1 / 0

“Truly creative thinking, of course , will always remain beyond the power of any machine.”

“Truly creative thinking, of course , will always remain beyond the power of any machine.”. Creativity and AI. Aakash Rao N S Gaurav Singh Chauhan Siram Bhargav Karnati. Outline. What is Creativity? Creativity : Is it Magic? Why Model Creativity? The Superhuman Fallacy

rianna
Télécharger la présentation

“Truly creative thinking, of course , will always remain beyond the power of any machine.”

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. “Truly creative thinking, of course, will always remain beyond the power of any machine.”

  2. Creativity and AI

    AakashRao N S Gaurav Singh Chauhan SiramBhargavKarnati
  3. Outline What is Creativity? Creativity : Is it Magic? Why Model Creativity? The Superhuman Fallacy Types of Creativity Combinational Creativity Computer Combinations Copycat JAPE Exploratory Creativity Computer Explorations BACON AARON EMI Transformational Creativity Computer Transformations Genetic Algorithms Genetic Images Evaluation of Ideas Summary Conclusion
  4. What is Creativity? The ability to generate novel, and valuable, ideas.
  5. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable:
  6. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting
  7. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting Useful
  8. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting Useful Beautiful
  9. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting Useful Beautiful Simple
  10. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting Useful Beautiful Simple Richly Complex
  11. What is Creativity? The ability to generate novel, and valuable, ideas. Valuable: Interesting Useful Beautiful Simple Richly Complex And so on…
  12. What is Creativity? The ability to generate novel, and valuable, ideas. Ideas :
  13. What is Creativity? The ability to generate novel, and valuable, ideas. Ideas : Concepts, Theories, Stories
  14. What is Creativity? The ability to generate novel, and valuable, ideas. Ideas : Concepts, Theories, Stories Sculptures, Buildings
  15. What is Creativity? The ability to generate novel, and valuable, ideas. Ideas : Concepts, Theories, Stories Sculptures, Buildings Paintings, Music, Movies
  16. What is Creativity? The ability to generate novel, and valuable, ideas. Ideas : Concepts, Theories, Stories Sculptures, Buildings Paintings, Music, Movies Almost absolutely anything
  17. What is Creativity? The ability to generate novel, and valuable, ideas. Novel :
  18. What is Creativity? The ability to generate novel, and valuable, ideas. Novel : P-creative : New to the person who generated it Jokes, Puns, Intuition, Insight
  19. What is Creativity? The ability to generate novel, and valuable, ideas. Novel : P-creative : New to the person who generated it Jokes, Puns, Intuition, Insight H-creative : P-creative and has never occurred before in history Discovery, Invention, Theory
  20. Creativity : Is it Magic? Creativity produces something out of nothing. How do you explain that? Original creations break the mold, they’re the products of geniuses. Creative Ideas are Flashes of Inspiration. They just happen. It is a special faculty granted to a tiny elite.
  21. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities.
  22. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities. Association of ideas
  23. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities. Association of ideas Perception
  24. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities. Association of ideas Perception Analogical thinking
  25. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities. Association of ideas Perception Analogical thinking Exploring a structured problem-space
  26. Creativity : Is it Magic? Creativity is not a special faculty, nor a psychological property confined to a tiny elite. It is a feature of human intelligence in general, grounded in everyday capacities. Association of ideas Perception Analogical thinking Exploring a structured problem-space Reflective self-criticism
  27. Why Model Creativity? Creativity is a fundamental feature of human intelligence, and thus is an inescapable challenge for AI. Even technologically oriented AI cannot ignore it, for creative programs could be very useful in the laboratory or the market-place. Creative Ideas have shaped human progress and development, and thus it is a natural goal for AI to develop artificial agents that can do the same. For a better understanding of how human themselves come up with creative ideas.
  28. The Superhuman Fallacy P-creativity in computers need not match all the previous achievements of human beings. We shouldn’t say that AI has failed simply because it can’t match the heights of human intelligence. After all, most of us can’t do that either. We should try to understand mundane thinking first, and worry about the exceptional cases only much later. If AI cannot simulate the rich creativity of Shakespeare and Mozart, it doesn’t follow that it can teach us nothing about the sorts of processes that go on in human minds—including theirs—when people think new thoughts
  29. Types of Creativity Combination : Unfamiliar combinations of familiar ideas Exploration: Generation of novel ideas by the exploration of structured conceptual spaces. Transformation : Transformation of one or more dimension of the space, so that new structures can be generated which could not have arisen before.
  30. Combinational Creativity By connecting ideas together creative leaps can be made, producing some of history’s biggest breakthroughs.
  31. Combinational Creativity By connecting ideas together creative leaps can be made, producing some of history’s biggest breakthroughs. Poetic Imagery and Figurative Language are examples of Combinational Creativity “All the world's a stage, And all the men and women merely players”
  32. Combinational Creativity By connecting ideas together creative leaps can be made, producing some of history’s biggest breakthroughs. Cartesian Geometry is the result of the combination of Geometry and Algebra.
  33. Combinational Creativity By connecting ideas together creative leaps can be made, producing some of history’s biggest breakthroughs. The Ford Motor Company didn’t invent the assembly line, interchangeable parts or even the automobile itself. But they combined all these elements in 1908 to produce the first mass market car, the Model T.
  34. Combinational Creativity By connecting ideas together creative leaps can be made, producing some of history’s biggest breakthroughs. The Internet slowly grew over several decades as networks and protocols merged. It finally hit critical mass in 1991 when Tim Berners-Lee added the World Wide Web.
  35. Computer Combinations Combinational creativity is the easiest for human beings to achieve. A rich store of world knowledge (including cultural knowledge) Making associations doesn’t have to be learned: it’s a natural feature of associative memory Perspectives and criteria to evaluate ideas and modify them
  36. Computer Combinations The very same factors make it the most difficult to achieve for a Computer. Lack of access to a rich store of knowledge Lack of an associative memory and search process Lack of evaluation criteria It is easy to generate random combinations, but not many of them may be valuable.
  37. Copycat A nondeterministic analogy-making computer program developed over several years by Melanie Mitchell and Douglas Hofstadter(1993) Uses an associative memory, whose nodes are connected by links that represent conceptual associations like sameness, successor, predecessor, alphabetic-first, alphabetic-last, left, right, direction, leftmost, rightmost, middle, group, sameness-group, successor-group, predecessor-group, group-length, 1, 2, 3, opposite, alphabetic-position, letter-category etc. Rates results i.e. analogies as creative based on several criteria, such as structure, order, degree of randomness.
  38. Copycat
  39. Copycat
  40. JAPE “Joke Analysis and Production Engine” Developed by Graeme Ritchie and Kim Binstedin 1994. Designed to generate question-answer-type puns.
  41. Some Punning Riddles by JAPE Q: What do you call a strange market? A: A bizarre bazaar. Q: What is the difference between leaves and a car? A: One you brush and rake, the other you rush and brake. Q : What kind of murderer has fibre? A : A cereal killer.
  42. Confusable Texts If an utterance of a spelling is ambiguous, then it has more than one written forms. Theses forms are completely confusable. Like “serial” and “cereal” The word segment “spec” and “spook” are not spelled or pronounced the same, yet one can be substituted for the other. Ex: What does a near-sighted ghost wear? Spooktacles!!! There are many other types of confusable texts
  43. Strategies used There are three main strategies used in puns to exploit spelling or word sense ambiguity Juxtaposition It is the most simple mechanism, simply placing the confusing segments near each other and treating them as normal construction Ex: What do you call a weird market? A bizarre bazaar.
  44. Strategies used There are three main strategies used in puns to exploit spelling or word sense ambiguity Juxtaposition It is the most simple mechanism, simply placing the confusing segments near each other and treating them as normal construction Jokes which use juxtaposition alone are very weak, as they do little more than demonstrate an unexpected phonological similarity so, it is mostly used in combination with others
  45. Strategies used Substitution This mechanism works by substituting one confusable segment for another, as part of a larger text, and using the resulting text as if it were a sensible construction. Ex:- The word “purr” is confusable with the first syllable of “purgatory”. If we substitute “purr” for “pur”, we get the constructed text “purrgatory”. We must also construct a plausible interpretation for the construction Where do cats go when they die? Purrgatory
  46. Strategies used Comparison This mechanism explicitly compares two confusable text, usually by asking for similarities or differences in the question part of the riddle. Positive Comparison (How is A like B?) Negative Comparison (What is difference between A and B?) What’s the difference between a short witch and a deer running from the hunters? A: One’s a stunted hag and the other is hunted stag.!!!! “stunted hag” and “hunted stag” are confusable texts
  47. Joke Generation : Step 1 Schemata It constrains and assert relations between lexical items and constructed items (including descriptions)
  48. Step 2 Small adequate description(SAD) SAD for lemon is {class(lemon, citrus fruit), specifier (lemon, yellow), adjective(lemon, sour), inact_verb(lemon, eat), has(seeds)} SAD generator for non-lexicalized items Ex: lemon-aide SAD for aide:{synonym(aide, assistant), act_verb(aide, help), adjective(aide, helpful)} Hence SAD FOR Lemon-aide iare set of SAD’S : {class(lemon_aide, assistant), specifier(lemon_aide, sour)}, {class(lemon_aide, assistant), inact_verb(eat)}
  49. Step 3 Templates: They transform a set of relations into a suitable surface form for a punning riddle(in our case question-answer) It consists of variables which are to be instantiated to text segments and lexemes, constraints on the relations the template can be given, and a sentence forminto which the words and lexemes the template is given can be slotted
  50. Step 3 Ex:{Relations :describes (NPWF, {class (Lex, Class), specifier (Lex, Spec)}) SF: What do you call np([Spec, Class]) ? Det (NPWF) NPWF where np, Det use a simple grammar to construct Instantiating NPWF by “low-comotive”, Class by “Train” , Spec by “depressed” Then we get Q:What do you call a depressed train ? “low-comotive”
  51. Exploratory Creativity In exploratory creativity, aperson moves through a conceptual space, exploring it to find out what’s there.
  52. Exploratory Creativity In exploratory creativity, aperson moves through a conceptual space, exploring it to find out what’s there. “Genres” in art, music etc. are examples of conceptual spaces. Most artists spend their lifetimes exploring and mastering these conceptual spaces, and produce marvelous results.
  53. Exploratory Creativity In exploratory creativity, aperson moves through a conceptual space, exploring it to find out what’s there. In the most interesting cases, we discover both the potential and the limits of the space in question.
  54. Exploratory Creativity The counter-intuitive results due to Einstein’s Theory of Relativity are due to systematic explorations based on two fundamental laws: The Principle of Relativity The Principle of Invariant Light Speed
  55. Computer Explorations Rules of the relevant thinking style need to be specified clearly enough to be put into a computer program Modeling exploratory creativity requires not only advanced AI skills but also expertise in, and deep insights into, the domain concerned.
  56. Computer Explorations Despite the difficulties, there has been much greater success here than in modeling combinational creativity. In many exploratory models, the computer comes up with results that are comparable to those of highly competent, sometimes even superlative, human professionals.
  57. BACON The BACON family of programs are programs meant to rediscover simple scientific equations given : Experimental data, Information on experimental apparatus. Some inherent feeling about the equation structure. Developed by Pat Langley, Herbert Simon, Gary Bradshaw, Jan Zytkow. The BACON family : BACON-1 (1978) BACON-3 (1979) BACON-4 (1980) BACON-5 (1981)
  58. BACON-1 Works by comparing attributes of objects and creating new ones until if invents one that does not work. It has 3 operators for creating a new object: If two existing attributes A1 and A2increase linearly relative to each other, it will create two new attributes S and I which are the slope and y-intercept of the line. These new attributes are constant across all objects. When this consistency is detected during the next round the equation finding process will stop. If two attributes A1 and A2 increase together, but not linearly then BACON will create a new attribute AR, which is the ratio between A1 and A2. If attribute A1 increases while A2 decreases then BACON will create a new attribute AP, which is the product of A1 and A2. Purely Observational
  59. BACON 3 Improvises on BACON-1 by adding the concept of active experimentation. Uses BACON-1’s heuristics and more. Partitions its attributes into independent ones and dependent ones. If dependent attributes exist, the system tries to fix all independent ones except one. Then varies the left over independent attribute systematically with a dependent one and, seeing the dependent variable’s change, tries to arrive at a relationship.
  60. BACON-3 in Action How does BACON-3 derive the Ideal Gas Law? Data on gas samples is supplied to BACON-3 to observe the dependent Pressure(P) by systematically varying quantity (n) with domain {1, 2, 3}, temperature(T) with domain {10, 20, 30} and volume (V) with domain {1000, 2000, 3000}. Since dependent attributes exist, the system tries to fix all independent ones except one. We choose to fix quantity(n) and temperature(T). Next the system derives 9 attribute-to-value sets: (n=1, T=10), (n=2, T=10), (n=3, T=10) (n=1, T=20), (n=2, T=20), (n=3, T=20) (n=1, T=30), (n=2, T=30), (n=3, T=30)
  61. BACON-3 in Action Within each set derived before, it varies pressure p and finds that volume(V) varies inversely. Creates new attribute A as the product of p & V (using heuristic 3) Next it systematically tries to vary A with the next independent attribute T to find a linear relationship. Given as A = bT +c. The constants b & c vary only with quantity (n). Finally it varies quantity (n) with constants b & c found previously and finds direct relationships between them. Creates new constants D = b/n & E = c/n. The equation them becomes : A = DnT + En. i.e. PV = nDT + nE With actual experimental data it would give PV = nR(T+273)
  62. Restrictions of BACON-3 Can handle only numeric attributes Attribute values in reality may not be numeric always For e.g. symbolic attribute substance can have values wood, metal, plastic and these values may have numeric attributes like density, conductivity etc. These describe the intrinsic properties of physical objects made using those materials.
  63. BACON-4 BACON-4 was built to overcome the limitations of BACON-3. It extended BACON-3 with an operator to define a new numeric attribute Anew given the pair (As, An). Here, AS is an existing symbolic attribute and AN is an existing numeric attribute. New attribute Anew' will be defined to be equal to existing attribute AN, but it represents a brand new attribute. BACON-4 rediscovered Black’s Law of Specific Heat.
  64. BACON-5 With new operators and more attributes, the search space grew exponentially. To reduce the search space, BACON-5 was implemented. BACON-5 has a heuristic to search for structurally symmetric equations before non-symmetric ones.
  65. Limitations of BACON Series The BACON series, and its allied programs were limited in their discovery ability. They were not equipped to handle noise or missing values. This limitation was a serious one in the sense that they always rediscovered rather than discovered any brand new scientific equation. They had to use data. It would have been impossible for them to postulate Newton's Law of Gravity or Einstein's E = mc2 given the same backgrounds as Newton or Einstein because both of these equations arose from purely theoretical considerations, not empirical ones.
  66. AARON – The Colorist First robot in human history to paint original art. AARON mixes its own paints, creates striking artwork and even washes its own brushes  “I’m a first class colorist, But AARON is a world class colorist.” -- Harold Cohen
  67. AARON’s Works
  68. How AARON Works We “tell” AARON what it knows. Declarative knowledge & Procedural knowledge. Declarative knowledge How long arms and legs are. Make a list of parts: -- left-upper-arm, torso, etc. each of which is a list of all points in that part with the position of the point in relation with the origin of the part. For example, the origin of "left-upper-arm" is "left-shoulder", and "left-elbow" is at some position in relation to it: so much below, so much to the left, so much in front. 
  69. How AARON Works The program has to know how to go about doing things, and this "procedural knowledge" is usually represented in the form of rules. if (left-arm-posture is "hand-on-hip")     (add-upper-arm left -.3 .5 .65)  else if (left-arm-posture is "arms-folded")    ... 
  70. EMI (Emmy) Developed by David Cope over 2001-2006, Emmy (EMI : Experiments in Musical Intelligence) composes music in the style of composers such as Bach, Beethoven, Chopin, Mahler. They are remarkably compelling, striking many musically literate listeners—though admittedly not all—as far superior to mere pastiche (an imitation). It employs powerful musical grammars expressed as ATNs. In addition, it uses lists of “signatures”: melodic, harmonic, metric, and ornamental motifs characteristic of individual composers Using general rules to vary and intertwine these, it often composes a musical phrase near-identical to a signature that has not been provided.
  71. Transformational Creativity In transformational creativity, the space or style itself is transformed by altering (or dropping) one or more of its defining dimensions. As a result, ideas can now be generated that simply could not have been generated before the change. The more fundamental the dimension concerned, and the more powerful the transformation, the more surprising the new ideas will be.
  72. Transformational Creativity The Three Musicians, Pablo Picasso, 1921 (Cubism)
  73. Transformational Creativity Impression, Soleil Levant, by Claude Monet, 1872 (Impressionism)
  74. Transformational Creativity The Electric Guitar
  75. Transformational Creativity Non-Euclidean Geometry: The essential difference between Euclidean and non-Euclidean geometry is the nature of parallel lines. It states that, within a two-dimensional plane, for any given line l and a point A, which is not on l, there is exactly one line through A that does not intersect l. In hyperbolic geometry, by contrast, there are infinitely many lines through A not intersecting l, while in elliptic geometry, any line through A intersects l
  76. Transformational Creativity
  77. Computer Transformations Given a style, a computer can explore it, but how can it come up with a new style? The rules and instructions specified in the program determine its possible performance (including its responses to input from the outside world), and there’s no going beyond them. The program must, therefore, include rules for changing itself.
  78. Genetic Algorithms GAs can make random changes in the program’s own task-oriented rules. Selection : Parent chromosomes for a new generation are selected from the current generation using a fitness function Crossover: Selects genes from parent chromosomes, combines them and creates a new offspring. Mutation: Alters one or more chromosomes from their initial states.
  79. Karl Sims’ Genetic Images Genetic Images (1991), is a graphics program that produces varied images from which a human being selects one or two for breeding the next generation (due to lack of an automatic fitness function) Image generating programs from the parents are concatenated, nested and mutations are applied to produce a new generation of images. This system often generates images that differ radically from their predecessors, with no visible family resemblance, due to mutations.
  80. Karl Sims’ Genetic Images
  81. Evaluation of Ideas It is more difficult to express (verbally or computationally) just what it is that we like about a painting, or a song, than it is to recognize something as a valuable idea. And to say what it is that we like (or even dislike) about a new, or previously unfamiliar, form of music or painting is even more challenging. Identifying the criteria we use in our evaluations is hard enough, but justifying, or even (causally) explaining, our reliance on those criteria is more difficult still. To make matters worse, human values-and therefore the novelties which we are prepared to approve as “creative” change from culture to culture, and from time to time.
  82. Evaluation of Ideas The main reason why most current AI-models of creativity attempt only exploration, and not transformation and combination, is that the resulting structures may not have any interest or value. A truly automatic AI-creator would have to have evaluative mechanisms sufficiently powerful to realize the poor quality of the new constructions, and drop (or amend) the transformations and combinations accordingly. Currently, most systems depend either on constant interaction and feedback from the human world, or on culturally accepted notions of value.
  83. Summary Creativity is the ability to generate novel and valuable ideas. Creativity can be of three types : Combinational, Exploratory and Transformational. Combinational creativity involves generation of unfamiliar ideas by combining familiar but unrelated ideas. JAPE and Copycat employ combinational creativity. Exploratory creativity involves exploration of a structured conceptual space to discover new ideas . Emmy, AARON and BACON employ exploratory creativity. In transformational creativity, one or more dimensions of the conceptual space are altered/dropped to allow for generation of ideas that could not be generated earlier. Transformational creativity is implemented using genetic algorithms, as seen in Karl Sims’s Genetic Images.
  84. “Truly creative thinking, of course, will always remain beyond the power of any machine.”

  85. Conclusion Most people today would not agree that artificial agents can be creative, for numerous reasons. It is not the job of AI scientists to fight philosophical battles over whether computers can be “truly” creative. As programmers, they can certainly do what they do best, ask for a set of requirements and constraints, and build a system that satisfies them. As is clear from the examples covered here, it would be foolish to completely dismiss the idea of creativity in artificial agents. This is, therefore, probably more a question of societal acceptance, rather than plausibility. As more challenges are met, and better creative agents are built, the current view is bound to change.
  86. References Boden, M. A. 2009. Computer Models of Creativity, Artificial Intelligence Magazine, Fall 2009 Issue. Boden, M. A. 1998. Creativity and Artificial Intelligence, Science Direct, Volume 103, August 1998. Binsted, K., Pain, H., and Ritchie, G. D. 1997. Children’s Evaluation of Computer-Generated Punning Riddles. Pragmatics and Cognition5(2): 305–354. Hofstadter, D. R. 2002. How Could a COPYCAT Ever be Creative? In Creativity, Cognition, and Knowledge: An Interaction, ed. T. Dartnall, 405–424. London: Praeger. Cohen, H. 1995. The Further Exploits of AARON Painter. In Constructions of the Mind: Artificial Intelligence and the Humanities, ed. S. Franchi and G. Guzeldere. Special edition of Stanford Humanities Review, 4(2): 141–160.
  87. References Langley, P. W.; Simon, H. A.; Bradshaw, G. L.; and ZytkowJ. M. 1987. Scientific Discovery: Computational Explorations of the Creative Process. Cambridge, MA: The MIT Press. Simon, H. A. 1995. Explaining the Ineffable: AI on the Topics of Intuition, Insight, and Inspiration. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, vol. 1, 939–948. San Francisco: Morgan Kaufmann Publishers. Cope, D. 2006. Computer Models of Musical Creativity. Cambridge, Mass.: The MIT Press. Ferguson K. 2011. Everything is a Remix, Part 3 of 4. Wikipedia, The Free Encyclopedia Google Images
  88. Thank You.

More Related