Transcript New AI 2
A New Artificial Intelligence 2 Kevin Warwick Nature –versus-Nurture • How much of our intelligence is due to our genes (programming) and how much is due to learning (the environment)? • Twin studies • Adopted siblings • Typical latest – 80/20 genes/nature. Intelligence Tests • An intelligence test measures the degree to which tester and subject think alike – Subjective Intelligence. • “Everyone is World Champion at some game” (Ross Ashby) – You are the most intelligent person in the world !!! • Positive links with examination results – indicates performance in specific areas Subjective Intelligence reminder • What we regard as being an intelligent act, and what not, is very subjective • Human centred – e.g. jokes • Animal centred – bat’s ultrasonic sense • Machine centred – maths, memory Intelligence Hypersphere • Intelligence is not something that can be indicated by a single number (IQ) • Intelligence is multifacetted Intelligence Tests? • In the 19th Century Frances Galton exhibited tests • • • • in Science Museum, London. These included: putting in weight order boxes different by 1 gram The closeness of two points on the back of your hand before you couldn’t tell the difference measuring the speed of response to a noise. Felt to be no scientific basis for such tests as there were no statistical links between the results of the test and how well individuals performed at school Binet • The IQ test was formulated by Alfred Binet in • • • • 1904. Asked to develop a simple method to identify children who would struggle in the normal school environment Concentrated on memory, comprehension, imagination, moral understanding, motor skills and attention. His test (for children between 3 and 12) was made up of 30 parts which children worked through from the start until they could no longer continue - oral The number reached was the ‘mental’ age. Subtracting the answer from the actual age the intellectual level was given Can we improve mental performance? • What have studies shown thus far? (Orange/Lead) • Fun with IQ tests!!! • Two sets of results. • Q1: If I wish to improve my mental performance in the short term – what can I do? • Q2: Can the food I regularly eat affect my mental performance? What do Intelligence tests show? • Validity of (IQ) tests has been questioned • • • recently Validity has been shown by statistical links those who do well in school exams tend to also do well in IQ tests. Whether IQ tests indicate anything about intelligence remains unanswered. But there are strong statistical correlations between exam performance (hence IQ test performance) and job status (not with job performance). IQ test performance gives a likely indication of general exam performance and such tests have been used to show the effects of lifestyle and activity – whatever one thinks of IQ tests such results are fascinating. Statistical Evidence? • Changing an IQ score by 3 points is a strong indicator • Regular Vitamin C intake in children has been shown to • • • • • improve their IQ score by 8 points (on average). Pollution appears to have little effect – doubling lead intake (pretty heavy) reduces a score by only 1 point. Bottle-fed babies fare worse on IQ tests than breast-fed babies (3 points). Children who regularly use a dummy score 3 to 5 points lower (in later life) than those who do not. Children whose Mothers were above 35 score 6 points higher Obviously there are social links associated with each of the indicators - it is difficult to separate items from such aspects Experiment 1: • 200 first year students at Reading University given an IQ test, then an activity + food, then another IQ test. • How did their results change over that short time period? (IQ points score) • Trying to answer the question “What is it best to do immediately before an examination, to maximize your performance?” Results - Food • Alcohol • Chocolate • Coffee • Orange Juice • Peanuts 0 -2 +3 -2 +1 Results - Activity • • • • • • • • • • Reading/Swatting Listening to classical music Watching a chat show on TV Playing with a construction toy Sitting/Chatting Watching a documentary on TV Walking Meditating Watching Friends on TV Completing a crossword puzzle -6 -2 +5 -4 -2 +4 +1 +2 +1 0 Experiment 2: • 50 children (aged 8-11) at Thameside School, Caversham were given regular breakfasts over a one month period. • Tested at start/end to see how they altered • IQ equivalent tests in numerical ability/word association/shape recognition. Results – Breakfast • Toast + Orange Juice • Bacon Sandwich • Control • Cereal • Eggs (various) +3 +3 0 -1 -5 Nature – v - Nurture • One of the most important and yet contentious issues with regard to intelligence is how does it originate? • Is it natural/programmed or is it learnt through education and experience? • In the make up of an individual’s intelligence, what percentage is inherited and what percentage is due to life’s environmental effects? Intelligence in History • Plato saw a person’s intelligence as being class related - to maintain the status quo people should only produce offspring with members of their own class. • At that time, average levels of intelligence were further maintained by killing children at birth (or infancy) if they were seen to display characteristics of ‘idiocy’. • In Aristotle’s time, things had changed. Levels of intelligence were considered to be dependent on teaching and life experience. Aristotle said that intelligence was present in all citizens. • This may sound quite radical, however slaves, laborers, many women and most foreigners were all excluded from citizenship and therefore from being intelligent. Charles Darwin • Darwin’s publication of Origin of species by means of natural selection (1859) led to huge • • • support for the genetic nature of intelligence Bolstered the idea of different levels of intelligence between nations, races, classes and individuals - to justify slavery/oppression. Poorer people should be allowed to die out in order that society can maintain a higher average level of intelligence. This meant that poor people were not given social welfare and, in some parts of the world, were not allowed to breed. Foetal Development • Some recent studies have put great emphasis on • • • the environment before birth. An article in Nature claimed that foetal development in the womb accounted for 20% of an individual’s total intelligence - genetic influences only accounted for 34%, the remaining 46% due to environmental factors. But these percentages are contrary to the norm Straw pole of research papers indicates 60-80% being down to inheritance with the remaining 40-20% being due to education and training. Adoptees • A study in Denmark looked at 100 men and • • women adopted in and around Copenhagen between 1924 and 1947. The adoptees in the study had little in common, in terms of their environment/education, with their biological siblings, but shared a common upbringing with their adoptive siblings. Biologically related siblings correlated well in terms of occupational status - there was no correlation between adoptive siblings. Twins • Identical twins have a close genetic make up of their • • brains –including the period in the womb. In 1976, in a detailed study of 850 twins, John Loehlin concluded that the make up of intelligence was 80/20 inheritance/environment The group of twins of most interest are those separated at birth and brought up in different environments. In 1966 Cyril Burt presented results on 53 pairs of identical twins who, he claimed, had been separated at birth, randomly placed in their adoptive homes and had had no further contact since birth. He came up with a figure of 86/14 - his results were discredited due to the validity of the twins used! Further Twins • At the University of Minneapolis, a special unit was set • • • up for the study of twins and many interesting statistics have subsequently been obtained Results were pooled on a total of 122 pairs of identical twins in terms of IQ test scores. Similarities between pairs of twins correlated to be 82% (pretty similar to the other results). However, unlike Burt’s claimed study, twins tended to be brought up in similar home backgrounds due to that being the strategy of the social services responsible. In fact results did not correlate so well for those twins who had grown up in dissimilar backgrounds. Anecdotal Example • • • • • • Twins - Jim Springer and Jim Lewis. Adopted by separate Ohio families - grew up independently Met at the age of 39. They found that they: drank the same brand of beer - smoked the same number of the same brand of cigarettes. Both had a basement workshop, both built a circular bench, which they painted white, around a tree trunk. In their youth they both hated spelling but enjoyed mathematics and both had owned dogs which they called ‘Toy’. Both joined the local police force, got promoted to the rank of Deputy Sheriff and left after 7 years. Both married and divorced women called Linda - then married women called Betty, with whom they had 1 son, although Jim Lewis’ child was named James Alan whilst Jim Springer’s child was called James Allan. Both took annual holidays in the same week at the same beach - but they never met. They both took an IQ test and gave identical answers. Comparative Intelligence • Mental and physical abilities are different between species. • Difficult to compare the performance of an individual in • • • one species with an individual from another species, other than in the sense of performance in a specific task. The ability to cover a distance over land in a minimum time – compare a cheetah, a human, an automobile and a snail. The human might finish in the top 3. But the result would only relate to one specific result – speed over a distance – could be seen as silly! The same could be said if we compared a human, with a rabbit and a computer in terms of ability to interact with a second human in a Chinese conversation. Certain computers could do a lot better than many humans who cannot communicate at all in Chinese. Think the Same • Comparing individuals from different species is • • • • meaningless other than in terms of the skills required to complete a task. To compare humans and machines in terms of intelligence we need to be clear which humans we are talking about and which machines. Is the comparison being made in terms of a human-centric task? Can we expect the machine to carry out the task in exactly the same way as a human? The end result is the critical thing, not how the machine performed? One human plays another at chess. The winner is not disqualified because they were thinking about food If a machine beats a human at chess we should not say, yes but it wasn’t thinking in the same way as the human therefore it has lost. Conclusions • Important to consider intelligence in other • • • creatures as well as humans Look at intelligence in humans in terms of the broad spread of individuals that form humanity as a whole and not simply an ‘ideal’. Can be tempting to compare the intellectual abilities of a machine with those of an ‘ideal’ human – in order to assess the standing of AI We need to make sure that we do not make fools of ourselves with naïve conclusions Next • Classical AI Contact Information - UK • Web site: www.kevinwarwick.com • Email: [email protected] • Tel: (44)-1189-318210 • Fax: (44)-1189-318220 • Professor Kevin Warwick, Department of Cybernetics, University of Reading, Whiteknights, Reading, RG6 6AY,UK