Home | MIT Sloan Fellows | Leadership Blog

Category: Machine Learning

How to Build a Supermind

One of the prodigious challenges of the 21stcentury—of perhaps any century—is effectively harnessing the intelligence of a group of people to solve intractable problems. The MIT Center for Collective Intelligence (CCI) is dedicated to exploring that challenge. CCI brings together faculty from across MIT to conduct research on how new communications technologies can enhance the way people work together.

Founded by MIT Sloan Professor Thomas W. Malone, who serves as center director, this first-of-its-kind research effort draws on the strengths of many diverse organizations across the Institute including MIT Sloan, the MIT Media Lab, the MIT Computer Science and Artificial Intelligence Laboratory, and the MIT Department of Brain and Cognitive Sciences. Researchers at CCI are working to achieve a deep, scientific understanding of collective intelligence and to further productive advances across business and society.

One of the center’s most ambitious real-world initiatives is the Collective Intelligence Design Lab(CIDL), which is helping groups develop innovative collective intelligent systems or “superminds” to solve important problems. Researchers from the CIDL work with groups to examine big challenges in new ways and to understand how AI, other digital technologies, and new ways of organizing people could help solve their problems. The idea is to harness the intelligence of the people in a group or organization—and perhaps outside it—to create significant, sustainable solutions.

Tapping all intelligent resources

The CIDL is not focused on the solutions themselves but on helping organizations to develop superminds that can solve problems now and in the future. Nor are they focused simply on new information technologies. The CIDL team works to organize people in new ways so that the smart integration of human and artificial intelligence can lead to more inventive solutions. Drawing on a wide range of resources, from software design to crowdsourcing, economics to artificial intelligence, cognitive science to organization theory, CIDL researchers are working to reinvent existing problem-solving models.

Thomas W. Malone, the Patrick J. McGovern Professor of Management, was also the founder and director of the MIT Center for Coordination Science and one of the two founding codirectors of the MIT Initiative on Inventing the Organizations of the 21st Century.  He teaches classes on organizational design, information technology, and leadership. His research focuses on how new organizations can be designed to take advantage of the possibilities provided by information technology

The CIDL is seeking sponsoring organizations—companies, governments, nonprofits—with interesting problems to solve and a commitment to discovering new, collective methods of solving them. Learn more about participating in the MIT Collective Intelligence Design Lab.

Problem-led leadership: A new entrepreneurial model

CEO, entrepreneur, and theoretical neuroscientist, Vivienne Ming believes we should—and will—embrace cyborgs. Within the next generation, she recently told a packed audience at MIT Sloan, cognitive neuroprosthetics will “fundamentally change the definition of what it means to be human.” Cofounder of the machine learning company Socos Labs and a visiting scholar at UC Berkeley’s Redwood Center for Theoretical Neuroscience, Ming’s goal is to solve sticky dilemmas at the intersection of advanced technology, learning, and labor economics.

Named one of Ten Women to Watch in Tech by Inc. magazine, Ming is renowned for her heady predictions about the future direction of tech, but it’s her leadership model that most intrigues MIT Leadership Center Executive Director Hal Gregersen and Faculty Director Deborah Ancona. They expound on her distinctive style in a recent Harvard Business Review post, because it’s a model they are observing more and more in contemporary C-suites.

Don’t do as I say. Do what I can’t do.

Gregersen and Ancona say that Ming has come to the conclusion that she can make her strongest contributions as an individual, rather than as a team booster. “For a long time, I tried to be the whole package. I put a lot of energy into making certain that I was shepherding everyone along, doing all the right things for my teams. Then I realized: You know what? If I can get some people that are really good at the things that I’m not, then I can focus on my strengths. And my strengths are in creative problem solving — all the way down to writing the code myself.”

As directors of the MIT Leadership Center, Gregersen and Ancona have been trying to get to the bottom of this new style. Is it a trend? A future best practice? “We weren’t sure if it was because we spent so much time with MIT-trained people,” they note in their Harvard Business Review post, “or if there was a much more widespread shift under way, but the people we saw driving impactful, world-changing initiatives just didn’t look like old-school leadership material—and didn’t seem to want to. Cautiously, we called it problem-led leadership and launched into all the interviewing, case studying, and literature review that goes into a leadership research project.”

Gregersen and Ancona found several common threads in the work of problem-led leaders. Most noteworthy, they say, is that none of these leaders appear to harbor any expectations that they will attract “followers” by the sheer power of their charisma or status. Instead, they note, “their method is to get others excited about whatever problem they have identified as ripe for a novel solution.” They take a leadership role only to bring together the problem-solvers necessary to reach a solution. For Ming, the style is simply a tool, a means to an end. “The only reason I do it is because it is an amazingly effective way to have an impact on the world.”

Read more in the MIT Sloan Experts blog.

Read the full post at Harvard Business Review.

Intentional analytics

How fresh is your data? Do you know why you gathered it in the first place? Is there a rhyme to your reason when it comes to analytics? Abhi Yadav, SF ’13, launched the MIT spinout ZyloTech because he realized that even the best data-educated personnel at major companies were unable to deal with the continual stream, variety, and mind-bending complexity of omnichannel customer data.

ZyloTech was actually born in the New Enterprise class taught by MIT Sloan Professor Bill Aulet. Yadav then recruited Michael Cusumano, who taught his Business of Software class, to the company’s board. Along with a team of data scientists, engineers, and digital marketers from the Cambridge ecosystem, Yadav wanted to make it possible for companies to leverage all their customer data in near real time so as to continuously access advanced customer analytics that deliver vastly more accurate and actionable insights.

“It’s futile to try and get good results from a marketing campaign when you’re working off old, incomplete data and ad-hoc analytics,” Yadav says. “What we’re doing is bottom-up analytics. We are unifying and curating a customer’s identity, which includes past behavior, intent-based data points, and basic contact info. We continuously track each existing customer action as it’s happening to determine what that customer likes an doesn’t like and what their signature behaviors are.”

Leveraging MIT research

Yadav and his team tapped MIT research in consumer science and automated machine learning to create a proprietary technology that performs entity resolution while integrating a probabilistic and a deterministic data unification approach. “When you combine these two approaches with deep-learning (AI) to discover patterns,” he explains, “you attain an unprecedented level of knowledge about your customer from raw data.”

Setting aside the technical terms, what Yadav and his team are doing is distilling all that information to get the real juice out of it, to take timely action, and to discover what a company really needs to know about the individualized motivations, habits, and predilections of its customers. As a result, they will be able to offer individualized promotions at the right time and through the right channel.

Given the volume and variety of data getting generated every second, Yadav says, it’s essential to make the most of it through timely insights. “We see businesses getting frustrated with the classic modern challenge of big data versus big insights,” he notes. “They don’t see where it’s getting them, because running after IT or hiring lots of engineers has not furthered their objectives. My goal is to help a business go beyond the lip service on customer centricity with real customer-centric marketing that unlocks the riches that lie in customer data. The result: a better, smarter experience for consumers—and for the companies that hope to win them.”


Tapping data to gain an analytic edge

Competitive companies have been dutifully gathering data for years, many of them amassing an extensive and revealing body of analytics. The reality, however, is that a shocking number of those organizations just aren’t quite sure what to do with that information. As a result, the data often remains untouched, untapped, and uninterpreted says Taylor Reynolds, SF ’15, Technology Policy Director of MIT’s Internet Policy Research Initiative (IPRI).

“AI has been around for a long time, but the recent advances in scale open an almost infinite range of new possibilities,” Reynolds says. “The bottomless storage available through cloud computing as well as the complexity of calculations we can do now make it possible to store data and crunch numbers on a scale we never knew possible.”

Reynolds notes the importance of tapping this new avenue of information. “There’s so much low-hanging fruit out there in terms of revealing data. And there’s power in that data. Companies leveraging that information have an advantage over those that don’t. In fact, if we were to freeze technological development right here and now, and we had to live with any advances that have already taken place, we’d still likely have ten years of productivity gains we could make with our untapped data.”

And data analytics aren’t just a matter of due diligence, Reynolds says. They can be the key to transformational innovations. He points to work being done at MIT at the Laboratory for Social Machines, the Moral Machine, and the Machine Understanding Group, which is part of MIT’s Internet Policy Research Initiative. “Researchers at MIT are working on projects like a self-driving car that can explain itself. If it gets into an accident, the vehicle will be able to provide a detailed analysis of why it made the decisions it made: ‘I was starting to make a left turn but took evasive action because I saw a pedestrian.’ That’s critical information in the development of systems that will hold life-or-death responsibility. We’re not quite there, yet, but that’s where we’re headed.”

Can data be biased?

Reynolds, who often helps policymakers address cybersecurity and Internet public policy challenges, notes that data also poses dangers for society—for example, when inherent biases are built into algorithms. He cites the work of investigative tech reporter and machine bias expert Julia Angwin. Angwin and her team at ProPublica revealed that an algorithm employed by the criminal justice system to predict repeat criminals had been designed with inherent racial biases, consistently assigning high risk scores to blacks who did not merit that distinction. “People aren’t perfect,” Reynolds says, “and if people aren’t perfect, neither are the algorithms they design. If a person is biased, the algorithm may be built with that bias. That’s an authentic risk of AI that we, as a society, have to guard against.”

Reynolds, in his role at IPRI, is pulling together researchers and students from departments and labs across the Institute to increase the trustworthiness and effectiveness of interconnected digital systems. The initiative just made news by awarding $1.5 million to researchers across campus working on Internet policy and cybersecurity-related research projects.

Read the IPRI blog.

AI to the Rescue: Recovering Lost Customers

Alan Ringvald MIT SF '16Is a customer who has been gone a few months a lost cause? Not according to Alan Ringvald, SF ’16, founder and CEO of the startup Relativity6. Ringvald believes that the time lapsed is not actually the most helpful metric. The focus, he says, should be less on how long they’ve been gone and more on why they went and what you need to do to get them back.

Ringvald and company cofounder and CTO Abraham Rodriguez, SF ’16, share a long obsession with decoding the behavior of lapsed customers. The seasoned entrepreneurs launched Relativity6 while collaborating on their Sloan Fellows’ master’s thesis, which explored the reactivation of unresponsive customers. Their research points to one silver bullet: machine learning. “We are trying to teach a machine to think like a human, to conclude from a customer’s past purchasing actions when and what they might purchase next.”

Relativity6 looks for the hidden variables that will reveal why a customer has been  inactive. “Companies lose a lot of customers, and they won’t get them all back. We’re looking for the ones with the highest likelihood of returning. We examine customers’ historical transactions and purchasing behavior. Over time, we find behavioral similarities.”

Not your grandmother’s demographics

Ringvald also believes that the marketplace has been guilty of an over-reliance on demographics. “Our culture is no longer so rigidly segmented by age. Game of Thrones might be the favorite TV show of a 25-year-old student in Michigan and a 70-year-old retiree in San Diego—and both might have downloaded a related feature. Their shared interests may well be more illuminating than their disparate ages when analyzing purchasing trends.”

Relativity6 is poised to help any organization with a sufficient customer database that has been collecting data for more than two or three years. The founders are so confident in their premise that they set up the company using a pay-by-performance model, reducing the risk for prospective customers. Establishing a 90 percent accuracy rate hasn’t hurt either, nor does the 5 percent average increase in revenue streams they’ve been delivering. And the model works as well for business-to-business as it does for business-to-consumer challenges.

Although a young company, Relativity6 has worked with small businesses and mega-companies representing a wide range of markets, including retailers, financial institutions, insurance agencies, hospitals, political organizations, universities, and nonprofits. NutraClick, a technology-driven company that provides leading health and wellness products, engaged Relativity6 to reactivate subscription customers and tripled their ROI in just one month.

The startup’s MIT roots run deep. In addition to Ringvald and Rodriguez, the team includes two additional Sloan Fellows alumni, Silvana Lopez Diaz, SF ’16, and Aaron Howell, SF ’15. MIT Sloan professor Duncan Simester sits on the board. MIT’s Industrial Liaison Program and the MIT Sandbox Innovation Fund have provided pivotal support.


Kidney-matching by algorithm

According to the National Kidney Foundation, thirteen people die every day while awaiting a kidney transplant. More than 3,000 new patients are added to the waiting list every month—a new name every 14 minutes. But the length of the waiting list and the insufficient supply aren’t the only issues in those deaths. The entire system is slowed by a time-consuming decision-making process that relies on individual discernment. “Who might be best suited to this kidney?” “Is this kidney the best possible match?” “Will a better match be coming in the next few months?”

Dimitris Bertsimas

Dimitris Bertsimas, the Boeing Leaders for Global Operations Professor of Management and the co-director of the MIT Sloan Operations Research Center, is cutting through red tape with an elegant algorithm designed to streamline the waiting list process, getting the right kidney to the right recipient in the shortest amount of time. In a new paper, he and MIT Sloan Assistant Professor of Operations Management Nikos Trichakis describe a pioneering model that applies machine-learning to historical data about all kidney transplants over the last decade to guide future donations.

Nikos Trichakis

At present, when a kidney is offered to a wait-listed candidate, the decision to accept or decline the organ relies primarily upon a surgeon’s experience and intuition. The physician might take into consideration the location and condition of the kidney. And might there be a higher-quality kidney or a better match available in the future? The authors maintain that the current experience-based paradigm lacks scientific rigor and is subject to the inaccuracies that plague anecdotal decision-making. As a result, as many as 20% of all kidneys obtained are discarded as unsuitable—when, in fact, they might well have been the best option.

Bertsimas’ and Trichakis’ data-driven analytics-based model predicts whether a patient will receive an offer for a deceased-donor kidney at KDPI thresholds of 0.2, 0.4, and 0.6, and at time frames of 3, 6, and 12 months. The model accounts for OPO, blood group, wait time, DR antigens, and prior offer history to provide accurate and personalized predictions. They tested datasets spanning various lengths of time to understand the adaptability of the method.

The pair is working with surgeons at Massachusetts General Hospital to create a support tool that leverages their model. They hope to give surgeons a reality check about kidneys, providing them with hard evidence of whether they can realistically expect a better donation if they decline a kidney—ultimately reducing the number of kidneys that are discarded because physicians are pessimistic about the match.

Find out more about their research.

Read the abstract.