Genetic programming is waiting for better tools

Genetic programming seems to be applicable in a very limited domain. Could the reason be the current state of software engineering?

What if we imagine the future of programming languages (or maybe the forgotten past), functional and homoiconic languages and better APIs (or lack thereof) in the next 20-40 years: could they enable genetic programming to finally thrive and work for common problems?

Computer scientists and engineers were thinking about universal modes of communication between computer programs since the 30s. The current state of API-driven communication is a nightmare by their standards, but we seem to only go deeper in it, solving more and more accidental complexity and losing energy and time.

LISPS continue to show us that good ideas come from simplicity, and the power of composition is the key to fighting complexity. I’d love to discuss how these ideas, if explored deeper, can allow genetic programming to evolve (pun intended) and solve real world problems at last.

Transcript

1. Genetic programming is waiting for better tools Rakhim Davletkaliyev
2. Genetic programming is waiting for better tools Rakhim Davletkaliyev
3. whoami codexpanse.com blog → rakhim.org twitter → @freetonik emacscast.org
4. Nature inspired solutions
5. Nature inspired computing
14. selection crossover mutation
15. Mutation in nature not perfect copying particle interference radiation etc human papillomavirus
16. Mutation in nature not perfect copying particle interference radiation etc human papillomavirus
17. problem domain a solution crossover small change a b + mutation a solution
18. math system 101010011101 crossover + mutation x = 3, y = 12, z = sin 101010011101 111110101010 101010101010 111010101010
19. Genetic programming computing environment AST crossover + mutation import x; var a = 12; modify nodes
20. Not a new idea Automatic programming
 “a euphemism for programming in a higher-level language than was then available to the programmer.” — David Parnas
21. Not a new idea Automatic programming
 “a euphemism for programming in a higher-level language than was then available to the programmer.” — David Parnas Computing Machinery and Intelligence (1950)
 “the Child Machine - simple program which can learn from rewards and punishment” — Alan Turing
22. 1950 "Computing Machinery and Intelligence", Alan Turing 1980 "A Darwinian Approach to Pattern Recognition", Richard Forsyth 1988 "Non-Linear Genetic Algorithms for Solving Problems", John Koza Applications: Curve fitting, data modeling, Symbolic regression, feature selection, classification 2004 — ∞ Human-Competitive Awards ("Humies") History of genetic programming
23. Why is domain limited? • Nature's encoding is richer? • Control and error handling • Scale and computational power
24. Imagine: end-user aps with GP
25. UI/UX Performance Accessibility Language Billing
26. UI/UX Performance Accessibility Language Billing UPDATE
27. UI/UX Performance Accessibility Language Billing Automated Software Transplantation
28. UI/UX Performance Accessibility Language Billing Automated Software Transplantation
29. UI/UX Performance Accessibility Language Billing UPDATE
30. crossover
31. UI/UX Performance Accessibility Language Billing UPDATE crossover INITIAL BUILD
32. Problems 1. Code is fragile 2. Communication is fragile 3. Computing is fragile
33. 1. Code is fragile
34. 1. Code is fragile
35. http://colinm.org/sigbovik/
36. 7 figigigg INVALID
37. Code is fragile • Engineering discipline • Optimistic: SE will improve • Intrinsic fragility / error handing • Interpreter does't really "interpret" • Ignore compilation errors
38. Which is more fragile? 1. Imperative: code → code 2. Constraints, declarative, logical: constraints → code Sketchpad, Ivan Sutherland 1963
39. Problems 1. Code is fragile 2. Communication is fragile 3. Computing is fragile it's gonna be ok
40. "Communicating with aliens"
41. "Communicating with aliens" GET / HTTP/1.1
42. Carl Hewitt actor model Alan Kay OOP Alan Turing duh Claude Shannon information theory
43. Alan Kay OOP method :( parent state
44. Alan Kay OOP
45. Alan Kay OOP negotiate negotiate no assumptions no ack local or remote?
46. Carl Hewitt actor model Alan Kay OOP Alan Turing duh Claude Shannon information theory
47. Carl Hewitt actor model ACTOR ACTOR ACTOR ACTOR
48. Carl Hewitt actor model
49. Carl Hewitt actor model
50. Problems 1. Code is fragile 2. Communication is fragile 3. Computing is fragile it's gonna be ok it's gonna be ok
51. https://rakhim.org/
52. https://rakhim.org/
53. Ferris Wheel of Software Engineering™
54. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas
55. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas Hmm, immutable data structures....
56. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas Hmm, functional programming...
57. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas It's amazing! WHEEE!!!
58. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas Let's do everything in the browser!
59. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas We need a new package manager!!!
60. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas SHIT SHIT SHIT!!!
61. Ferris Wheel of Software Engineering™ fundamental concepts and old ideas Oooh, higher order functions....
62. Problems 1. Code is fragile 2. Communication is fragile 3. Computing is fragile it's gonna be ok it's gonna be ok
63. Computing is fragile
64. Non-quantum scale
65. Non-quantum scale Quantum scale
66. Non-quantum scale Quantum scale
71. "I think I can safely say that nobody understands quantum mechanics." — Richard Feynman
72. "I think I can safely say that nobody understands quantum mechanics." — Richard Feynman "Hvis kvantemekanikken ikke gør dig svimmel, har du ikke forstået noget som helst". — Nils Bohr
73. Conclusions We're not starting from scratch
74. Conclusions We're not starting from scratch This is just the beginning
75. Conclusions We're not starting from scratch This is just the beginning Maybe embrace uncertainty perhaps why not I don't know maybe?..
76. We can only see a short distance ahead, but we can see plenty there that needs to be done. — Alan Turing
77. Thank you rakhim.org @freetonik