Posted on

New neural-network algorithm learns directly from human instructions instead of examples

Conventional neural-network image-recognition algorithm trained to recognize human hair (left), compared to the more precise heuristically trained algorithm (right) (credit: Wenzhangzhi Guo and Parham Aarabi/IEEE Trans NN & LS)

A new machine learning algorithm designed by University of Toronto researchers Parham Aarabi and Wenzhi Guo learns directly from human instructions, rather than an existing set of examples, as in traditional neural networks. In tests, it outperformed existing neural networks by 160 per cent.

Their “heuristically trained neural networks” (HNN) algorithm also outperformed its own training by nine per cent — it learned to recognize hair in pictures with greater reliability than that enabled by the training.

Aarabi and Guo trained their HNN algorithm to identify people’s hair in photographs, a challenging task for computers. “Our algorithm learned to correctly classify difficult, borderline cases — distinguishing the texture of hair versus the texture of the background,” says Aarabi. “What we saw was like a teacher instructing a child, and the child learning beyond what the teacher taught her initially.”

Heuristic training

Humans conventionally “teach” neural networks by providing a set of labeled data and asking the neural network to make decisions based on the samples it’s seen. For example, you could train a neural network to identify sky in a photograph by showing it hundreds of pictures with the sky labeled.

With HNN, humans provide direct instructions that are used to pre-classify training samples rather than a set of fixed examples. Trainers program the algorithm with guidelines such as “Sky is likely to be varying shades of blue,” and “Pixels near the top of the image are more likely to be sky than pixels at the bottom.”

Their work is published in the journal IEEE Transactions on Neural Networks and Learning Systems.

This heuristic-training approach addresses one of the biggest challenges for neural networks: making correct classifications of previously unknown or unlabeled data, the researchers say. This is crucial for applying machine learning to new situations, such as correctly identifying cancerous tissues for medical diagnostics, or classifying all the objects surrounding and approaching a self-driving car.

“Applying heuristic training to hair segmentation is just a start,” says Guo. “We’re keen to apply our method to other fields and a range of applications, from medicine to transportation.”

Abstract of Hair Segmentation Using Heuristically-Trained Neural Networks

We present a method for binary classification using neural networks (NNs) that performs training and classification on the same data using the help of a pretraining heuristic classifier. The heuristic classifier is initially used to segment data into three clusters of high-confidence positives, high-confidence negatives, and low-confidence sets. The high-confidence sets are used to train an NN, which is then used to classify the low-confidence set. Applying this method to the binary classification of hair versus nonhair patches, we obtain a 2.2% performance increase using the heuristically trained NN over the current state-of-the-art hair segmentation method.

Posted on

Google’s new multilingual Neural Machine Translation System can translate between language pairs even though it has never been taught to do so

Google Neural Machine Translation (credit: Google)

Google researchers have announced they have implemented a neural machine translation system in Google Translate that improves translation quality and enables “Zero-Shot Translation” — translation between language pairs never seen explicitly by the system.

For example, in the animation above, the system was trained to translate bidirectionally between English and Japanese and between English and Korean. But the new system can also translate between Japanese and Korean — even though it has never been taught to do so.

Google calls this “zero-shot” translation — shown by the yellow dotted lines in the animation. “To the best of our knowledge, this is the first time this type of transfer learning has worked in machine translation,” the researchers say.

Neural networks for machine translation

What drove this development was the fact that in the last ten years, Google Translate has grown from supporting just a few languages to 103, translating more than 140 billion words every day. That required building and maintaining many different systems to translate between any two languages — incurring significant computational cost.

That meant the researchers needed to rethink the technology behind Google Translate.

As a result, the Google Brain Team and Google Translate team announced the first step in September: a new system called Google Neural Machine Translation (GNMT), which learns from millions of examples.

“GNMT reduces translation errors by more than 55%–85% on several major language pairs measured on sampled sentences from Wikipedia and news websites with the help of bilingual human raters,” the researchers note. “However, while switching to GNMT improved the quality for the languages we tested it on, scaling up to all the 103 supported languages presented a significant challenge.”

So as explained in a paper, “Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation” (open access), the researchers extended the GNMT system, allowing for a single system to translate between multiple languages. “Our proposed architecture requires no change in the base GNMT system,” the authors explain, “instead it uses an additional ‘token’ at the beginning of the input sentence to specify the required target language to translate to.”

An interlingua

But the success of the zero-shot translation raised another important question: Is the system learning a common representation in which sentences with the same meaning are represented in similar ways regardless of language — i.e. an “interlingua”?

To find out, the researchers used a 3-dimensional representation of internal network data, allowing them to take a peek into the system as it translated a set of sentences between all possible pairs of the Japanese, Korean, and English languages, for example.

Developing an “interlinqua” (credit: Google)

Part (a) from the figure above shows an overall geometry of these translations. The points in this view are colored by the meaning; a sentence translated from English to Korean with the same meaning as a sentence translated from Japanese to English share the same color. From this view we can see distinct groupings of points, each with their own color. Part (b) zooms in to one of the groups, and the colors in part (c) show the different source languages.

Within a single group, we see a sentence with the same meaning (“The stratosphere extends from about 10km to about 50km in altitude”), but from three different languages, the researchers explain. “This means the network must be encoding something about the semantics [meaning] of the sentence, rather than simply memorizing phrase-to-phrase translations. We interpret this as a sign of existence of an interlingua in the network.”

The new Multilingual Google Neural Machine Translation system is running in production today for all Google Translate users, according to the researchers.

Abstract of Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation

We propose a simple, elegant solution to use a single Neural Machine Translation (NMT) model to translate between multiple languages. Our solution requires no change in the model architecture from our base system but instead introduces an artificial token at the beginning of the input sentence to specify the required target language. The rest of the model, which includes encoder, decoder and attention, remains unchanged and is shared across all languages. Using a shared wordpiece vocabulary, our approach enables Multilingual NMT using a single model without any increase in parameters, which is significantly simpler than previous proposals for Multilingual NMT. Our method often improves the translation quality of all involved language pairs, even while keeping the total number of model parameters constant. On the WMT’14 benchmarks, a single multilingual model achieves comparable performance for English→French and surpasses state-of-the-art results for English→German. Similarly, a single multilingual model surpasses state-of-the-art results for French→English and German→English on WMT’14 and WMT’15 benchmarks respectively. On production corpora, multilingual models of up to twelve language pairs allow for better translation of many individual pairs. In addition to improving the translation quality of language pairs that the model was trained with, our models can also learn to perform implicit bridging between language pairs never seen explicitly during training, showing that transfer learning and zero-shot translation is possible for neural translation. Finally, we show analyses that hints at a universal interlingua representation in our models and show some interesting examples when mixing languages.

Posted on

New MIT technique reveals the basis for machine-learning systems’ hidden decisions

A Stanford School of Medicine machine-learning method for automatically analyzing images of cancerous tissues and predicting patient survival was found more accurate than doctors in breast-cancer diagnosis, but doctors still don’t trust this method, say MIT researchers (credit: Science/AAAS)

MIT researchers have developed a method to determine the rationale for predictions by neural networks, which loosely mimic the human brain. Neural networks, such as Google’s Alpha Go program, use a process known as “deep learning” to look for patterns in training data.

An ongoing problem with neural networks is that they are “black boxes.” After training, a network may be very good at classifying data, but even its creators will have no idea why. With visual data, it’s sometimes possible to automate experiments that determine which visual features a neural net is responding to, but text-processing systems tend to be more opaque.

In the deep learning process, training data is fed to a network’s input nodes, which modify it and feed it to other nodes, which modify it and feed it to still other nodes, and so on. The values stored in the network’s output nodes are then correlated with the classification category that the network is trying to learn — such as the objects in an image, or the topic of an essay.

“In real-world applications, sometimes people really want to know why the model makes the predictions it does,” says Tao Lei, an MIT graduate student in electrical engineering and computer science and first author on the new paper. “One major reason that doctors don’t trust machine-learning methods is that there’s no evidence.” Another critical example is self-driving cars.

At the Association for Computational Linguistics’ Conference on Empirical Methods in Natural Language Processing, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) will present a new way to train neural networks so that they provide not only predictions and classifications but rationales for their decisions.

“There’s a broader aspect to this work, as well,” says Tommi Jaakkola, an MIT professor of electrical engineering and computer science and a coauthor on the open-access paper. “You may not want to just verify that the model is making the prediction in the right way; you might also want to exert some influence in terms of the types of predictions that it should make. How does a layperson communicate with a complex model that’s trained with algorithms that they know nothing about? They might be able to tell you about the rationale for a particular prediction. In that sense it opens up a different way of communicating with the model.”

Interpreting a neural net’s decisions

In the new paper, the CSAIL researchers specifically address neural nets trained on textual data. To enable interpretation of a neural net’s decisions, the researchers divide the net into two modules. The first module extracts segments of text from the training data, and the segments are scored according to their length and their coherence: The shorter the segment, and the more of it that is drawn from strings of consecutive words, the higher its score.

The segments selected by the first module are then passed to the second module, which performs the prediction or classification task. The modules are trained together, and the goal of training is to maximize both the score of the extracted segments and the accuracy of prediction or classification.

An example of a beer review with ranking in two categories. The rationale for Look prediction is shown in bold. (credit: MIT CSAIL)

One of the data sets on which the researchers tested their system is a group of reviews from a website where users evaluate different beers. The data set includes the raw text of the reviews and the corresponding ratings, using a five-star system, on each of three attributes: aroma, palate, and appearance.

What makes the data attractive to natural-language-processing researchers is that it’s also been annotated by hand, to indicate which sentences in the reviews correspond to which scores. For example, a review might consist of eight or nine sentences, and the annotator might have highlighted those that refer to the beer’s “tan-colored head about half an inch thick,” “signature Guinness smells,” and “lack of carbonation.” Each sentence is correlated with a different attribute rating.

As such, the data set provides an excellent test of the CSAIL researchers’ system. If the first module has extracted those three phrases, and the second module has correlated them with the correct ratings, then the system has identified the same basis for judgment that the human annotator did.

In experiments, the system’s agreement with the human annotations was 96 percent and 95 percent, respectively, for ratings of appearance and aroma, and 80 percent for the more nebulous concept of palate.

In the paper, the researchers also report testing their system on a database of free-form technical questions and answers, where the task is to determine whether a given question has been answered previously.

In unpublished work, they’ve applied it to thousands of pathology reports on breast biopsies, where it has learned to extract text explaining the bases for the pathologists’ diagnoses. They’re even using it to analyze mammograms, where the first module extracts sections of images rather than segments of text.

Abstract of Rationalizing Neural Predictions

Prediction without justification has limited applicability. As a remedy, we learn to extract pieces of input text as justifications – rationales – that are tailored to be short and coherent, yet sufficient for making the same prediction. Our approach combines two modular components, generator and encoder, which are trained to operate well together. The generator specifies a distribution over text fragments as candidate rationales and these are passed through the encoder for prediction. Rationales are never given during training. Instead, the model is regularized by desiderata for rationales. We evaluate the approach on multi-aspect sentiment analysis against manually annotated test cases. Our approach outperforms attention-based baseline by a significant margin. We also successfully illustrate the method on the question retrieval task.

Posted on

New MIT technique reveals the basis for machine-learning systems’ hidden decisions

A Stanford School of Medicine machine-learning method for automatically analyzing images of cancerous tissues and predicting patient survival was found more accurate than doctors in breast-cancer diagnosis, but doctors still don’t trust this method, say MIT researchers (credit: Science/AAAS)

MIT researchers have developed a method to determine the rationale for predictions by neural networks, which loosely mimic the human brain. Neural networks, such as Google’s Alpha Go program, use a process known as “deep learning” to look for patterns in training data.

An ongoing problem with neural networks is that they are “black boxes.” After training, a network may be very good at classifying data, but even its creators will have no idea why. With visual data, it’s sometimes possible to automate experiments that determine which visual features a neural net is responding to, but text-processing systems tend to be more opaque.

In the deep learning process, training data is fed to a network’s input nodes, which modify it and feed it to other nodes, which modify it and feed it to still other nodes, and so on. The values stored in the network’s output nodes are then correlated with the classification category that the network is trying to learn — such as the objects in an image, or the topic of an essay.

“In real-world applications, sometimes people really want to know why the model makes the predictions it does,” says Tao Lei, an MIT graduate student in electrical engineering and computer science and first author on the new paper. “One major reason that doctors don’t trust machine-learning methods is that there’s no evidence.” Another critical example is self-driving cars.

At the Association for Computational Linguistics’ Conference on Empirical Methods in Natural Language Processing, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) will present a new way to train neural networks so that they provide not only predictions and classifications but rationales for their decisions.

“There’s a broader aspect to this work, as well,” says Tommi Jaakkola, an MIT professor of electrical engineering and computer science and a coauthor on the open-access paper. “You may not want to just verify that the model is making the prediction in the right way; you might also want to exert some influence in terms of the types of predictions that it should make. How does a layperson communicate with a complex model that’s trained with algorithms that they know nothing about? They might be able to tell you about the rationale for a particular prediction. In that sense it opens up a different way of communicating with the model.”

Interpreting a neural net’s decisions

In the new paper, the CSAIL researchers specifically address neural nets trained on textual data. To enable interpretation of a neural net’s decisions, the researchers divide the net into two modules. The first module extracts segments of text from the training data, and the segments are scored according to their length and their coherence: The shorter the segment, and the more of it that is drawn from strings of consecutive words, the higher its score.

The segments selected by the first module are then passed to the second module, which performs the prediction or classification task. The modules are trained together, and the goal of training is to maximize both the score of the extracted segments and the accuracy of prediction or classification.

An example of a beer review with ranking in two categories. The rationale for Look prediction is shown in bold. (credit: MIT CSAIL)

One of the data sets on which the researchers tested their system is a group of reviews from a website where users evaluate different beers. The data set includes the raw text of the reviews and the corresponding ratings, using a five-star system, on each of three attributes: aroma, palate, and appearance.

What makes the data attractive to natural-language-processing researchers is that it’s also been annotated by hand, to indicate which sentences in the reviews correspond to which scores. For example, a review might consist of eight or nine sentences, and the annotator might have highlighted those that refer to the beer’s “tan-colored head about half an inch thick,” “signature Guinness smells,” and “lack of carbonation.” Each sentence is correlated with a different attribute rating.

As such, the data set provides an excellent test of the CSAIL researchers’ system. If the first module has extracted those three phrases, and the second module has correlated them with the correct ratings, then the system has identified the same basis for judgment that the human annotator did.

In experiments, the system’s agreement with the human annotations was 96 percent and 95 percent, respectively, for ratings of appearance and aroma, and 80 percent for the more nebulous concept of palate.

In the paper, the researchers also report testing their system on a database of free-form technical questions and answers, where the task is to determine whether a given question has been answered previously.

In unpublished work, they’ve applied it to thousands of pathology reports on breast biopsies, where it has learned to extract text explaining the bases for the pathologists’ diagnoses. They’re even using it to analyze mammograms, where the first module extracts sections of images rather than segments of text.

Abstract of Rationalizing Neural Predictions

Prediction without justification has limited applicability. As a remedy, we learn to extract pieces of input text as justifications – rationales – that are tailored to be short and coherent, yet sufficient for making the same prediction. Our approach combines two modular components, generator and encoder, which are trained to operate well together. The generator specifies a distribution over text fragments as candidate rationales and these are passed through the encoder for prediction. Rationales are never given during training. Instead, the model is regularized by desiderata for rationales. We evaluate the approach on multi-aspect sentiment analysis against manually annotated test cases. Our approach outperforms attention-based baseline by a significant margin. We also successfully illustrate the method on the question retrieval task.

Posted on

Will AI replace judges and lawyers?

(credit: iStock)

An artificial intelligence method developed by University College London computer scientists and associates has predicted the judicial decisions of the European Court of Human Rights (ECtHR) with 79% accuracy, according to a paper published Monday, Oct. 24 in PeerJ Computer Science.

The method is the first to predict the outcomes of a major international court by automatically analyzing case text using a machine-learning algorithm.*

“We don’t see AI replacing judges or lawyers,” said Nikolaos Aletras, who led the study at UCL Computer Science, “but we think they’d find it useful for rapidly identifying patterns in cases that lead to certain outcomes. It could also be a valuable tool for highlighting which cases are most likely to be violations of the European Convention on Human Rights.”

Judgments correlated with facts rather than legal arguments

(credit: European Court of Human Rights)

In developing the method, the team found that judgments by the ECtHR are highly correlated to non-legal (real-world) facts, rather than direct legal arguments, suggesting that judges of the Court are, in the jargon of legal theory, “realists” rather than “formalists.”

This supports findings from previous studies of the decision-making processes of other high level courts, including the U.S. Supreme Court.

The team of computer and legal scientists extracted case information published by the ECtHR in its publically accessible database (applications made to the court were not available), explained UCL co-author Vasileios Lampos, PhD.

They identified English-language data sets for 584 cases relating to Articles 3, 6 and 8** of the Convention and applied an AI algorithm to find patterns in the text. To prevent bias and mislearning, they selected an equal number of violation and non-violation cases.

Predictions based of analysis of text

The most reliable factors for predicting the court’s decision were found to be the language used as well as the topics and circumstances mentioned in the case text (the “circumstances” section of the text includes information about the case factual background). By combining the information extracted from the abstract “topics” that the cases cover and “circumstances” across data for all three articles, an accuracy of 79% was achieved.

“Previous studies have predicted outcomes based on the nature of the crime, or the policy position of each judge, so this is the first time judgments have been predicted using analysis of text prepared by the court. We expect this sort of tool would improve efficiencies of high level, in demand courts, but to become a reality, we need to test it against more articles and the case data submitted to the court,” added Lampos.

Researchers at the University of Sheffield and the University of Pennsylvania where also involved in the study.

* “We define the problem of the ECtHR case prediction as a binary classification task. We utilise textual features, i.e., N-grams and topics, to train Support Vector Machine (SVM) classifiers. We apply a linear kernel function that facilitates the interpretation of models in a straightforward manner.” — Authors of PeerJ Computer Science paper.

** Article 3 prohibits torture and inhuman and degrading treatment (250 cases); Article 6 protects the right to a fair trial (80 cases); and Article 8 provides a right to respect for one’s “private and family life, his home and his correspondence” (254 cases).

Abstract of Predicting Judicial Decisions of the European Court of Human Rights: A Natural Language Processing Perspective

Recent advances in Natural Language Processing and Machine Learning provide us with the tools to build predictive models that can be used to unveil patterns driving judicial decisions. This can be useful, for both lawyers and judges, as an assisting tool to rapidly identify cases and extract patterns which lead to certain decisions. This paper presents the first systematic study on predicting the outcome of cases tried by the European Court of Human Rights based solely on textual content. We formulate a binary classification task where the input of our classifiers is the textual content extracted from a case and the target output is the actual judgment as to whether there has been a violation of an article of the convention of human rights. Textual information is represented using contiguous word sequences, i.e. N-grams, and topics. Our models can predict the court’s decisions with a strong accuracy (79% on average). Our empirical analysis indicates that the formal facts of a case are the most important predictive factor. This is consistent with the theory of legal realism suggesting that judicial decision-making is significantly affected by the stimulus of the facts. We also observe that the topical content of a case is another important feature in this classification task and explore this relationship further by conducting a qualitative analysis.

Posted on

Will AI replace judges and lawyers?

(credit: iStock)

An artificial intelligence method developed by University College London computer scientists and associates has predicted the judicial decisions of the European Court of Human Rights (ECtHR) with 79% accuracy, according to a paper published Monday, Oct. 24 in PeerJ Computer Science.

The method is the first to predict the outcomes of a major international court by automatically analyzing case text using a machine-learning algorithm.*

“We don’t see AI replacing judges or lawyers,” said Nikolaos Aletras, who led the study at UCL Computer Science, “but we think they’d find it useful for rapidly identifying patterns in cases that lead to certain outcomes. It could also be a valuable tool for highlighting which cases are most likely to be violations of the European Convention on Human Rights.”

Judgments correlated with facts rather than legal arguments

(credit: European Court of Human Rights)

In developing the method, the team found that judgments by the ECtHR are highly correlated to non-legal (real-world) facts, rather than direct legal arguments, suggesting that judges of the Court are, in the jargon of legal theory, “realists” rather than “formalists.”

This supports findings from previous studies of the decision-making processes of other high level courts, including the U.S. Supreme Court.

The team of computer and legal scientists extracted case information published by the ECtHR in its publically accessible database (applications made to the court were not available), explained UCL co-author Vasileios Lampos, PhD.

They identified English-language data sets for 584 cases relating to Articles 3, 6 and 8** of the Convention and applied an AI algorithm to find patterns in the text. To prevent bias and mislearning, they selected an equal number of violation and non-violation cases.

Predictions based of analysis of text

The most reliable factors for predicting the court’s decision were found to be the language used as well as the topics and circumstances mentioned in the case text (the “circumstances” section of the text includes information about the case factual background). By combining the information extracted from the abstract “topics” that the cases cover and “circumstances” across data for all three articles, an accuracy of 79% was achieved.

“Previous studies have predicted outcomes based on the nature of the crime, or the policy position of each judge, so this is the first time judgments have been predicted using analysis of text prepared by the court. We expect this sort of tool would improve efficiencies of high level, in demand courts, but to become a reality, we need to test it against more articles and the case data submitted to the court,” added Lampos.

Researchers at the University of Sheffield and the University of Pennsylvania where also involved in the study.

* “We define the problem of the ECtHR case prediction as a binary classification task. We utilise textual features, i.e., N-grams and topics, to train Support Vector Machine (SVM) classifiers. We apply a linear kernel function that facilitates the interpretation of models in a straightforward manner.” — Authors of PeerJ Computer Science paper.

** Article 3 prohibits torture and inhuman and degrading treatment (250 cases); Article 6 protects the right to a fair trial (80 cases); and Article 8 provides a right to respect for one’s “private and family life, his home and his correspondence” (254 cases).

Abstract of Predicting Judicial Decisions of the European Court of Human Rights: A Natural Language Processing Perspective

Recent advances in Natural Language Processing and Machine Learning provide us with the tools to build predictive models that can be used to unveil patterns driving judicial decisions. This can be useful, for both lawyers and judges, as an assisting tool to rapidly identify cases and extract patterns which lead to certain decisions. This paper presents the first systematic study on predicting the outcome of cases tried by the European Court of Human Rights based solely on textual content. We formulate a binary classification task where the input of our classifiers is the textual content extracted from a case and the target output is the actual judgment as to whether there has been a violation of an article of the convention of human rights. Textual information is represented using contiguous word sequences, i.e. N-grams, and topics. Our models can predict the court’s decisions with a strong accuracy (79% on average). Our empirical analysis indicates that the formal facts of a case are the most important predictive factor. This is consistent with the theory of legal realism suggesting that judicial decision-making is significantly affected by the stimulus of the facts. We also observe that the topical content of a case is another important feature in this classification task and explore this relationship further by conducting a qualitative analysis.

Posted on

IBM announces AI-powered decision-making

Project DataWorks predictive model (credit: IBM)

IBM today announced today Watson-based “Project DataWorks,” the first cloud-based data and analytics platform to integrate all types of data and enable AI-powered decision-making.

Project DataWorks is designed to make it simple for business leaders and data professionals to collect, organize, govern, and secure data, and become a “cognitive business.”

Achieving data insights is increasingly complex, and most of this work is done by highly skilled data professionals who work in silos with disconnected tools and data services that may be difficult to manage, integrate, and govern, says IBM. Businesses must also continually iterate their data models and products — often manually — to benefit from the most relevant, up-to-date insights.

IBM says Project DataWorks can help businesses break down these barriers by connecting all data and insights for their users into an integrated, self-service platform.

Available on Bluemix, IBM’s Cloud platform, Project DataWorks is designed to help organizations:

  • Automate the deployment of data assets and products using cognitive-based machine learning and Apache Spark;
  • Ingest data faster than any other data platform, from 50 to hundreds of Gbps, and all endpoints: enterprise databases, Internet of Things, weather, and social media;
  • Leverage an open ecosystem of more than 20 partners and technologies, such as Confluent, Continuum Analytics, Galvanize, Alation, NumFOCUS, RStudio, Skymind, and more.

 

Posted on

A thought-controlled robotic exoskeleton for the hand

A robotic hand exoskeleton helps stroke patients integrate rehabilitation exercises into their everyday lives. (credit: Gerber Loesch Photography)

Scientists at ETH Rehabilitation Engineering Laboratory in Switzerland have invented a robotic system that they say could fundamentally change the daily lives of stroke patients.

According to the ETH scientists, one in six people will suffer a stroke in their lifetime; two thirds of those affected suffer from paralysis of the arm. Intensive clinical training, including robot-assisted therapy, can help patients regain a degree of limited control over their arms and hands.

But now Roger Gassert, Professor of Rehabilitation Engineering at ETH Zurich, has a better idea. “My vision is that instead of performing exercises in an abstract situation at the clinic, patients will be able to integrate them into their daily life at home, supported in some cases by a robot” — using an exoskeleton mounted on the hand.

A lightweight exoskeleton that extends the patient’s hand

The problem: existing exoskeletons are heavy, so patients can’t lift their hands, Gassert says, and patients have difficulty feeling objects and exerting the right amount of force. “That’s why we wanted to develop a model that leaves the palm of the hand more or less free, allowing patients to perform daily activities that support not only motor (movement) functions but somatosensory functions as well.”

The initial solution, developed with Professor Jumpei Arata from Kyushu University (Japan), was a mechanism for the finger featuring three overlapping leaf springs. A motor moves the middle spring, which transmits the force to the different segments of the finger through the other two springs. The fingers thus automatically adapt to the shape of the object the patient wants to grasp. But the motors brought the weight of the exoskeleton to 250 grams, which in clinical tests proved too heavy for patients.

The new solution: remove the motors from the hand and fix them to the patient’s back. The force is transmitted to the exoskeleton using a bicycle brake cable. The hand module now weighs slightly less than 120 grams and is strong enough to lift a liter bottle of mineral water.

A hand exoskeleton with motors that can be fixed to the patient’s back: A bicycle brake cable transmits enough force to lift a liter bottle of mineral water. (credit: Stefan Schneller)

Strengthening existing neural connections between brain and hand

Another problem was making sure commands from the brain can reach the extremities after a stroke. “Especially with seriously affected patients, the connection between the brain and the hand is often severely or completely disrupted,” Gassert explains.

The idea is to enable the brain to detect a patient’s intention to move his or her hand and directly pass this information on to the exoskeleton.

Gassert says a number of studies show that it is possible to strengthen existing neural connections between the brain and the hand with regular exercise if the brain can receive somatosensory feedback from the hand when it produces a command to move.

Gassert is using electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) to study this. An interaction between the brain and the exoskeleton could lead to a device that is ideally suited for therapy — without requiring brain implants.

Even if the deficits are permanent, a robotic device could still offer vital long-term support.

Posted on

How AI may affect urban life in 2030

(credit: AI100)

Specialized robots that clean and provide security, robot-assisted surgery, natural language processing-augmented instruction, and helping people adapt as old jobs are lost and new ones are created: these are some of the profound challenges explored by a panel of academic and industrial thinkers that has looked ahead to 2030 to forecast how advances in artificial intelligence (AI) might affect life in a typical North American city.

Titled “Artificial Intelligence and Life in 2030,” this open-access year-long investigation is the first product of the One Hundred Year Study on Artificial Intelligence (AI100), an ongoing project hosted by Stanford University to inform society and provide guidance on the ethical development of smart software, sensors and machines.

The new report traces its roots to a 2009 study by AI scientists in 2014, when Eric and Mary Horvitz created the AI100 endowment through Stanford.

The 28,000-word report includes a glossary to help nontechnical readers understand how AI applications such as computer vision might help screen tissue samples for cancers or how natural language processing will allow computerized systems to grasp not simply the literal definitions, but the connotations and intent, behind words.

“Currently in the United States, at least sixteen separate agencies govern sectors of the economy related to AI technologies,” the researchers write. “Who is responsible when a self-driven car crashes or an intelligent medical device fails? How can AI applications be prevented from [being used for] racial discrimination or financial cheating?”

Beyond science fiction books and movies

“Until now, most of what [the public knows] about AI comes from science fiction books and movies,” Stone said. “This study provides a realistic foundation to discuss how AI technologies are likely to affect society.”

The study raises questions such as: How can AI applications be prevented from unlawful discrimination? Who should reap the gains of efficiencies enabled by AI technologies and what protections should be afforded to people whose skills are rendered obsolete?

To help address concerns about the individual and societal implications of rapidly evolving AI technologies, the Study Panel offers three general policy recommendations:

  • Define a path toward accruing technical expertise in AI at all levels of government.
  • Remove impediments to research on the fairness, security, privacy, and social impacts of AI systems.
  • Increase public and private funding for interdisciplinary studies of the societal impacts of AI.