Universal programmability how ai can help
1 like1,929 views
The document discusses the potential for natural language programming to allow more people to program computers. It notes that while billions of devices are programmable, few owners can program them. Natural language could make programming accessible to anyone by allowing people to explain tasks in everyday language. Recent advances in language technology such as translation devices, IBM Watson, and Siri show the potential is now higher. The document outlines challenges but argues that with clarification from smart devices, natural language programming may now be viable and help address issues like the growing shortage of programmers. It proposes a "grand challenge" to enable anyone to program through approaches like natural language.
Universal programmability how ai can help
- 1. KIT – Universität des Landes Baden-Württemberg und nationales Forschungszentrum in der Helmholtz-Gemeinschaft RAISE’13 – DAY 2: OVER THE HORIZON www.kit.edu Universal Programmability How AI Can Help Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | May 26, 2013
- 2. 2 26.05.2013 Problem 1 Almost everyone has one or more programmable devices. There are literally billions of chips out there. But only a tiny fraction of the owners of programmable devices can program. Programmability, the most fundamental aspect of computers, is inaccessible to almost everyone. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 3. 3 26.05.2013 Problem 2 Shortage of programmers. In the EU alone there are estimates that the shortage will grow to 900,000 programmers in ten years. Possible fixes: Train more of them. Import them. Select only the best. Match tasks with skill sets. Deskill programming, so more can program. (Automate it.) Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability None of this scales.
- 4. 4 26.05.2013 A Possible Solution Programming in natural language could be done by anyone. People are good at explaining. We teach our children. Recipes, instruction manuals, assembly manuals, books, etc. Requirements elicitation is explaining the domain. Some of the challenges of natural language: Ambiguity Implicit context Implicit assumptions These challenges need to be met. A smart programmable device can ask for clarification. Programming is not a one-way street. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 5. 5 26.05.2013 Why now? Because Language Technology has made tremendous strides Translation devices: Jibbigo Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 6. 6 26.05.2013 IBM Watson wins Jeopardy Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 7. 7 26.05.2013 And the Siri Announcement... Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 8. 8 26.05.2013 Recent Successes of Applying NLP in SE Inferring code contracts from method descriptions Detecting incorrect code comments Programming Excel by example Test script generated from NL requirements Our own contributions: Automatic specification improvement (remove ambiguities by checking knowledge sources and asking the writer) Generating UML diagrams from textual specifications. Round-trip engineering (if you change the class diagram, the requirements text is updated automatically) Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 9. 9 26.05.2013 Example: IETF 3912 WHOIS Protocol A WHOIS server listens on TCP port 43 for requests from WHOIS clients. The WHOIS client makes a text request to the WHOIS server, then the WHOIS server replies with text content. All requests are terminated with ASCII CR and then ASCII LF. The response might contain more than one line of text, so the presence of ASCII CR or ASCII LF characters does not indicate the end of the response. The WHOIS server closes its connection as soon as the output is finished. The closed TCP connection is the indication to the client that the response has been received. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 10. 10 26.05.2013 IETF 3912: WHOIS Protocol Generated Class Diagram Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability ASCII_CR ASCII_LF response WHOIS_client TCP_port_43 WHOIS_server listen(pat : TCP_port_43) : void makes() : void replies() : text_content closes(pat : connection) : void connection closed : boolean requests treminated(instr : ASCII_CR,instr : ASCII_LF) : void end text_content HAB POSS line morethanone RECP DON RECPDON text QUALII QUALPOSSHAB output finished() : void HAB HAB
- 11. 11 26.05.2013 Animation Scripting CMU Alice is a framework for building 3D environments, animations, and games. It is graphic! Program it in English rather than Java. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 12. 12 26.05.2013 Animation Scripting: Prerequisites Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability Getting to know Alice → Ontology building from Alice. What objects are available for programming? Determine the features of the objects. How can objects and actions be described in natural language? (linking the API to NL) Determine method parameters (what if they are not given in the script? or in the wrong type?). Handle control flow (e.g. foreach). How do humans express repetition? They do not spell out a while-loop.
- 13. 13 26.05.2013 Animation Scripting: Scene Setup The ground is covered with grass, the sky is blue. In the background on the left hand side there is a palm tree. In the foreground on the left hand side there is a closed Mailbox facing southeast. Right to the mailbox there is a frog facing east. In the foreground on the right hand side there is a bunny facing southwest. In front of the bunny there is a broccoli. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 14. 14 26.05.2013 Application Area: Programming Robots ARMAR-III is a humanoid robot. Designed to be a household help. Example tasks: „Set the table for three persons.“ „Put cereal and milk on the table.“ “Clear the table and put the dishes in the dishwasher.” You may want to explain a “subroutine”, such as “clearing the table means pick up all the objects on the table and put them on the counter” Why would you want to instruct the robot any other way? Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability
- 15. 15 26.05.2013 The NL Programming Grand Challenge Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability 2014 2015 ??? ??? ???? Scene Descriptions Straight Line Scripts Robot Interaction Loops, Recursion Algorithms Multimodal Programming (pointing, touch, demonstrating)
- 16. 16 26.05.2013 Autonomous Vehicles: DARPA Grand Challenge 1 2007 DARPA Urban Challenge
- 17. 17 26.05.2013 Autonome Fahrzeuge: DARPA Herausforderungen 1 2007 DARPA Urban Challenge In all these cases, concentrated efforts led to substantial progress in a short time. Successful techniques were „borrowed“ from other teams and improved. Could we achieve a similar speed in NL programming?
- 18. 18 26.05.2013 Grand Challenge: Enabling anyone to program Which approaches will enable billions of users to program? To unleash the potential of programmable devices, programming should be accessible to anyone. Many fields have to work together. Walter F. Tichy, Mathias Landhäußer, Sven J. Körner | Universal Programmability Researchers Join Forces!