How five alumni are using artificial intelligence to make the world a better place.
Last November, research lab OpenAI publicly released ChatGPT — its artificial intelligence chat program — and ignited an AI mania that shows no signs of abating. This new breed of AI is driven by so-called “large language models,” which ingest massive amounts of textual data and use it to produce unique responses to user input. ChatGPT and other large language models are already transforming industries ranging from software development and finance to education and advertising. But ChatGPT and similar systems are just one facet of the AI revolution. Colgate Magazine spoke with five alumni who are using AI to make the world a better place by improving our mental health, protecting us from sophisticated cybercrime, finding solutions to our climate crisis, developing lifesaving drugs, and driving economic prosperity.
AI Is Transforming Mental Health
It’s lonely out in space. For decades, NASA has known that depression, anxiety, and other mental health issues may naturally arise when astronauts spend weeks or months orbiting hundreds of miles away from the Earth’s surface. As the space agency began to lay the groundwork for a crewed mission to Mars, it was apparent that finding a way to support the mental health of astronauts on long-duration spaceflight would be critical to the success of the most ambitious space mission ever undertaken. But in a tightly packed spacecraft, there’s no room for a therapist to fly along.
Approximately 15 years ago, Claudia Zayfert ’83, who was the director of the Anxiety Disorders Clinic at Dartmouth’s Geisel School of Medicine, teamed up with a colleague from Harvard Medical School who had received a grant from NASA to develop an AI system that could autonomously manage astronaut health on long-duration spaceflight missions. “It was designed with the Mars mission in mind, where you’re so far from Earth that you can’t have a conversation in real time,” Zayfert says. In an ideal scenario, a flight to Mars takes 21 months round trip — a long and dangerous journey, especially if astronauts lack access to personalized mental health support. It seemed like an area where an AI therapist might do a lot of good.
While collaborating on the NASA project, Zayfert and her colleague hatched a plan to apply the same methodologies used for the spacefaring AI therapist to people on Earth with limited access to therapists and other mental health professionals.
In 2019, Zayfert co-founded Evermind to commercialize the AI technology that had been developed with spacefaring applications in mind to provide access to mental health support for the multitudes of earthlings who — like an astronaut on a mission to the Red Planet — don’t have access to an in-person therapist. Evermind’s flagship product is ePST, a virtual, AI-assisted therapy that uses hundreds of short audio and video clips of a real, expert psychologist sitting in a warmly lit room to help treat depression.
Although the virtual therapist in Evermind’s videos is a real mental health professional, his pre-recorded responses are selected by an AI system based on a user’s input. Each week, the AI therapist learns more about the user based on their feedback about how the treatment is progressing. It integrates that feedback and advises them on skills to practice and steps to keep moving forward. Although there are dozens of companies providing AI-driven therapy today, most of these systems use generative AI, which produces novel responses to user input. While generative AI is useful in the case of a tool like ChatGPT for coming up with new ideas, it can be ineffective or even dangerous in a therapeutic context because the AI program might provide false or potentially harmful information to a patient.
At Evermind, Zayfert and her colleagues took a totally different approach to AI-driven therapy that keeps the best part of virtual automated mental health–support services while providing guardrails to ensure they are effective and safe for users. ePST is based on a type of AI known as an “expert system,” which is designed to emulate the thought process of a human expert — in this case, a therapist. The AI operates across a database of responses vetted by human therapists and uses well-established therapy protocols to intelligently select the best response for a given user input. This ensures the AI program stays on script and provides mental health support in a way that has been proven to help patients grappling with depression.
“The system isn’t generating its own responses,” Zayfert says. “All the responses are preprogrammed, and it follows a rigorous, evidence-based protocol every step of the way.”
Whether you’re an astronaut hurtling through the vastness of space toward another planet or one of the billions of people back on Earth, Zayfert believes that thoughtfully designed AI systems have an important role to play in helping us deal with our all-too-human problems. She says the point of applying AI to mental health is not to replace human therapists, but to empower individuals to learn the skills they need to cope with their mental health challenges. “This is a tool kit you carry around in your head, that you put to work in your life, and that’s what good therapy does,” she says. “It teaches you something that you carry with you, not spoon-feeding you the answers to life’s problems.”
AI Is Saving the Planet
In July, the U.N. secretary-general declared the world was entering an era of “global boiling” as countries broke heat records and experienced unprecedented wildfires and cataclysmic flooding. Finding new ways to deploy technology to stave off the worst impacts of climate change has never been more urgent, and Anthony Annunziata ’05 believes that AI may prove to be a decisive factor in our ability to survive on a warming planet.
As the director of product and business incubation, IBM’s AI and Quantum Accelerated Sciences division, Annunziata is leading a team of researchers who are experimenting with ways to leverage the company’s cutting-edge AI systems to tackle a host of large-scale problems in the realm of scientific discovery, health care, and climate. After graduating from Colgate with a degree in physics, Annunziata spent his early career seeking ways to apply his technical skills for social good at institutions, including MIT, Cambridge, and Yale. “I went into Colgate not knowing if physics could even be a viable major choice, never mind a career choice,” he says. “I emerged convinced that this is the thing I wanted to do in grad school and beyond.”
In 2017, Annunziata launched IBM’s quantum computing business, and since then he and his colleagues have been searching for ways to use quantum computing and AI to accelerate scientific discovery. If a lab needs a better catalyst to perform a chemical reaction, for example, or create a new-and-improved energy storage method, Annunziata and his team can train their AI on existing scientific literature and use that model to form new hypotheses and predict the outcomes of experiments. “I have a portfolio at IBM that’s trying to research and develop ways to use AI for these really tough societal challenges,” he says.
In the context of climate science, Annunziata and his crew at IBM are using AI for a host of applications that range from identifying promising new materials for carbon capture technologies or making improved predictions about how climate change could affect a particular region’s rainfall, soil, or cloud cover. He sees AI as a uniquely useful tool to apply to climate science because it is so multidisciplinary and the problem space is so complex. But connecting the dots between various fields of scientific research and technology development is an area where AI, with its superhuman ability to analyze data, can give human scientists and technologists a big advantage.
“AI isn’t going to be able to come up with all the solutions, but it will help us find solutions to societal problems faster, whether they are diseases, energy consumption and storage, or understanding how best to target climate change mitigation,” Annunziata says. “I think that’s going to have an impact across lots of areas. I’m a big optimist here.”
AI Is Accelerating the Development of Lifesaving Drugs
During the earliest days of the COVID-19 pandemic, the United States marshaled its scientific, technological, and industrial resources to develop three novel vaccines in record time as part of Operation Warp Speed. It was an unprecedented feat in terms of the amount of time between the emergence of a pathogen and the deployment of a vaccine, but if George Armstrong ’18 has his way, AI will make this sort of accelerated drug discovery routine — and save countless lives along the way.
As a deep learning engineer at Nvidia, Armstrong is helping the company find ways to enable researchers to use AI for life science and health care applications including AI models that could someday help with the development of precision drugs that are tailored to an individual patient’s genetic profile. It’s a skill set he honed at Colgate, where he contributed to a research project run by Professor Ahmet Ay that used machine learning to improve cancer subtyping — a way of grouping cancers based on their molecular characteristics. This project led to subsequent research on advanced computing approaches for studying the microbiome — the collection of microorganisms that live on our bodies and in our guts — throughout Armstrong’s PhD program at the University of California San Diego.
Today, he is developing techniques that use AI to rapidly analyze human DNA and use this information to identify new drug candidates for a wide variety of diseases. “One of the big things AI is doing right now for life sciences is helping to predict the structure of proteins,” Armstrong says. “That way you can better predict whether a drug is going to induce a change we want and not induce a change we don’t want.”
It’s a radically new approach to the traditional drug discovery paradigm, which typically involves running expensive and time-consuming computer simulations to identify promising new therapeutic molecules. But by leveraging large language models — the same type of AI that underpins applications like ChatGPT — this process can be done more rapidly and at a fraction of the cost. To put the magnitude of this new approach to drug discovery in perspective, Armstrong says that it can take “years and years” of compute time to identify the structure of a single promising protein. With AI, a similar result can be delivered in as little as 30 minutes.
“AI isn’t just impacting the speed of innovation; it can also be far more energy efficient for drug discovery,” Armstrong says. Energy-efficient computer systems will be especially important in a world in the throes of climate change — training the third generation of ChatGPT, for example, is estimated to have released more than 500 tons of carbon dioxide into the atmosphere — but this must be balanced with the profound benefits for the life sciences that will be driven by AI systems.
AI Is Driving Business Innovation and Worker Productivity
In April, Goldman Sachs released a report detailing how AI could create almost $7 trillion in value over the next decade and raise global gross domestic product by more than 7%. There’s certainly a lot of money to be made, but for Micah Kotch ’98, a partner at Blackhorn Ventures, there’s no reason that this wave of innovation can’t help digitize and decarbonize our largest industries in the process. Indeed, he believes it’s the businesses that use AI to reduce operating expenses and emissions while simultaneously improving labor productivity that stand to be the biggest winners of this new technological paradigm.
“We invest in enterprise SaaS [software as a service] start-ups building digital infrastructure for industries that power, move, and build our world, industries that generate $3T in annual revenue and account for approximately 90% of U.S. greenhouse gas emissions,” Kotch says. “Our investors are looking for impact and understand that digitization and decarbonization go hand in hand.”
At Blackhorn, Kotch and his partners invest in start-ups that use AI to drive resource efficiency in sectors, including construction, energy, supply chain logistics, and transportation. In a world where every business is looking for ways to incorporate AI into their products and workflows, Kotch says proprietary data is king and a key differentiating factor for businesses at the forefront of AI-driven technological innovation. “Companies generating their own unique datasets can leverage the full potential of large language models,” Kotch says.
AI isn’t just good for business, however. It also empowers frontline workers at companies whose work is vital to keeping America humming. As an example, Kotch points to Datch.io, a Blackhorn portfolio company that uses AI voice technology to enable “deskless” workers to talk through their everyday tasks and turn their speech into structured data. This saves countless hours that workers would otherwise spend doing data entry, providing more time to focus on manufacturing plant maintenance or installing electric utility infrastructure.
Blackhorn’s investment thesis means that Kotch has seen firsthand how AI is dramatically transforming a wide range of businesses that are critical to the U.S. economy, but rarely thought about in our day-to-day lives. Rail Vision, for example, is maximizing train efficiency by using AI to inform train operators about the best time to accelerate or brake, which has led to safety improvements, massive cost savings on fuel, and emissions reductions of up to 30%.
Kotch says these companies are examples of how venture capitalists like himself can simultaneously generate strong returns for investors while accelerating industrial transformation — and decarbonization of America’s largest and most important industries.
AI Is Protecting Businesses From a New Breed of Cybercrime
Earlier this year, the FBI released its annual internet crime report, which documented a “staggering” rise in cybercrimes that cost American consumers and businesses more than $10 billion. One of the largest areas of cybercrime is known as “business email compromise,” a scam where hackers impersonate company employees and convince unwitting victims to transfer them large sums of money. While this type of attack isn’t new, hackers are increasingly using AI to increase the scale and sophistication of their attacks. Although AI is driving a new wave of cybercrime, it may also be our defense.
As the vice president of legal at Abnormal Security, a cybersecurity company that uses AI to fight scammers before they get the chance to exploit companies, Evan LeBon ’05 has had a front-row seat to the emergence of this AI-driven cybercrime. Since he joined the company in 2021, LeBon has seen the company’s AI-driven security systems stop attackers using AI to commit invoice fraud, impersonate company employees and executives, and even launch targeted attacks on university students to convince them to fork over their student loan payments to criminals.
“Emails and cloud communications are a huge attack vector for nation state actors and privately backed syndicates of shady actors,” LeBon says. “These systems are notoriously hard to secure because they’re very open by nature.”
The old way of preventing these kinds of attacks on corporations and institutions involved identifying attackers’ IP addresses or other unique signatures of their communications, and blocking them as they came in. Those methods were one size fits all and didn’t take into consideration the specific communications happening within an organization. By using AI, companies like Abnormal can learn how an institution’s communications normally operate and tailor its defenses to a wider but more nimble range of AI-driven attacks.
“We are seeing the problem get worse because the level of sophistication of the attacks is exponentially growing,” LeBon says. “And a lot of that is because attackers are using the exact same tools we’re using to provide security services.”
The benefit of AI systems, says LeBon, is that they can analyze massive amounts of data much faster and more accurately than humans can when identifying communication patterns that look suspicious. When Abnormal’s AI defenders identify a suspect request, they can immediately flag it for a company’s security team for further investigation. Although the AI-driven techniques used by cybercriminals are constantly evolving, LeBon believes it’s possible to stay one step ahead of hackers by using the same AI systems for good.
“I think it’s so important for people to look at the positive use cases of AI,” LeBon says. “Sure, AI can be used for bad, but it can also be used for good. We need to look at both sides of the problem and understand how we’ll engage with it as humans and what kinds of limits we should place on it.”