100 Days Blog

Day 049 - Artificial intelligence and reverse cyborgs

Submitted by Sam on 8 July, 2011 - 21:52

A cyborg is a fully integrated man-machine system, where a human's natural tolerances or capabilities are extended beyond their normal capacity through machine augmentation. Various existing neural and physical prosthetics have given glimpses of the potentialities for such biological -machine syntheses, but we are objectively a very long way away from creating a perfect man-machine hybrid. The recent phenomenon of internet crowdsourcing, however, has already created hybrid intelligences which outperform our current artificial intelligent agents in a number of ways. Indeed, it has been speculated that perhaps the most intelligent machines in the near future may be “reverse cyborgs”, or artificial intelligences augmented by us 1.

Crowdsourcing allows companies or individuals to outsource work to an open, undefined community, rather than tasking specific employees or contractors with it. Web-based crowdsourcing has lead to the phenomenon of ubiquitous human computing, which is where a task is broken down into smaller, much more basic sub-tasks, which are then parcelled out around the world for completion by anyone with an internet connection. In theory, this allows computers to turn to the human crowd to assist when a problem is encountered that cannot be solved.

In 2005, Amazon launched Mechanical Turk, an online service which has made possible such reverse-cyborg-like systems. Mechanical Turk co-ordinates human intelligence from a crowd of “Workers” with HITs (Human Intelligence Tasks) posed by “Requesters”, who are typically corporations or researchers. Tasks like transcription can now be crowdsourced extremely cheaply, and with great rapidity – the crowd of workers is often thousands-strong, and so jobs can get completed in seconds.

One example of a human-computer hybrid built on Amazon's crowdsourcing platform is IQ Engine's oMoby smartphone app, which joins the company's “visual intelligence” image recognition software with human intelligence. Users take a picture with their mobile phone using the app, which then tries to identify the image with conventional image recognition algorithms. If it's attempts at categorization are unssucessful, the software will upload the image to the crowd on Mechanical Turk, or the company's own pool of workers. Their Director of research, Pierre Garrigues, claims around half of all queries are able to be answered by this hybridised method in under 25 seconds.

  • 1. Giles, Jim. "Brain Donors." New Scientist 2818 (2011). Print.

Day 048 - Imprinting

Submitted by Sam on 8 July, 2011 - 01:33

As genes cannot possibly encode a human's complete value system, our brains have evolved general purpose machinery, such as mirror-neurons, as a means to acquire and transmit goals and values from one person to another. A child therefore not only has to learn about causes and effects itself, but has also to internalize a coherent system of values, and there is no way for it to do so without basing it on a pre-existing model. The child therefore typically constructs its value system from those models expressed by older individuals, and most typically the ideals, values and goals provided by its parents.

Strong bonds of childhood attachment have therefore evolved in order to keep children within the protective and didactic sphere of its parents, giving the child every opportunity to learn and construct an understanding of what should be classed as good and bad. The strength of these filial bonds also ensure that the child isn't exposed to too large a number of close adult role-models, which could potentially result in a fragmentary pastiche of influence that would not result in a coherent personality. Instead, the attachment mechanism restricts the child's attention to only a few very close (genetically) models, simplifying the child's task.

Marvin Minsky has suggested that the strong attachment-bonds between human children and parents could be based on memories which are rapid to form but extremely slow to degrade, perhaps using memory structures descended from the forms of learning called “imprinting”, which is where some very young animals such as chicks, ducklings and goslings quickly 'imprint' a recognition of their parents in the early days of their life. Minsky reasons that whilst primates and humans don't actually imprint in this way, infants and parents do develop an analogous bond serving the same evolutionary purpose of providing offspring with a strong role-model as the basis of their personalities, and the foundation of their value system.

Minsky's postulation draws attention to a spectrum of behavioural patterns, ranging from the purely innate to the fully-learned. In many animals, innate behavioural mechanisms provide the methods to learn new behaviours, just as a kitten has the instinct to hunt a rat, but must learn how to follow it from its mother. The brain's innate actions facilitate the learning process, and help to maintain a proximity to the object of its attachment, affording the infant a good chance of survival.

In less complex animals, the hard-wired ability to rapidly imprint usually leads to an immediate recognition of the animal's parent, helping to protect the child from potentially threatening events that the adult can help protect against. However, as Austrian naturalist Konrad Lorenz showed in the mid-20th century, imprinting is a very rigid mechanism for forging filial bonds, and is vulnerable to attribution errors. Lorenz discovered that the greylag geese he reared right from hatching would treat him like the parent, imprinting him as the first sensory object that they met. The geese would follow him around, and as adults would court him in preference to other greylag geese. Mammals have a much more robust method of forming parental bonds, which are more complex and generate bonds which are strong and long-lasting, ensuring that children are likely to see their parents (and often their mother) as the ultimate protector and provider; a ready-made consolidated value machine ripe for emulation.

Day 047 - Male and female brains

Submitted by Sam on 7 July, 2011 - 00:25

In 2001, Simon Baron-Cohen and other researchers from the University of Cambridge published a paper exploring whether human sexual dimorphism in sociability (i.e. anecdotal and scientifically observed differences in behaviour between men and women) is a result of biological or socio-cultural differences between the two sexes. The study examined 102 new-born babies (only a day old so that they would not have had time to be influenced by social and cultural factors), testing whether there was a difference in the amount of time the boys and girls spent looking at a face compared to a mechanical mobile. Averaged over the group, their results showed that the male infants spent longer looking at mechanical objects whilst the females showed a stronger interest in the faces 1 , suggesting that the sex differences arise from a biological (i.e. innate neurological) difference, rather than through the social environment.

The male preference for systems is supported by other psychological tests, including the on-average better performance of males at mental rotation tests, where the task is to identify whether a shape is a rotation or a mirror of another. Males are generally both quicker and more accurate at such tests than females, who score more highly in tests measuring empathy. Twelve-month old baby girls, for example, expressively signal more emotional concern through facial expressions and vocalizations than boys of a similar age when confronted with the distress of other people. Women, in general, are more sensitive to facial expressions, and tend to score higher than men when asked to identify what expression a particular image depicts.

These differences, which emerge as statistical trends when groups are compared, have lead Simon Baron-Cohen to create the 'empathizing-systemizing theory', which identifies three common brain types. In this scheme, the 'male' brain is predisposed towards understanding and building of systems; the 'female' brain is hard-wired for empathy, and the 'balanced' brain is equally attuned to both systemizing and empathizing. Crucial to the theory is that your sex doesn't necessarily determine which type of brain you have, as men can have the female brain and women can have the male brain, but trends suggest that men are more likely to have 'male' brains and women more likely to have 'female' brains.

Neurologically, it could be suggested that each brain type is determined by the number or efficiency of its mirror neurons, and thus that male brains have fewer mirror-neurons than female brains.

The BBC has a comprehensive set of six online tests which help identify the 'sex' of your brain, whilst an adaptation of Baron-Cohen's emotional empathy image test is available online here.

I scored 30/36 on the emotional empathy image test, and the BBC test tells me I have a male brain, as I can't spot the difference, and have a preference for more feminine faces.

  • 1. Connellan, Jennifer, Simon Baron-Cohen, Sally Wheelwright, Anna Batki, and Jag Ahluwalia. "Sex Differences in Human Neonatal Social Perception." Infant Behavior and Development 23 (2001): 113-18. Print.

Day 046 - Mirror-touch synaesthesia

Submitted by Sam on 6 July, 2011 - 03:20

The hypothesis that we emphasize with others through a process of simulation enabled by our mirror-neurons is supported by a rare and recently discovered form of synaesthesia called mirror-touch synaesthesia. People with this condition are considered to be hyper-empathetic and actually feel tactile sensations in their own body when observing other humans being touched.

The somatosensory cortex of the brain receives inputs from the body when it is touched, with subregions corresponding to each part of the body becoming active in a systematic response to each stimulus. Neuroimaging studies have shown that people with mirror-touch synaesthesia have a hyperactivated somatosensory cortex when they observe other people being touched, personally feeling the sensation themselves. The synaesthetes' sensations of observed touches are indistinguishable from those produced when they themselves are actually touched, making the usual empathetic simulation a very real reality for them.

Synaesthetes with hyperactive mirror-neurons consistently score higher in questionnaires designed to measure empathy, correlating the relationship between mirror neurons and understanding of others. In her article for the Guardian earlier this year, synaesthete Fiona Torrance described herself as “hugely considerate of other people”, feeling as though she knew “exactly what it feels like to be them”. However, like the people at the other end of the empathetic scale (such as those autism spectrum disorders – a much more common condition), Fiona's extreme level of empathy carries with it a great many difficulties. She has struggled with weight loss through always feeling full by seeing other people eat, constantly crying because she felt someone else's pain, and suffering sleep deprivation as other people's feelings permeate and invade her own.

Fiona's condition gives an insight into the neurological basis for empathy, showing how the wiring of the brain can completely control a person's emotional quotient. The implications are profound, particularly for the criminal-justice system. A hard-wired lack of empathy (or the inverse) would act as a determinate of a person's behaviour that they could be totally unaware of, and helpless to override. How can a compassionate and civilized society punish offenders whose actions are determined, at least in part, by neurological machinery totally outside of their own conscious control?

Day 045 - Unpacking subconscious and involuntary social cues

Submitted by Sam on 5 July, 2011 - 01:07

We react to numerous social cues that we give off and perceive during conversation, but often only subconsciously. Unlike the easy-to-fake facial expressions that give indications of our emotional state, many of these signals – such as as variation in the tone and pitch of the voice and gesture mirroring – are transmitted involuntarily, and if we were able to consciously process and analyze them we might have a more holistic understanding of the dynamics of social situations, and perhaps might react more deliberately. Various technologies have been developed to capture and process these signals programmatically, driven by a proven ability to raise group productivity in commercial environments by drawing attention to and elucidating a previously unconscious layer of social interaction.

Vertex Data Science, a multi-million pound private company based in the UK and one of the largest providers of call centre outsourcing in the world deployed such socially-decoding technology in 2006 to analyze the speech patterns of its telephone operators. Alex Pentland and his colleagues from MIT's Media Lab developed the equipment to measure variations in tone and pitch, the physical voice signal, not the semantic content nor the logic of the conversations themselves. After only a few seconds of such audio data the team were able to accurately predict the ultimate success or failure of almost every sales call.

The sensors revealed that people are less susceptible to meaning and reasoning than they are to the unconscious and instinctual aspects of communication, showing in particular the technology’s commercial value by demonstrating how telephone operators could increase their sales performance by consciously controlling previously unconscious aspects of their communication, learning to vary the tone and pitch of their voice in the manner of the most successful callers.

Other work from the team supports this evidence, showing the subtle differences between successful teams and ineffective teams, revealing new ways in which companies can train and support collaboration in order to out-compete rivals using a new type of computer-augmented social awareness. A year before the Vertex Data Science deployment, the team worked with researchers from the MIT Sloan School of Management with similar equipment, recording and analyzing the bodily movements and tone of voice of participants and again ignoring the meaning of the spoken words themselves. The participants were assigned to play a middle manager taking a new job, and another playing the role of a vice president of that division. Their task was to negotiate the middle manager's salary package, with real monetary rewards on offer to motivate the actors. Whilst the negotiations could last for hours, the team's electronic sensor took only a few minutes to predict with 87% accuracy which party would win the negotiation, demonstrating how empirical data can lead not only to a more reliable understanding of how businesses work, but also how the unconscious side of human behaviour may one day be fully soluble through the aid of new technologies.

Day 044 - Mirror neurons and the foundations of civilization

Submitted by Sam on 3 July, 2011 - 22:36

VS Ramachandran identifies a point of significant development in the course of human evolution about 75,000 years ago, when we suddenly and rapidly acquired and spread a great number of skills, which were unique to our species. The development of tool use, the use of fire, shelters, language and the ability to read someone else's expressions and interpret their behaviour can all be traced to the sudden emergence of a sophisticated mirror neuron system, Ramachandran argues.

The very rapid emergence of these skills occurred with no significant change in the size of the human brain, which stabilized almost three or four thousand years ago. Mirror neurons constitute only a very small part of the brain, and so their emergence 75,000 years ago is certainly a possibility in this regard.

As we have seen, mirror neurons allow for virtual simulations of other people's behaviours and emotions, and so their development would have allowed accidental discoveries made by individuals in a group (such as the use of fire) to be emulated and imitated by others, allowing skills to spread throughout the population rather than remaining isolated with their discoverer. In this way skills would rapidly accumulate, propagating vertically across generations, enabling a new kind of social evolution whereby a child can learn complex skills from its parents in ten minutes that otherwise might take hundreds of thousands of years to become evolutionarily encoded.

Below is Ramachandran's TED talk from 2009 in which he discusses the implications of mirror neurons for human development.

Day 043 - Augmented emotional reality

Submitted by Sam on 3 July, 2011 - 01:13

Whilst emotional contagion is an effective means of transmitting information and engendering synchronized emotions in a group, the effects are only beneficial if the emotions being transmitted are appropriately recognized and interpreted. Misreading of emotion is surprisingly common in human social interaction as the spectrum of human emotion manifests itself through facial and bodily indicators so subtle and myriad that they are often only partially observed and partially classified, leading to miscommunication and potentially negative social consequences.

In the pursuit of improving our ability to correctly identify these signals (often referred to as our emotional intelligence), various technologies are being developed to artificially classify emotional expression with an above-human level of accuracy, aiming to help us better understand each other by removing ambiguity from the interpretation of physically expressed emotions.

Rosalind Picard from the MIT Media Lab has developed a prototype of one such technology to boost our emotional intelligence – a pair of glasses containing a camera connected to a computer which interprets facial expression. The camera sends a feed to software which analyzes faces for the thousands of tiny muscle movements which constitute expressions, interpreting them and relaying them back to the wearer either though earphones or a computer screen. By tracking twenty-four “feature points” on the face, the software analyzes micro-expressions and compares them with its database of six known expressions, divided into “thinking”, “agreeing”, “concentrating”, “interested”, “confused” and “disagreeing”, correctly identifying them 64% of the time, compared to the human accuracy of 54%. These figures indicate that this model could aid not only people suffering from impaired emotional intelligence (such as those with autism), but could in fact assist the majority of people in sensing the mood of the people they are talking to.

The prototype consequently carries commercial value, and the team behind the glasses have established a company called Affectiva selling their expression-recognition software to companies wishing to measure how people feel about their adverts or products, for instance.

However, the recognition of emotional states from facial expressions is still far from foolproof, and can be subject to the same manipulations and misreadings that effect human interpretation. The goal of complete emotional knowledge and perfect brain-state sharing remains the domain of connectome projects like The Blue Brain Project.

Day 042 - Mirroring brain states

Submitted by Sam on 2 July, 2011 - 02:46

The principle of emotional contagion describes the tendency of emotional states to quickly spread around a group through the mimicry of expressions and behaviour, allowing the rapid dissemination of information through the transference of emotional signals. This system of non-verbal interaction enables rapid communication about risk and reward, and is beneficial to groups of social animals, mediating and facilitating interaction.

The discovery of mirror neurons by a group of Italian neurophysiologists in the 1990s has triggered research which is uncovering the neural mechanisms responsible for this phenomenon, showing how the actions of individual neurons can correlate to emotions and the empathetic understanding of emotions in others.

When a person (or a primate – the original experimental evidence was derived from macaque monkeys) performs a specific action, certain neurons in the front of their brain fire, commanding or encoding that particular action. A subset of these neurons, the mirror neurons, will also fire when that person sees another person performing the same action, “mirroring” the other person's view by performing a neural-simulation of their action.

A variety of research has corroborated the relationships between mirror neurons and emotional transference. An fMRI study was conducted in which participants were presented with malodorous smells and then videos showing the emotional facial expressions of disgust. The core findings of the study concluded that the same sites in the brain were activated by the actual disgusting stimulus as by the observation of faces expressing disgust, showing that merely observing an emotion activates the neural representation of that emotion. 1

The findings of this study illustrate how emotional contagion can function as a primitive mechanism to protect young infants (and young monkeys) from food-poisoning, by transferring the neural correlate of the emotion of disgust through facial expressions alone.

This is a highly effective method to transfer knowledge, goals and values without the need for sophisticated cognitive skills, relying on a “common substrate for feeling” which allows people to simulate each other in order to empathetically understand them, all without the need for language.

  • 1. Wicker, Bruno, Christian Keysers, Jane Plailly, Jean-Pierre Royet, Vittorio Gallese, and Giacomo Rizzolatti. "Both of Us Disgusted in My Insula The Common Neural Basis of Seeing and Feeling Disgust." Neuron 40.3 (2003): 655-64. Print.

Day 041 - Emotional transference of goals and values

Submitted by Sam on 1 July, 2011 - 01:55

In order for a system to be able to have goals, it must be able to make value judgements. Genetic inheritance can account for some extremely primitive judgements about what is good for the organism and what is bad, but most such distinctions have to be learned. For highly complex organisms in highly dynamical environments, it becomes evolutionary advantageous (and then necessary) to parse experience into transferable wisdom to allow useful goals and values to be transmitted from organism to organism without the need for each individual in each generation to invent them themselves through first-hand experience. Accordingly, in human society, we have established various forms of cultural knowledge, traditions and heritages which can be passed on from one group to another in a variety of forms. The easiest way to encode this information now is to use language, but there are more basic, “general-purpose” mechanisms whereby higher-organisms can transfer things like values and goals between each other. Perhaps the expression of recognizable emotions is one of the most effective ways of transmitting ready-made brain-states from one organism to another.

If we learn to associate the visual input of a tiger with the emotion “fear” (perhaps by first learning how “fear” expresses itself through the body-language and facial expressions of peers, and then observing these indications expressed when they also see the tiger), then we can rapidly activate all the agents in our brain associated with “fear” ourselves, thereby bypassing the learn-by-first-hand-experience mode of knowledge acquisition and instantly activating the same groups of agents in our brain that have proven evolutionarily advantageous to be activated in similar situations.

In this way, emotions might be seen to suppress certain features of normal thinking, and activating specific thought-processes associated with each particular emotion. By having a common emotional heritage (reflecting a common neural architecture), we can reliably learn which stimuli should trigger (sub-consciously perhaps) the group of agents associated with emotions such as love and fear, dread and happiness, by seeing when they are expressed by others.

The elemental distinctions of pain and pleasure constitute the primary drivers of emotion, functioning respectively to suppress all but one goal – remove the pain and maintain the pleasurable state. Emotions are complexes of pleasure and pain, leading to aggregates like “anger” which can perform a number of inhibitory and excitatory roles, such as disabling slower brain processes, most long-range goals and plans, high-level reasoning, but activating fast-reacting processes and special gestures and expressions.

Animals which can express and 'read' emotions gain an ability to turn on specific tailored ways of thinking when confronted with new problems, activating and suppressing agents in an evolutionarily optimized fashion.

Minsky touches on some of these implications of emotions in his rather sprawling talk at MIT in September 2007, below.

Day 040 - Think the same way with the investment principle

Submitted by Sam on 30 June, 2011 - 01:47

It is extremely difficult to challenge long-held beliefs. The modular, additive mechanism of mind proposed by Minsky offers a compelling argument as to why this is the case, through what he describes as “the investment principle”. This is the tendency for humans to approach new situations and unfamiliar problems using modes of thinking that they have learned perform well in other scenarios, rather than try to confront novel environments with entirely novel ways of thinking. This tendency arises because our oldest ideas have an unfair advantage over ideas which arrive later – we have invested more in them by layering dependent skills on top of them. Older ideas will therefore have a greater number of skills (or ways of thinking, beliefs, conceptions) reliant on them than newer ideas, which will have to compete against this large mass of skills in order to become established as beneficial modes of thinking.

Whilst this pattern of thinking minimizes the amount of energy that the brain uses by providing it with strategies which work most of the time, thereby preventing the constant reformulation of policies for dealing with the world, the investment principle does however place certain constraints on how open-minded one can ever really be when confronted with entirely new events. Through the hierarchical layering of agents relying on other agents creating interdependent skill-sets and thought-pathways, we come to apply patterns of past experience to present and future behaviours, reinforcing our longest-held beliefs by continuing to invest time and energy in them, galvanising them by mobilizing them in novel situations, even if they are not adequately equipped to deal with them.

Evolution provides clear illustrations of how this process of investment can become enslaving to development. Once the pattern of centralizing neural networks in an organism's head became evolutionarily beneficial, a delicate network of anatomical dependencies became established around this structure that could not be drastically disturbed by random mutation without a high chance of deleterious side-effects. Whilst parcelling the brain into the head is excellent for most organisms, some could perhaps be better served by a brain placed in a different location – woodpeckers, for example. So many fundamental structures have based themselves around the generally reliable and generally efficient principle of “brain goes in the head” that some organisms are stuck with sub-optimal anatomies.

As a generalizing tool, the investment principle works most of the time, but it is critical that we are aware of its effects. As our brains mature, and as our patterns of thought become more and more entrenched, it becomes harder and harder to dislodge them even in the face of highly compelling new ideas. An awareness of the investment principle can help understand why it is so difficult to challenge working beliefs, and perhaps facilitate introspective change.

Syndicate content
Drupal theme by Kiwi Themes.