial intelligence; it’s a favorite topic for me. I am a big proponent of the discipline of data science being a team sport. We currently have our engineers and our analysts and our operators. And the roles and disciplines around data science and data engineers have been morphing out of an additional duty on analysts and engineers into its own sub sector, its own discipline. We’re looking at a cadre of data professionals that serve almost as a logistics function to our operators who are doing the mission-level analysis. If you treat data as an asset that has to be moved and prepared and cleaned and readied, all terms in the data science and data engineering world now, you start to realize that it requires logistics functions similar to any other asset that has to be moved.
If you get professionals dedicated to that end, you will be able to scale to the data problems you have without overburdening your current engineers who are building the compute platforms, or your current mission analysts who are trying to interpret the data and apply the insights to your stakeholders. You will have more team members moving data to the right places, making data-driven decisions.
Are you able to hire the help you need to do the job? Are you able to find qualified people? Where are the gaps?
As the domain continues to mature, as we understand more about the different roles, we begin to see gaps—education programs and training programs that need to be developed. I think maybe three, five years ago, you would see certificates from higher education in data science. Now we’re starting to see full-fledged degrees as concentrations out of computer science or mathematics. Those graduates are the pipeline to help us fill the gaps we currently have. So as far as our current problems, there’s never enough people. It’s always hard to get the good ones and then keep them because the competition is so high.
Here at CISA, we continue to invest not only in our own folks that are re-training, but in the development of a cyber education and training group, which is looking at the partnerships with academia to help shore up that pipeline. It continually improves.
Do you have a message for high school or college students interested in pursuing a career in AI, either in the government or in business, as to what they should study?
Yes and it’s similar to the message I give to the high schoolers that live in my house. That is, don’t give up on math so easily. Math and science, the STEM subjects, have foundational skills that may be applicable to your future career. That is not to discount the diversity and variety of thought processes that come from other disciplines. I tell my kids they need the mathematical foundation to be able to apply the thought processes you learn from studying music or studying art or studying literature. And the different ways that those disciplines help you make connections. But have the mathematical foundation to represent those connections to a computer.
One of the fallacies around machine learning is that it will just learn [by itself]. That’s not true. You have to be able to teach it, and you can only talk to computers with math, at the base level.
So if you have the mathematical skills to relay your complicated human thought processes to the computer, and now it can replicate those patterns and identify what you’re asking it to do, you will have success in this field. But if you give up on the math part too early—it’s a progressive discipline—if you give up on algebra two and then come back years later and jump straight into calculus, success is going to be difficult, but not impossible.
You sound like a math teacher.
A simpler way to say it is: if you say no to math now, it’s harder to say yes later. But if you say yes now, you can always say no later, if data science ends up not being your thing.
Are there any incentives for young people, let’s say a student just out of college, to go to work for the government? Is there any kind of loan forgiveness for instance?
We have a variety of programs. The one that I really like, that I have had a lot of success with as a hiring manager in the federal government, especially here at DHS over the past 10 years, is a program called Scholarship for Service. It’s a CyberCorps program where interested students, who pass the process to be accepted can get a degree in exchange for some service time. It used to be two years; it might be more now, but they owe some time and service to the federal government after the completion of their degree.
I have seen many successful candidates come out of that program and go on to fantastic careers, contributing in cyberspace all over. I have interns that I hired nine years ago that are now senior leaders in this organization or have departed for private industry and are making their difference out there. It’s a fantastic program for young folks to know about.
What advice do you have for other government agencies just getting started in pursuing AI to help them meet their goals?
My advice for my peers and partners and anybody who’s willing to listen to it is, when you’re pursuing AI, be very specific about what it can do for you.
I go back to the decisions you make, what people are counting on you to do. You bear some responsibility to know how you make those decisions if you’re really going to leverage AI and machine learning to make decisions faster or better or some other quality of goodness. The speed at which you make decisions will go both ways. You have to identify your benefit of that decision being made if it’s positive and define your regret if that decision is made and it’s negative. And then do yourself a simple HIGH-LOW matrix; the quadrant of high-benefit, low-regret decisions is the target. Those are ones that I would like to automate as much as possible. And if artificial intelligence and machine learning can help, that would be great. If not, that’s a decision you have to make.
I have two examples I use in our cyber mission to illustrate the extremes here. One is for incident triage. If a cyber incident is detected, we have a triage process to make sure that it’s real. That presents information to an analyst. If that’s done correctly, it has a high benefit because it can take a lot of work off our analysts. It has low–to–medium regret if it’s done incorrectly, because the decision is to present information to an analyst who can then provide that additional filter. So that’s a high benefit, low regret. That’s a no-brainer for automating as much as possible.
On the other side of the spectrum is protecting next generation 911 call centers from a potential telephony denial of service attack. One of the potential automated responses could be to cut off the incoming traffic to the 911 call center to stunt the attack. Benefit: you may have prevented the attack. Regret: potentially you’re cutting off legitimate traffic to a 911 call center, and that has life and safety implications. And that is unacceptable. That’s an area where automation is probably not the right approach. Those are two extreme examples, which are easy for people to understand, and it helps illustrate how the benefit regret matrix can work. How you make decisions is really the key to understanding whether to implement AI and machine learning to help automate those decisions using the full breadth of data.