Advancing AI to Redefine How We Solve the World’s Problems
Maneesh talks about the ways that human-machine interactions are changing, the impact AI has on nearly every industry, and how Verisk leverages innovations in machine learning to solve real-world problems.
My Role
My Role
I’m the head of Verisk’s Innovative Analytics (VIA) Global Research team. We help Verisk’s business units create innovative, top-of-market technologies and services for clients.
We use advanced technology to create assets and workflows that cater to use cases across the business.
We conduct upstream transformational research and development (R&D) in artificial intelligence (AI) and machine learning (ML) with university collaborations.
We create compelling intellectual property for Verisk and visibility in the R&D communities to attract talent.
How My Team Makes a Difference to Customers and Society
How My Team Makes a Difference to Customers and Society
Our work transcends individual, business, and government needs to serve societal needs. Insurance and technological safety protect against risks, both from natural disasters and from human activities like malicious attacks.
My team conducts applied research in AI/ML with university collaborators to create data, models, software modules, and workflows that form the foundation of downstream products.
We work with faculty and graduate students at universities to train the engineers and scientists of tomorrow via summer/semester-long internships or longer residency/post-doctoral stints.
For example, our collaboration with Professor Sriraam Ganapathy at the Indian Institute of Science created learning-based models for diagnosing respiratory pathologies from acoustic data. The driving application is diagnosing COVID-19 from cough recordings and using acoustics for life insurance. In another collaboration, with Professor Terrance Boult at the University of Colorado, Colorado Springs, we created state-of-the-art, deep learning-based models for detecting manipulation of digital imagery, which helps prevent claims fraud and fight fake news.
How Verisk Leverages AI to Solve Real-World Problems
How Verisk Leverages AI to Solve Real-World Problems
The use of AI and ML across Verisk is extensive and more than I can describe here. Some examples follow:
Insurance: to automate underwriting and claims workflows, including information extraction from images, video, text, and documents; forensic defenses against fraudulent claims that use forged digital media; monitoring of the dark web; and detection of websites engaged in fraudulent and illegal activities.
Catastrophe modeling: simulation and creation of large portfolios of extreme weather events, modeling of pandemics, and the impact of policy interventions.
Financial services: to detect fraudulent credit transactions and fraud networks.
Energy: automatic information extraction from scientific documents to identify a variety of environmental, political, and social risks, and strategic forecasting, etc.
How We Foster Innovation at Verisk
How We Foster Innovation at Verisk
My team has a mandate to help Verisk ride the rapid pace of AI technology innovation. We've filed over 30 patent applications in the last three years.
We compete against the top research labs in the world through solving AI/ML challenges and shared tasks (through our academic collaborations). In the last four years, we’ve had over 25 publications in top-tier AI conferences and journals with over 1,200 citations.
We bring the latest research into Verisk and convert that into consumable data, analytics, and technology. Data and metadata assets make their way into Verisk data lakes; state-of-the-art ML models are provisioned into our model stores; code modules are made available through internal development platform repositories; reports are released and used for filing invention disclosures.
Knowledge, skills, and technology assets we create provide novel solutions for addressing problems via collaborative projects with our downstream teams in VIA and other Verisk business units. Sharing our work helps disseminate our learnings into the collaborative business teams and the broader Verisk audience.
Why Verisk
Why Verisk
Our current Chief Analytics Officer led an effort to bootstrap a big-data, cross-business analytics activity at the center, under our Chief Information Officer. He recruited me in 2016 to create an applied R&D team for using AI/ML for images and video analytics to subsequently be scaled to other unstructured (text, documents, credit transaction graphs, etc.) and structured data sets (loss cost data for underwriting).
The opportunity to create a top-of-its-class AI/ML R&D lab producing high research, business, and societal impact was exciting.
Working on diverse sets of problems with multiple use cases let me solve broader classes of problems and be part of the revolution to create Artificial General Intelligence (AGI).
Working alongside the entire technology innovation pipeline gave me an opportunity to take transformative ideas from conception to market deployment. Collaborating with the world’s leading academic R&D groups meant I’d constantly learn from the best.
My Career Growth Over Time
My Career Growth Over Time
I started out as the Director of Image and Video Analytics with a team of two people and a couple of interns.
Now, I head Global Research in AI/ML: a team of approximately 20—comprised of R&D managers, scientists, and engineers—and a rotating staff of faculty scientific advisors from academia, post-docs, residents, and interns.
We’re working on approximately 25 projects with more than 10 Verisk teams and 10+ universities across the U.S., the U.K., and India.
A Typical Day
A Typical Day
The day usually starts with five to six meetings or presentations with management peers and executives for reporting, planning, and guidance—or technical project teams, which may include academic or business collaborators.
I spend most of the remaining time attending to logistical work, authoring research papers, or reviewing and critiquing them. The latter is a pro bono service to the AI/ML research community—conducted as a member of 10+ program committees for top tier conferences—and is essential to cutting-edge research.
I spend the rest of my time reading up on hot-off-the-press AI/ML research.
The Most Significant Advances in AI in the Last 10 Years
The Most Significant Advances in AI in the Last 10 Years
One of the most revolutionary impacts is the rapid proliferation of AI in virtually all technology that surrounds us.
AI has replaced decades of traditional approaches for analyzing unstructured data in entire fields: computer vision (images and videos), natural language processing (text), and automatic speech recognition (voice).
Some significant advances are concerning as well. The rise of General Adversarial Networks (GANs) has led to the spread of so-called deep fakes (images, videos, audio, textual and other data that’s fake but indistinguishable from real data). This lowers the bar for fraud and poses many risks to businesses and society at large via fake news, political propaganda, etc.
While cybersecurity benefits from AI, AI poses a massive risk via malicious attacks.
AI and Machine Learning 10 Years from Now
AI and Machine Learning 10 Years from Now
The proliferation of AI and ML will continue to grow everywhere exponentially: healthcare (medical diagnostics, surgical and pharmaceutical interventions, prosthetics, etc.), personal and commercial property (vehicles, IoT, smart homes, etc.), robotic manufacturing and nanotechnology, technology enablers for people to work, educational assistants, cybersecurity, fraud, and deception for economic and political ends.
Niche domains where AI can outperform humans will rapidly expand. Self-regulated lifelong learning technology– ‘AI agents’ –will proliferate.
Questions concerning statistical risk (including replicability, generalizability, etc.), safety, transparency, and ethics will need to be addressed by insurance agencies and regulatory bodies to mediate how such technology is adopted.
For companies like Verisk, a big impact will be automation of all kinds of workflows. These include the creation of AI-based analytical models and even automation of regulatory workflows overseeing the former. AI will likely also proliferate into our interactions with clients and customers.
The Way Humans and Machines Interact is Changing
The Way Humans and Machines Interact is Changing
A machine is basically a piece of technology with some automation capability, but it’s expected to be more than just a tool: it can execute predefined functionalities automatically with human interaction through a defined interface. As machines advance, automatic functionalities become more complex, non-deterministic, and humanlike.
Human interaction is becoming reduced and human-machine interactions are becoming more abstract, perhaps akin to human-human interactions.
Think of a human-like active agent instead of a machine. The machine will become more proactive. It’ll ask questions like a true collaborator than just executing what it’s told and be capable of learning on the fly. It’ll be cognizant of its failures.
Humans will oversee high-level designs, overall responsibility, and accountability—including responding to changing end-user requirements—while machines will execute more well-defined functionalities. Interfaces will become more natural as well (using natural language, gestures, and freehand illustrations).
My Career Advice
I advise candidates to have a mid-to-long-term focus and to think hard about what they want to do and how they want to grow in their careers.
Candidates should always have a three-to-five-year plan.
Verisk’s diversity of domains, data, and tasks allows for experience and growth that’s perhaps unique in the fintech industry. I advise candidates to plan their time with Verisk accordingly to maximally leverage these opportunities.