Finger-shaped sensor enables more dexterous robots

Imagine grasping a heavy object, like a pipe wrench, with one hand. You would likely grab the wrench using your entire fingers, not just your fingertips. Sensory receptors in your skin, which run along the entire length of each finger, would send information to your brain about the tool you are grasping.

In a robotic hand, tactile sensors that use cameras to obtain information about grasped objects are small and flat, so they are often located in the fingertips. These robots, in turn, use only their fingertips to grasp objects, typically with a pinching motion. This limits the manipulation tasks they can perform.

MIT researchers have developed a camera-based touch sensor that is long, curved, and shaped like a human finger. Their device provides high-resolution tactile sensing over a large area. The sensor, called the GelSight Svelte, uses two mirrors to reflect and refract light so that one camera, located in the base of the sensor, can see along the entire finger’s length.

In addition, the researchers built the finger-shaped sensor with a flexible backbone. By measuring how the backbone bends when the finger touches an object, they can estimate the force being placed on the sensor.

They used GelSight Svelte sensors to produce a robotic hand that was able to grasp a heavy object like a human would, using the entire sensing area of all three of its fingers. The hand could also perform the same pinch grasps common to traditional robotic grippers.

“Because our new sensor is human finger-shaped, we can use it to do different types of grasps for different tasks, instead of using pinch grasps for everything. There’s only so much you can do with a parallel jaw gripper. Our sensor really opens up some new possibilities on different manipulation tasks we could do with robots,” says Alan (Jialiang) Zhao, a mechanical engineering graduate student and lead author of a paper on GelSight Svelte.

Zhao wrote the paper with senior author Edward Adelson, the John and Dorothy Wilson Professor of Vision Science in the Department of Brain and Cognitive Sciences and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL). The research will be presented at the IEEE Conference on Intelligent Robots and Systems.

Mirror mirror

Cameras used in tactile sensors are limited by their size, the focal distance of their lenses, and their viewing angles. Therefore, these tactile sensors tend to be small and flat, which confines them to a robot’s fingertips.

With a longer sensing area, one that more closely resembles a human finger, the camera would need to sit farther from the sensing surface to see the entire area. This is particularly challenging due to size and shape restrictions of a robotic gripper.

Zhao and Adelson solved this problem using two mirrors that reflect and refract light toward a single camera located at the base of the finger.

GelSight Svelte incorporates one flat, angled mirror that sits across from the camera and one long, curved mirror that sits along the back of the sensor. These mirrors redistribute light rays from the camera in such a way that the camera can see the along the entire finger’s length.

To optimize the shape, angle, and curvature of the mirrors, the researchers designed software to simulate reflection and refraction of light.

“With this software, we can easily play around with where the mirrors are located and how they are curved to get a sense of how well the image will look after we actually make the sensor,” Zhao explains.

The mirrors, camera, and two sets of LEDs for illumination are attached to a plastic backbone and encased in a flexible skin made from silicone gel. The camera views the back of the skin from the inside; based on the deformation, it can see where contact occurs and measure the geometry of the object’s contact surface.

In addition, the red and green LED arrays give a sense of how deeply the gel is being pressed down when an object is grasped, due to the saturation of color at different locations on the sensor.

The researchers can use this color saturation information to reconstruct a 3D depth image of the object being grasped.

The sensor’s plastic backbone enables it to determine proprioceptive information, such as the twisting torques applied to the finger. The backbone bends and flexes when an object is grasped. The researchers use machine learning to estimate how much force is being applied to the sensor, based on these backbone deformations.

However, combining these elements into a working sensor was no easy task, Zhao says.

“Making sure you have the correct curvature for the mirror to match what we have in simulation is pretty challenging. Plus, I realized there are some kinds of superglue that inhibit the curing of silicon. It took a lot of experiments to make a sensor that actually works,” he adds.

Versatile grasping

Once they had perfected the design, the researchers tested the GelSight Svelte by pressing objects, like a screw, to different locations on the sensor to check image clarity and see how well it could determine the shape of the object.

They also used three sensors to build a GelSight Svelte hand that can perform multiple grasps, including a pinch grasp, lateral pinch grasp, and a power grasp that uses the entire sensing area of the three fingers. Most robotic hands, which are shaped like parallel jaw drippers, can only perform pinch grasps.

A three-finger power grasp enables a robotic hand to hold a heavier object more stably. However, pinch grasps are still useful when an object is very small. Being able to perform both types of grasps with one hand would give a robot more versatility, he says.

Moving forward, the researchers plan to enhance the GelSight Svelte so the sensor is articulated and can bend at the joints, more like a human finger.

“Optical-tactile finger sensors allow robots to use inexpensive cameras to collect high-resolution images of surface contact, and by observing the deformation of a flexible surface the robot estimates the contact shape and forces applied. This work represents an advancement on the GelSight finger design, with improvements in full-finger coverage and the ability to approximate bending deflection torques using image differences and machine learning,” says Monroe Kennedy III, assistant professor of mechanical engineering at Stanford University, who was not involved with this research. “Improving a robot’s sense of touch to approach human ability is a necessity and perhaps the catalyst problem for developing robots capable of working on complex, dexterous tasks.”

This research is supported, in part, by the Toyota Research Institute.

de MIT News https://ift.tt/35E6MsB

AI copilot enhances human precision for safer aviation

Imagine you're in an airplane with two pilots, one human and one computer. Both have their “hands” on the controllers, but they're always looking out for different things. If they're both paying attention to the same thing, the human gets to steer. But if the human gets distracted or misses something, the computer quickly takes over.

Meet the Air-Guardian, a system developed by researchers at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). As modern pilots grapple with an onslaught of information from multiple monitors, especially during critical moments, Air-Guardian acts as a proactive copilot; a partnership between human and machine, rooted in understanding attention.

But how does it determine attention, exactly? For humans, it uses eye-tracking, and for the neural system, it relies on something called "saliency maps," which pinpoint where attention is directed. The maps serve as visual guides highlighting key regions within an image, aiding in grasping and deciphering the behavior of intricate algorithms. Air-Guardian identifies early signs of potential risks through these attention markers, instead of only intervening during safety breaches like traditional autopilot systems. 

The broader implications of this system reach beyond aviation. Similar cooperative control mechanisms could one day be used in cars, drones, and a wider spectrum of robotics.

"An exciting feature of our method is its differentiability," says MIT CSAIL postdoc Lianhao Yin, a lead author on a new paper about Air-Guardian. "Our cooperative layer and the entire end-to-end process can be trained. We specifically chose the causal continuous-depth neural network model because of its dynamic features in mapping attention. Another unique aspect is adaptability. The Air-Guardian system isn't rigid; it can be adjusted based on the situation's demands, ensuring a balanced partnership between human and machine."

In field tests, both the pilot and the system made decisions based on the same raw images when navigating to the target waypoint. Air-Guardian’s success was gauged based on the cumulative rewards earned during flight and shorter path to the waypoint. The guardian reduced the risk level of flights and increased the success rate of navigating to target points. 

"This system represents the innovative approach of human-centric AI-enabled aviation," adds Ramin Hasani, MIT CSAIL research affiliate and inventor of liquid neural networks. "Our use of liquid neural networks provides a dynamic, adaptive approach, ensuring that the AI doesn't merely replace human judgment but complements it, leading to enhanced safety and collaboration in the skies."

The true strength of Air-Guardian is its foundational technology. Using an optimization-based cooperative layer using visual attention from humans and machine, and liquid closed-form continuous-time neural networks (CfC) known for its prowess in deciphering cause-and-effect relationships, it analyzes incoming images for vital information. Complementing this is the VisualBackProp algorithm, which identifies the system's focal points within an image, ensuring clear understanding of its attention maps. 

For future mass adoption, there's a need to refine the human-machine interface. Feedback suggests an indicator, like a bar, might be more intuitive to signify when the guardian system takes control.

Air-Guardian heralds a new age of safer skies, offering a reliable safety net for those moments when human attention wavers.

"The Air-Guardian system highlights the synergy between human expertise and machine learning, furthering the objective of using machine learning to augment pilots in challenging scenarios and reduce operational errors," says Daniela Rus, the Andrew (1956) and Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT, director of CSAIL, and senior author on the paper.

"One of the most interesting outcomes of using a visual attention metric in this work is the potential for allowing earlier interventions and greater interpretability by human pilots," says Stephanie Gil, assistant professor of computer science at Harvard University, who was not involved in the work. "This showcases a great example of how AI can be used to work with a human, lowering the barrier for achieving trust by using natural communication mechanisms between the human and the AI system."

This research was partially funded by the U.S. Air Force (USAF) Research Laboratory, the USAF Artificial Intelligence Accelerator, the Boeing Co., and the Office of Naval Research. The findings don't necessarily reflect the views of the U.S. government or the USAF.

de MIT News https://ift.tt/CGp7BLS

AI copilot enhances human precision for safer aviation

Imagine you're in an airplane with two pilots, one human and one computer. Both have their “hands” on the controllers, but they're always looking out for different things. If they're both paying attention to the same thing, the human gets to steer. But if the human gets distracted or misses something, the computer quickly takes over.

Meet the Air-Guardian, a system developed by researchers at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). As modern pilots grapple with an onslaught of information from multiple monitors, especially during critical moments, Air-Guardian acts as a proactive copilot; a partnership between human and machine, rooted in understanding attention.

But how does it determine attention, exactly? For humans, it uses eye-tracking, and for the neural system, it relies on something called "saliency maps," which pinpoint where attention is directed. The maps serve as visual guides highlighting key regions within an image, aiding in grasping and deciphering the behavior of intricate algorithms. Air-Guardian identifies early signs of potential risks through these attention markers, instead of only intervening during safety breaches like traditional autopilot systems. 

The broader implications of this system reach beyond aviation. Similar cooperative control mechanisms could one day be used in cars, drones, and a wider spectrum of robotics.

"An exciting feature of our method is its differentiability," says MIT CSAIL postdoc Lianhao Yin, a lead author on a new paper about Air-Guardian. "Our cooperative layer and the entire end-to-end process can be trained. We specifically chose the causal continuous-depth neural network model because of its dynamic features in mapping attention. Another unique aspect is adaptability. The Air-Guardian system isn't rigid; it can be adjusted based on the situation's demands, ensuring a balanced partnership between human and machine."

In field tests, both the pilot and the system made decisions based on the same raw images when navigating to the target waypoint. Air-Guardian’s success was gauged based on the cumulative rewards earned during flight and shorter path to the waypoint. The guardian reduced the risk level of flights and increased the success rate of navigating to target points. 

"This system represents the innovative approach of human-centric AI-enabled aviation," adds Ramin Hasani, MIT CSAIL research affiliate and inventor of liquid neural networks. "Our use of liquid neural networks provides a dynamic, adaptive approach, ensuring that the AI doesn't merely replace human judgment but complements it, leading to enhanced safety and collaboration in the skies."

The true strength of Air-Guardian is its foundational technology. Using an optimization-based cooperative layer using visual attention from humans and machine, and liquid closed-form continuous-time neural networks (CfC) known for its prowess in deciphering cause-and-effect relationships, it analyzes incoming images for vital information. Complementing this is the VisualBackProp algorithm, which identifies the system's focal points within an image, ensuring clear understanding of its attention maps. 

For future mass adoption, there's a need to refine the human-machine interface. Feedback suggests an indicator, like a bar, might be more intuitive to signify when the guardian system takes control.

Air-Guardian heralds a new age of safer skies, offering a reliable safety net for those moments when human attention wavers.

"The Air-Guardian system highlights the synergy between human expertise and machine learning, furthering the objective of using machine learning to augment pilots in challenging scenarios and reduce operational errors," says Daniela Rus, the Andrew (1956) and Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT, director of CSAIL, and senior author on the paper.

"One of the most interesting outcomes of using a visual attention metric in this work is the potential for allowing earlier interventions and greater interpretability by human pilots," says Stephanie Gil, assistant professor of computer science at Harvard University, who was not involved in the work. "This showcases a great example of how AI can be used to work with a human, lowering the barrier for achieving trust by using natural communication mechanisms between the human and the AI system."

This research was partially funded by the U.S. Air Force (USAF) Research Laboratory, the USAF Artificial Intelligence Accelerator, the Boeing Co., and the Office of Naval Research. The findings don't necessarily reflect the views of the U.S. government or the USAF.

de MIT News https://ift.tt/RQFwJXh

“A whole world of potential learners and potential knowledge to gain”

When Aya Khalifa came to MIT from Egypt for her master’s degree in chemical engineering, she adapted well to a new educational system thanks to class 10.MBC (Math Boot Camp for Engineers). This online resource was developed by the MIT Digital Learning Lab (DLL) and the MIT Department of Chemical Engineering for first-year graduate students who might need a refresher on the math skills needed for their core classes. 

“It exposed me to different ways of solving problems,” Khalifa says, adding that the resource was a “huge gain” for her academic progress. She initially took the course during the summer before her program officially started, but she also used Math Boot Camp for Engineers to revisit concepts throughout the semester. 

This online MIT resource is now also available as a massive open online course (MOOC) to any learner in the world. Through serving learners on MIT campus and online, the DLL advances quality digital learning initiatives at the Institute and extends MIT’s teaching and knowledge globally.

Digital learning on campus and beyond 

The DLL, a joint program between MIT Open Learning and MIT’s academic departments, is composed of academic staff and postdocs who collaborate on digital learning innovations. With their combined subject-matter expertise of their respective departments and instructional design innovation, DLL staff promote the latest findings in the learning sciences and educational technologies to develop and update courses.

“The DLL team does so many cool things — creative, hands-on, and informed by the best evidence in teaching and learning,” says Christopher Capozzola, senior associate dean for open learning. “From bringing cutting-edge technologies and ideas into classroom teaching to working alongside faculty as thought partners in developing new courses, programs, and projects, they bring a unique academic and digital agility to every department at MIT.”

Over the last decade, the DLL has grown to encompass all aspects of online learning research and deployment. Faculty work closely with digital learning scientists to incorporate digital technologies into the design and teaching practices of MITx courses and MITx MicroMasters programs for everyone and on-campus courses for MIT students. MITx online courses embody the rigor and quality of MIT’s residential courses, and MicroMasters are credential-bearing programs that can help individuals fast-track and save on costs of their master’s degree. In trying to identify the most effective teaching methods, the Digital Learning Lab ultimately discovers how to better support online learners and MIT students. 

Khalifa says the material she learned in Math Boot Camp for Engineers was relevant to her core graduate courses, including classes 10.50 (Analysis of Transport Phenomena) and 10.65 (Chemical Reactor Engineering). “I didn’t struggle with solving mathematical equations when there was already new content to learn in the course itself,” she says.

This was the outcome the chemical engineering department hoped for when they first approached Joey Gu MS ’16, PhD ’19, lecturer and digital learning scientist in chemical engineering, about improving the first semester experience for graduate students. Gu collaborated with departmental leadership, faculty, and graduate students to develop the first iteration of 10.MBC for summer 2020. Today, Math Boot Camp for Engineers is composed of six self-paced, self-guided, active-learning modules that cover different math topics. Students can take the modules in any order, depending on their needs, as identified by a diagnostic quiz. 

“I liked the option of doing the course in my own time and pace,” says Khalifa, adding that she thought the platform was “very user-friendly.” She found it helpful when concepts were divided into multiple short instructional videos, as opposed to hour-long lectures. 

Immediacy was key to her online learning experience. “I was not waiting for an instructor to give feedback,” Khalifa says. “I solved the problem right then and there, got the answer, and the explanation of the correct answer. I really appreciated that as a student.”

Sharing the latest advances in online learning

As some of the early pioneers and today’s leaders in designing open online courses, digital learning scientists publish research in the fields of learning science and their respective academic areas. They speak at conferences, lead workshops, and share their insights and innovations with MIT faculty and the learning community at large.

This semester, Mary Ellen Wiltrout PhD ’09 and Jessica Sandland ’99, PhD ’04 are serving as general chairs for the 2023 IEEE Learning with MOOCs Conference (LWMOOCs) taking place Oct. 11-13 at MIT. LWMOOCs is an international forum for academic and industry professionals to discuss the latest advances in MOOCs. This year, the conference will focus on blended learning with an emphasis on key topics such as strategies and opportunities for implementing open online courses in today’s world, using these courses to increase educational opportunities for more learners, especially those facing an opportunity gap, and impacting sustainability education. The conference is returning to campus for the first time since its inception in 2014.

“Ten years ago, we were talking about individual, stand-alone online courses, but now there’s more interest in exploring a variety of different educational spaces that are trying to use online modalities to make education more accessible, more affordable,” says Sandland, principal lecturer and digital learning scientist in materials science and engineering.

Participants will also have the opportunity to join workshops on AI in education and inclusive teaching, and learn evidence-based practices from experts developing and managing MOOCs  and other open online courses.

“We want to highlight case studies, research, and frameworks from those creating, running, or studying MOOCs for the community to learn from each other, driving the field to evolve,” says Wiltrout, who is the director of blended and online initiatives, lecturer, and digital learning scientist in biology and has managed over 100 course runs of MOOCs since 2013.

Through participation in LWMOOCs and their own research at the DLL, MIT’s digital learning scientists have been on the forefront of best practices for online teaching and innovations in online learning. “There’s a whole world of potential learners and potential knowledge to gain,” Sandland says. “The more we understand that, the more we can make rich learning experiences for all sorts of different learners.”

de MIT News https://ift.tt/2JtYZcO

Finding solidarity in the teachers’ lounge

In the United States, social institutions from church organizations to sports leagues occupy key roles in shaping political life, with unions perhaps the most familiar player, affecting change in realms from protest movements to elections.   

But while these civil society institutions draw little notice in a democracy, they turn heads in settings where political life is more constrained.     

Elizabeth “Biff” Parker-Magyar, a sixth-year doctoral student in political science at MIT, is investigating this phenomenon.  

“It’s quite puzzling when some organizations manage to form and exert influence in a setting where civil society movements face high barriers to independence,” she says.   

Her dissertation is focused on the small Middle Eastern nation of Jordan. She locates civil society there in an unexpected setting: public-sector work spaces. “Teachers there are highly impactful not only in shaping the contours of education, but in national politics generally, and we don’t really have good explanations for why,” she says.   

Parker-Magyar has immersed herself in the dynamics of Jordanian public-sector workplaces, focusing on the contrasting cases of visibly influential teachers and more isolated public health-care employees. Her research, which paints a fine-grained picture of employee interactions and how these interactions affect political behavior, points to the centrality of social networks within the workplace.  

“I believe my data will help answer some really big questions about both political economy and contentious politics,” she says. “I also hope it will answer some related questions around the impact of political reforms — like how state workers are hired and whether they find their work satisfying — and how decentralization matters for how public sector workers do their jobs.”  

Teachers as activists  

Political science research does not often take up the topics of teachers’ political behavior and the role of state workers in social movements — especially movements that emerge outside of democracies, according to Parker-Magyar. Through detailed field work in Jordan and elsewhere, she hopes to fill that lacuna in the literature. Her research and analysis to date have already borne fruit.   

“We often think of public employees as extremely close to the party in power, because they’re perceived to get their jobs as some sort of quid pro quo” she says. “But we need to distinguish among employees — especially those in the military, for instance — and those who work in more public-facing jobs, like public school teachers and public-sector doctors.” Once employed, these workers occupy influential positions in society, with considerable agency.   

“By creating a social movement, teachers in Jordan have, to a certain extent, been granted a seat at the table,” she says, in ways that shape government policies and actions, if not necessarily electoral politics.   

Before starting her PhD, Parker-Magyar had already spent considerable time in Jordan. She traveled there to study Arabic while at Hamilton College in spring 2011. She pursued a Fulbright in the country right after graduation, serving as a teacher during a momentous time when teachers called a national strike.   

“I had become hyperaware of the social dynamics in the teachers’ lounge, and very engaged with issues in education” she says. The striking teachers wanted some changes in the curriculum and the school calendar. “They also were laying claim to a certain level freedom of association, a right to be represented as a group,” says Parker-Magyar. This was a level of organizing that bore all the attributes of political mobilization.  

Mapping workplace networks  

Parker-Magyar’s dissertation has brought her back to the teachers’ lounges, to test out her idea that social networks in schools, resulting from close ties among colleagues, are a foundational form of political engagement.  

A graduate research fellow with the MIT Governance Lab (GOV/LAB) and MIT Global Diversity Lab, Parker-Magyar deployed such cutting-edge methodologies as a networks elicitation survey to map and measure workplace social networks. She interviewed and surveyed hundreds of teachers and health care employees, in workplace after workplace. She also digitized elections data in Jordan down to the level of local polling places — a real challenge in a setting where data are scarce — to make possible fine-grained analysis of the voting characteristics of communities where public-sector employees have influence.   

Parker-Magyar’s analysis supports her hypothesis that collegial ties help drive teachers’ influence. Teachers who form strong day-to-day bonds also benefit from those bonds when they decide to lobby — sometimes on key issues like curricular reform or wages. Significantly, these connections develop across religious and identity lines. “Having really strong social networks at work and these close relationships with your colleagues can help propel a movement forward, even in really challenging conditions,” she says.  

The same organization is more challenging for doctors and hospital workers. “It is very difficult for health care workers to organize, in part because they are so non-homogeneous and hierarchical,” she says. “Doctors have a harder time coordinating with one another, let alone nurses and X-ray technicians, because their social networks are more fragmented.” While the architecture of these two public sector institutions may look the same, “teachers have major advantages in terms of their relationships with each other that allow them to sustain a social movement.”  

Parker-Magyar believes that autonomous movements that spring up among labor groups must be taken more seriously. She points out that the United States provides enormous amounts of aid to nations like Jordan: “Given the U.S. investment in education and health care practices in Jordan and around the world, we should take seriously the representation of front-line teachers, doctors, and nurses.”    

Public sector employees as political actors  

A New Jersey native, Parker-Magyar grew up in a household steeped in labor politics: Her grandfather was a steelworker and her dad helped develop a history museum related to the state’s labor movement. “I knew I was going to study politics and government from a really young age,” she says, “though the fact that my experiences in the Middle East led me to focus on labor movements has been a bit of a surprise.”   

Parker-Magyar first studied in Jordan in spring 2011, when protest movements across the Arabic-speaking world heralded a period of political change. By the time she had returned to the country for her Fulbright fellowship in 2012, the protest movement in neighboring Syria had developed into a violent civil war, sending hundreds of thousands to Jordan; nearly overnight, one in 10 people in Jordan was a refugee fleeing Syria’s conflict.   

“There was this huge humanitarian need, and I ended up staying an additional year after my Fulbright, working as a journalist in a small newsroom alongside Syrians covering what was happening in their country,” she says. “I gained a deep understanding of the costs of conflict, and it made me very passionate about having a career that attempts to communicate the realities of politics in the region to a broader audience.” 

After several years working on Syria, Parker-Magyar returned to those initial impressions from her days teaching soon after beginning her doctoral work. “I hope my work shows how seriously we should be taking teachers and public-sector workers as political actors,” she says. “These individuals provide important services, and their collective politics also shapes how those services are provided and the broader landscapes of civil society.” 

de MIT News https://ift.tt/azZmcGp

A more effective experimental design for engineering a cell into a new state

A strategy for cellular reprogramming involves using targeted genetic interventions to engineer a cell into a new state. The technique holds great promise in immunotherapy, for instance, where researchers could reprogram a patient’s T-cells so they are more potent cancer killers. Someday, the approach could also help identify life-saving cancer treatments or regenerative therapies that repair disease-ravaged organs.

But the human body has about 20,000 genes, and a genetic perturbation could be on a combination of genes or on any of the over 1,000 transcription factors that regulate the genes. Because the search space is vast and genetic experiments are costly, scientists often struggle to find the ideal perturbation for their particular application.   

Researchers from MIT and Harvard University developed a new, computational approach that can efficiently identify optimal genetic perturbations based on a much smaller number of experiments than traditional methods.

Their algorithmic technique leverages the cause-and-effect relationship between factors in a complex system, such as genome regulation, to prioritize the best intervention in each round of sequential experiments.

The researchers conducted a rigorous theoretical analysis to determine that their technique did, indeed, identify optimal interventions. With that theoretical framework in place, they applied the algorithms to real biological data designed to mimic a cellular reprogramming experiment. Their algorithms were the most efficient and effective.

“Too often, large-scale experiments are designed empirically. A careful causal framework for sequential experimentation may allow identifying optimal interventions with fewer trials, thereby reducing experimental costs,” says co-senior author Caroline Uhler, a professor in the Department of Electrical Engineering and Computer Science (EECS) who is also co-director of the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard, and a researcher at MIT’s Laboratory for Information and Decision Systems (LIDS) and Institute for Data, Systems and Society (IDSS).

Joining Uhler on the paper, which appears today in Nature Machine Intelligence, are lead author Jiaqi Zhang, a graduate student and Eric and Wendy Schmidt Center Fellow; co-senior author Themistoklis P. Sapsis, professor of mechanical and ocean engineering at MIT and a member of IDSS; and others at Harvard and MIT.

Active learning

When scientists try to design an effective intervention for a complex system, like in cellular reprogramming, they often perform experiments sequentially. Such settings are ideally suited for the use of a machine-learning approach called active learning. Data samples are collected and used to learn a model of the system that incorporates the knowledge gathered so far. From this model, an acquisition function is designed — an equation that evaluates all potential interventions and picks the best one to test in the next trial.

This process is repeated until an optimal intervention is identified (or resources to fund subsequent experiments run out).

“While there are several generic acquisition functions to sequentially design experiments, these are not effective for problems of such complexity, leading to very slow convergence,” Sapsis explains.

Acquisition functions typically consider correlation between factors, such as which genes are co-expressed. But focusing only on correlation ignores the regulatory relationships or causal structure of the system. For instance, a genetic intervention can only affect the expression of downstream genes, but a correlation-based approach would not be able to distinguish between genes that are upstream or downstream.

“You can learn some of this causal knowledge from the data and use that to design an intervention more efficiently,” Zhang explains.

The MIT and Harvard researchers leveraged this underlying causal structure for their technique. First, they carefully constructed an algorithm so it can only learn models of the system that account for causal relationships.

Then the researchers designed the acquisition function so it automatically evaluates interventions using information on these causal relationships. They crafted this function so it prioritizes the most informative interventions, meaning those most likely to lead to the optimal intervention in subsequent experiments.

“By considering causal models instead of correlation-based models, we can already rule out certain interventions. Then, whenever you get new data, you can learn a more accurate causal model and thereby further shrink the space of interventions,” Uhler explains.

This smaller search space, coupled with the acquisition function’s special focus on the most informative interventions, is what makes their approach so efficient.

The researchers further improved their acquisition function using a technique known as output weighting, inspired by the study of extreme events in complex systems. This method carefully emphasizes interventions that are likely to be closer to the optimal intervention.

“Essentially, we view an optimal intervention as an ‘extreme event’ within the space of all possible, suboptimal interventions and use some of the ideas we have developed for these problems,” Sapsis says.    

Enhanced efficiency

They tested their algorithms using real biological data in a simulated cellular reprogramming experiment. For this test, they sought a genetic perturbation that would result in a desired shift in average gene expression. Their acquisition functions consistently identified better interventions than baseline methods through every step in the multi-stage experiment.

“If you cut the experiment off at any stage, ours would still be more efficient than the baselines. This means you could run fewer experiments and get the same or better results,” Zhang says.

The researchers are currently working with experimentalists to apply their technique toward cellular reprogramming in the lab.

Their approach could also be applied to problems outside genomics, such as identifying optimal prices for consumer products or enabling optimal feedback control in fluid mechanics applications.

In the future, they plan to enhance their technique for optimizations beyond those that seek to match a desired mean. In addition, their method assumes that scientists already understand the causal relationships in their system, but future work could explore how to use AI to learn that information, as well.

This work was funded, in part, by the Office of Naval Research, the MIT-IBM Watson AI Lab, the MIT J-Clinic for Machine Learning and Health, the Eric and Wendy Schmidt Center at the Broad Institute, a Simons Investigator Award, the Air Force Office of Scientific Research, and a National Science Foundation Graduate Fellowship.

de MIT News https://ift.tt/JGOhZFK

Is AI in the eye of the beholder?

Someone’s prior beliefs about an artificial intelligence agent, like a chatbot, have a significant effect on their interactions with that agent and their perception of its trustworthiness, empathy, and effectiveness, according to a new study.

Researchers from MIT and Arizona State University found that priming users — by telling them that a conversational AI agent for mental health support was either empathetic, neutral, or manipulative — influenced their perception of the chatbot and shaped how they communicated with it, even though they were speaking to the exact same chatbot.

Most users who were told the AI agent was caring believed that it was, and they also gave it higher performance ratings than those who believed it was manipulative. At the same time, less than half of the users who were told the agent had manipulative motives thought the chatbot was actually malicious, indicating that people may try to “see the good” in AI the same way they do in their fellow humans.

The study revealed a feedback loop between users’ mental models, or their perception of an AI agent, and that agent’s responses. The sentiment of user-AI conversations became more positive over time if the user believed the AI was empathetic, while the opposite was true for users who thought it was nefarious.

“From this study, we see that to some extent, the AI is the AI of the beholder,” says Pat Pataranutaporn, a graduate student in the Fluid Interfaces group of the MIT Media Lab and co-lead author of a paper describing this study. “When we describe to users what an AI agent is, it does not just change their mental model, it also changes their behavior. And since the AI responds to the user, when the person changes their behavior, that changes the AI, as well.”

Pataranutaporn is joined by co-lead author and fellow MIT graduate student Ruby Liu; Ed Finn, associate professor in the Center for Science and Imagination at Arizona State University; and senior author Pattie Maes, professor of media technology and head of the Fluid Interfaces group at MIT.

The study, published today in Nature Machine Intelligence, highlights the importance of studying how AI is presented to society, since the media and popular culture strongly influence our mental models. The authors also raise a cautionary flag, since the same types of priming statements in this study could be used to deceive people about an AI’s motives or capabilities.

“A lot of people think of AI as only an engineering problem, but the success of AI is also a human factors problem. The way we talk about AI, even the name that we give it in the first place, can have an enormous impact on the effectiveness of these systems when you put them in front of people. We have to think more about these issues,” Maes says.

AI friend or foe?

In this study, the researchers sought to determine how much of the empathy and effectiveness people see in AI is based on their subjective perception and how much is based on the technology itself. They also wanted to explore whether one could manipulate someone’s subjective perception with priming.

“The AI is a black box, so we tend to associate it with something else that we can understand. We make analogies and metaphors. But what is the right metaphor we can use to think about AI? The answer is not straightforward,” Pataranutaporn says.

They designed a study in which humans interacted with a conversational AI mental health companion for about 30 minutes to determine whether they would recommend it to a friend, and then rated the agent and their experiences. The researchers recruited 310 participants and randomly split them into three groups, which were each given a priming statement about the AI.

One group was told the agent had no motives, the second group was told the AI had benevolent intentions and cared about the user’s well-being, and the third group was told the agent had malicious intentions and would try to deceive users. While it was challenging to settle on only three primers, the researchers chose statements they thought fit the most common perceptions about AI, Liu says.

Half the participants in each group interacted with an AI agent based on the generative language model GPT-3, a powerful deep-learning model that can generate human-like text. The other half interacted with an implementation of the chatbot ELIZA, a less sophisticated rule-based natural language processing program developed at MIT in the 1960s.

Molding mental models

Post-survey results revealed that simple priming statements can strongly influence a user’s mental model of an AI agent, and that the positive primers had a greater effect. Only 44 percent of those given negative primers believed them, while 88 percent of those in the positive group and 79 percent of those in the neutral group believed the AI was empathetic or neutral, respectively.

“With the negative priming statements, rather than priming them to believe something, we were priming them to form their own opinion. If you tell someone to be suspicious of something, then they might just be more suspicious in general,” Liu says.

But the capabilities of the technology do play a role, since the effects were more significant for the more sophisticated GPT-3 based conversational chatbot.

The researchers were surprised to see that users rated the effectiveness of the chatbots differently based on the priming statements. Users in the positive group awarded their chatbots higher marks for giving mental health advice, despite the fact that all agents were identical.

Interestingly, they also saw that the sentiment of conversations changed based on how users were primed. People who believed the AI was caring tended to interact with it in a more positive way, making the agent’s responses more positive. The negative priming statements had the opposite effect. This impact on sentiment was amplified as the conversation progressed, Maes adds.

The results of the study suggest that because priming statements can have such a strong impact on a user’s mental model, one could use them to make an AI agent seem more capable than it is — which might lead users to place too much trust in an agent and follow incorrect advice.

“Maybe we should prime people more to be careful and to understand that AI agents can hallucinate and are biased. How we talk about AI systems will ultimately have a big effect on how people respond to them,” Maes says.

In the future, the researchers want to see how AI-user interactions would be affected if the agents were designed to counteract some user bias. For instance, perhaps someone with a highly positive perception of AI is given a chatbot that responds in a neutral or even a slightly negative way so the conversation stays more balanced.

They also want to use what they’ve learned to enhance certain AI applications, like mental health treatments, where it could be beneficial for the user to believe an AI is empathetic. In addition, they want to conduct a longer-term study to see how a user’s mental model of an AI agent changes over time.

This research was funded, in part, by the Media Lab, the Harvard-MIT Program in Health Sciences and Technology, Accenture, and KBTG. 

de MIT News https://ift.tt/KxAaePI