These are the latest articles and videos I found most interesting.

1. Artificial General Intelligence: Why Aren’t We There Yet?
2. Machine Learning and AI for the Sciences – Towards Understanding
3. Deep Learning in Speech Recognition
4. Face Value: The Irresistible Influence of First Impressions
5. Brain waves reflect different types of learning
6. Brain scans reveal why rewards and punishments don’t seem to work on teenagers

You cannot swim for new horizons until you have courage to lose sight of the shore.

‒ William Faulkner

---

Artificial General Intelligence: Why Aren’t We There Yet?

Talk by Gary Marcus at Allen Institute for Artificial Intelligence (AI2)

All purpose, all-powerful AI systems, capable of catering to our every intellectual need, have been promised for six decades, but thus far still not arrived. What will it take to bring AI to something like human-level intelligence? And why haven’t we gotten there already? Scientist, author, and entrepreneur Gary Marcus (Founder and CEO of Geometric Intelligence, recently acquired by Uber) explains why deep learning is overrated, and what we need to do next to achieve genuine artificial intelligence.

Marcus and LeCun in Complete Agreement on Seven Points (October 5, 2107):

• AI is still in infancy
• Machine learning is fundamentally necessary for reaching strong AI
• Deep learning is a powerful technique for machine learning
• Deep learning is not sufficient on its own for cognition
• [mode-free] Reinforcement learning is not the answer, either
• AI systems still need better internal forward models
• Commonsense reasoning remains fundamentally unsolved

Utter optimist Andrew Ng versus the realists:

If a typical person can do a mental task with less than one second of thought, we can probably automate it using AI either now or in the near future.

‒ Andrew Ng

If a typical person can do a mental task with less than one second of thought, and we can gather an enormous amount of directly relevant data, we have a fighting chance – so long as the test data aren’t too terribly different from the training data, and the domain doesn’t change too much over time

‒ Garry Marcus

---

Machine Learning and AI for the Sciences – Towards Understanding

Talk by Klaus-Robert Müller, Technische Universität Berlin

In recent years, machine learning (ML) and artificial intelligence (AI) methods have begun to play a more and more enabling role in the sciences and in industry. In particular, the advent of large and/or complex data corpora has given rise to new technological challenges and possibilities. In his talk, Müller will touch upon the topic of ML applications in the sciences, in particular in neuroscience, medicine and physics. He will also discuss possibilities for extracting information from machine learning models to further our understanding by explaining nonlinear ML models. E.g. Machine Learning Models for Quantum Chemistry can, by applying interpretable ML, contribute to furthering chemical understanding. Finally, Müller will briefly outline perspectives and limitations.

---

Deep Learning in Speech Recognition

Talk by Alex Acero, Apple Computer at Stanford

While neural networks had been used in speech recognition in the early 1990s, they did not outperform the traditional machine learning approaches until 2010, when Alex’s team members at Microsoft Research demonstrated the superiority of Deep Neural Networks (DNN) for large vocabulary speech recognition systems. The speech community rapidly adopted deep learning, followed by the image processing community, and many other disciplines. In this talk I will give an introduction to speech recognition, go over the fundamentals of deep learning, explained what it took for the speech recognition field to adopt deep learning, and how that has been contributed to popularize personal assistants like Siri.

Machine learning:

Field of study that gives Computers the ability to learn without being explicitly programmed.

‒ Arthur Samuel (1959)

Improve on Task T, with respect to performance metric P, based on experience E

‒ Tom Mitchell (1988)

---

Face Value: The Irresistible Influence of First Impressions

Talk by Alexander Todorov at Google

Dr. Alexander Todorov, head of the Social Perception Lab at Princeton University, discusses his new book, “Face Value: The Irresistible Influence of First Impressions”.

The book discusses his research, as well as that of others in his field, on the ways in which we make up our minds about strangers after seeing their faces for less than a second. These snap judgments can predict who wins elections, who gets bank loans, and even who is wrongfully convicted of crimes.

I hate interviews, because unstructured interviews have zero utility. Well, they do, correlation between future success is .10 (point ten).That is basically 3% of the variance. Why do we interview people? In this particular case, you have enough evidence for past performance and that should be enough to make your decision. You interact with a person for 30 minutes and 30 minutes is too little time and you know, if you are shy you may not do well because it is anxiety provoking situation. If you look closely at the evidence, interview are not very good. Letters of reference are much better. They are not perfect but they summarize samples of observation over much longer time. It is not my impression over 30 minutes of informal discussion. If it’s an important decision you should try to look for good reliable evidence that is indicative of whatever you are trying to achieve in this particular situation.

Darwin was almost denied the chance to take the historic Beagle voyage on account of his nose. The captain [a fan of Lavater] did not believe that a person with such a nose would “possess sufficient energy and determination.” “But I think, he was afterwards well-satisfied that my nose had spoken falsely.”

I hate interviews, because unstructured interviews have zero utility. Well, they do, correlation between future success is .10 (point ten).That is basically 3% of the variance. Why do we interview people? In this particular case, you have enough evidence for past performance and that should be enough to make your decision. You interact with a person for 30 minutes and 30 minutes is too little time and you know, if you are shy you may not do well because it is anxiety provoking situation. If you look closely at the evidence, interview are not very good. Letters of reference are much better. They are not perfect but they summarize samples of observation over much longer time. It is not my impression over 30 minutes of informal discussion. If it’s an important decision you should try to look for good reliable evidence that is indicative of whatever you are trying to achieve in this particular situation.

The scientific story of first impressions—and why the snap character judgments we make from faces are irresistible but usually incorrect

We make up our minds about others after seeing their faces for a fraction of a second—and these snap judgments predict all kinds of important decisions. For example, politicians who simply look more competent are more likely to win elections. Yet the character judgments we make from faces are as inaccurate as they are irresistible; in most situations, we would guess more accurately if we ignored faces. So why do we put so much stock in these widely shared impressions? What is their purpose if they are completely unreliable? In this book, Alexander Todorov, one of the world’s leading researchers on the subject, answers these questions as he tells the story of the modern science of first impressions.

Drawing on psychology, cognitive science, neuroscience, computer science, and other fields, this accessible and richly illustrated book describes cutting-edge research and puts it in the context of the history of efforts to read personality from faces. Todorov describes how we have evolved the ability to read basic social signals and momentary emotional states from faces, using a network of brain regions dedicated to the processing of faces. Yet contrary to the nineteenth-century pseudoscience of physiognomy and even some of today’s psychologists, faces don’t provide us a map to the personalities of others. Rather, the impressions we draw from faces reveal a map of our own biases and stereotypes.

A fascinating scientific account of first impressions, Face Value explains why we pay so much attention to faces, why they lead us astray, and what our judgments actually tell us.

---

Brain waves reflect different types of learning

Reprinted from MIT News
by Becky Ham, MIT News correspondent

Figuring out how to pedal a bike and memorizing the rules of chess require two different types of learning, and now for the first time, researchers have been able to distinguish each type of learning by the brain-wave patterns it produces.

These distinct neural signatures could guide scientists as they study the underlying neurobiology of how we both learn motor skills and work through complex cognitive tasks, says Earl K. Miller, the Picower Professor of Neuroscience at the Picower Institute for Learning and Memory and the Department of Brain and Cognitive Sciences, and senior author of a paper describing the findings in the Oct. 11 edition of Neuron.

When neurons fire, they produce electrical signals that combine to form brain waves that oscillate at different frequencies. “Our ultimate goal is to help people with learning and memory deficits,” notes Miller. “We might find a way to stimulate the human brain or optimize training techniques to mitigate those deficits.”

The neural signatures could help identify changes in learning strategies that occur in diseases such as Alzheimer’s, with an eye to diagnosing these diseases earlier or enhancing certain types of learning to help patients cope with the disorder, says Roman F. Loonis, a graduate student in the Miller Lab and first author of the paper. Picower Institute research scientist Scott L. Brincat and former MIT postdoc Evan G. Antzoulatos, now at the University of California at Davis, are co-authors.

Explicit versus implicit learning

Scientists used to think all learning was the same, Miller explains, until they learned about patients such as the famous Henry Molaison or “H.M.,” who developed severe amnesia in 1953 after having part of his brain removed in an operation to control his epileptic seizures. Molaison couldn’t remember eating breakfast a few minutes after the meal, but he was able to learn and retain motor skills that he learned, such as tracing objects like a five-pointed star in a mirror.

“H.M. and other amnesiacs got better at these skills over time, even though they had no memory of doing these things before,” Miller says.

The divide revealed that the brain engages in two types of learning and memory — explicit and implicit.

Explicit learning “is learning that you have conscious awareness of, when you think about what you’re learning and you can articulate what you’ve learned, like memorizing a long passage in a book or learning the steps of a complex game like chess,” Miller explains.

“Implicit learning is the opposite. You might call it motor skill learning or muscle memory, the kind of learning that you don’t have conscious access to, like learning to ride a bike or to juggle,” he adds. “By doing it you get better and better at it, but you can’t really articulate what you’re learning.”

Many tasks, like learning to play a new piece of music, require both kinds of learning, he notes.

Brain waves from earlier studies

When the MIT researchers studied the behavior of animals learning different tasks, they found signs that different tasks might require either explicit or implicit learning. In tasks that required comparing and matching two things, for instance, the animals appeared to use both correct and incorrect answers to improve their next matches, indicating an explicit form of learning. But in a task where the animals learned to move their gaze one direction or another in response to different visual patterns, they only improved their performance in response to correct answers, suggesting implicit learning.

What’s more, the researchers found, these different types of behavior are accompanied by different patterns of brain waves.

During explicit learning tasks, there was an increase in alpha2-beta brain waves (oscillating at 10-30 hertz) following a correct choice, and an increase delta-theta waves (3-7 hertz) after an incorrect choice. The alpha2-beta waves increased with learning during explicit tasks, then decreased as learning progressed. The researchers also saw signs of a neural spike in activity that occurs in response to behavioral errors, called event-related negativity, only in the tasks that were thought to require explicit learning.

The increase in alpha-2-beta brain waves during explicit learning “could reflect the building of a model of the task,” Miller explains. “And then after the animal learns the task, the alpha-beta rhythms then drop off, because the model is already built.”

By contrast, delta-theta rhythms only increased with correct answers during an implicit learning task, and they decreased during learning. Miller says this pattern could reflect neural “rewiring” that encodes the motor skill during learning.

“This showed us that there are different mechanisms at play during explicit versus implicit learning,” he notes.

Future Boost to Learning

Loonis says the brain wave signatures might be especially useful in shaping how we teach or train a person as they learn a specific task. “If we can detect the kind of learning that’s going on, then we may be able to enhance or provide better feedback for that individual,” he says. “For instance, if they are using implicit learning more, that means they’re more likely relying on positive feedback, and we could modify their learning to take advantage of that.”

The neural signatures could also help detect disorders such as Alzheimer’s disease at an earlier stage, Loonis says. “In Alzheimer’s, a kind of explicit fact learning disappears with dementia, and there can be a reversion to a different kind of implicit learning,” he explains. “Because the one learning system is down, you have to rely on another one.”

Earlier studies have shown that certain parts of the brain such as the hippocampus are more closely related to explicit learning, while areas such as the basal ganglia are more involved in implicit learning. But Miller says that the brain wave study indicates “a lot of overlap in these two systems. They share a lot of the same neural networks.”

The research was funded by the National Institute of Mental Health and the Picower Institute Innovation Fund.

---

Brain scans reveal why rewards and punishments don’t seem to work on teenagers

Reprinted from The Conversation
by Gina Rippon, Professor Emeritus of Cognitive NeuroImaging, Aston University

You cannot swim for new horizons until you have courage to lose sight of the shore.

‒ William Faulkner

Parents and teachers are painfully aware that it’s nearly impossible to get a teenager to focus on what you think is important. Even offering them a bribe or issuing a stern warning will typically fail. There may be many reasons for that, including the teenager’s developing sense of independence and social pressure from friends.

Now a new study, published in Nature Communications, shows that this behaviour may actually be down to how the adolescent brain is wired.

Adolescence is defined as the period of life that starts with the biological changes of puberty and ends when the individual attains a stable, independent role in society. (This definition may leave some readers wistfully pondering the second half of that equation). We now know that it is also a time of tremendous brain reorganisation, which we are only just beginning to understand.

At this point, the brain’s grey matter, which has been growing exuberantly since birth, starts to thin. This is probably due to a system of synaptic pruning, ridding the brain of unnecessary nerve cell connections and resulting in boosted neural efficiency. This thinning occurs from the back to the front of the brain, with the prefrontal cortex, responsible for executive functions such as cognitive control and decision-making, being the last to be tidied up.

Associated with this maturing process are “upgrades” of key structural and functional networks – a shift from local connections to more widespread global links between different parts of the brain.

You don’t need to be a neuroscientist to know that adolescence is also a time of greatly increased impulsivity, sensation-seeking and risk-taking. One aspect of risk behaviour in adolescents appears to be an apparent inability to match their behaviour to the likely rewards (or punishments) that might follow.

A mature brain is quite good at predicting the necessary balance between effort and reward. It does this by using links between the cognitive control systems, found in the highly evolved prefrontal cortex, and the reward circuitry, made up of evolutionarily older sub-cortical structures, which controls motivation and “wanting”. These include the striatum and the anterior cingulate cortex.

Psychologists would describe this skill as the ability to adjust one’s cognitive performance to environmental demands, whereas business gurus would refer to it as “cost-benefit analysis”. Colloquially we might decide whether or not “the game is worth the candle”.

So is it possible that the adolescent brain organisation is not yet up to the task of this careful balancing act? This would come from an unsophisticated reward system, which has not yet been dampened by input from a more conservative, forward-planning prediction system based on cognition.

High versus low stakes

The new study shows that this really is the case – looking at the brains of individuals from 13-20 years old. They did this by collecting data from functional magnetic resonance imaging, which measures brain activity indirectly by tracking changes in blood flow, from participants while they played a video game. This was a cognitive test giving players either high or low financial rewards or punishments in return for correctly sorting pictures of planets.

In this kind of task, you would expect to see improved performance when there are higher stakes involved. But the study showed that this was only the case for older participants (19-20 years old). Younger players were less efficient at the task whether the stakes were high or low. The defining characteristic of brain activity in the better performers was increased use of the prefrontal areas and, perhaps crucially, more powerful connections between the prefrontal cortex and the sub-cortical striatal areas.

Effectively, this study demonstrates the emerging efficiency of a “cool” cognitive control system moderating a “hot” motivational assessment system, resulting in the appropriate balance between the rewards offered and the actions required to maximise performance. If your brain is younger, you are simply not very good at matching what you need to do with what you will gain if you get it right or lose if you get it wrong. This is indeed evidence of an adolescent lack of the necessary fine-tuning in the reward system which (thankfully) appears to emerge with age.

Interestingly, this is different from brain activity linked to adult impulsivity and sensation-seeking – which is associated with general under-responsiveness of the reward system rather than a simultaneous lack of connectivity with the control system.

Knowing about this effect could be of value in educational and training fields. Just increasing any reward/bribe you might be tempted to offer to get a teenager to do something may not have the desired effect. Instead, try to give young adolescents as much information as possible about an upcoming decision – this could help redress the imbalance between cognition and motivation.

For example, instead of bribing them to apply to a certain university, taking them on multiple visits to university open days might just be worthwhile. That said, it may not be easy. There’s also the risk that you’ll be faced with another aspect of adolescent behaviour – a refusal to listen to adult words of wisdom.

We also need to acknowledge that this kind of behaviour is not always a bad thing. There is an evolutionary take that a newly emerging adult needs to take risks, with youthful enthusiasm and excitement unfettered by worthy cognitive controls. As author William Faulkner said: “You cannot swim for new horizons until you have courage to lose sight of the shore.”