Model-based programming and AI-assisted software development
0 likes695 views
Jussi Rintanen Professor of Artificial Intelligence and Software Systems – Aalto University Jussi is a computer scientist with interests in automating the solution of complex problems which earlier have required the use of human intelligence. His research belongs to the areas of automated reasoning, combinatorial search and optimization, and artificial intelligence.
![How do we program computers?
mov ax,@data
mov ds,ax
mov bx,si_ze
dec bx
xx:
mov cx,bx
mov si,0
yy:
mov al,a[si]
inc si
cmp al,a[si]
jb zz
xchg al,a[si]
mov a[si-1],al
zz:
loop yy
dec bx
jnz xx](https://image.slidesharecdn.com/2-190109125651/85/Model-based-programming-and-AI-assisted-software-development-3-320.jpg)
![How do we program computers?
for (c = 0; c < n; c++)
scanf("%d", &array[c]);
for (c = 0 ; c < n - 1; c++)
{
for (d = 0 ; d < n - c - 1; d++)
{
if (array[d] > array[d+1]) /* For decreasing order use < */
{
swap = array[d];
array[d] = array[d+1];
array[d+1] = swap;
}
}
}](https://image.slidesharecdn.com/2-190109125651/85/Model-based-programming-and-AI-assisted-software-development-5-320.jpg)
![Example
A Facebook-style social media application
Who can post on a wall?
• The owner of the wall
• Friends of the wall owner
Result: Post appears on the
wall and in all relevant feeds.
action post(p : feedData, wallOwner : user) {
wallOwner == currentuser
or currentuser in wallOwner.friends
==>
new ent : feedEntry [ t = now,
entry = p,
comments = [],
auth = currentuser.auth,
owner = currentuser ];
wallOwner.wall = concat(wallOwner.wall,[ent]);
currentuser.posts = concat(wallOwner.wall,[ent]);
forall (u in currentuser.friends)
u.feed = concat(u.feed,[ent]);
}](https://image.slidesharecdn.com/2-190109125651/85/Model-based-programming-and-AI-assisted-software-development-18-320.jpg)
![Example
Peer-reviewing for an academic journal
Some articles can be published
as a new volume if
• the current user is the editor
of the journal,
• the articles are unpublished
and belong to the journal,
• the set of articles is
non-empty.
action createNewVolume( p : article set,
j : journal ) {
currentuser == j.editor
and subseteq(p,j.unpublishedarticles)
and card(p) > 0
==>
j.lastvolumenumber = j.lastvolumenumber + 1;
j.unpublishedarticles = j.unpublishedarticles p;
new v : volume [ volumenumber = j.lastvolumenumber,
publicationtime = now,
articles = p ];
j.publishedvolumes = j.publishedvolumes U { v };
forall (a in p) a.volume = v;
}](https://image.slidesharecdn.com/2-190109125651/85/Model-based-programming-and-AI-assisted-software-development-19-320.jpg)
More Related Content
Similar to Model-based programming and AI-assisted software development (20)
More from Eficode (20)
Recently uploaded (20)
Model-based programming and AI-assisted software development
- 1. Model-Based Methods and Automation of Software Production Jussi Rintanen Department of Computer Science, Aalto University December 13, 2018
- 2. The Problem • Massive programmer shortage: • Finland: 7000+ • USA: 200000+ • Programmer productivity too low • Programming is a craft industry Completely new kind of solutions are needed!
- 3. How do we program computers? mov ax,@data mov ds,ax mov bx,si_ze dec bx xx: mov cx,bx mov si,0 yy: mov al,a[si] inc si cmp al,a[si] jb zz xchg al,a[si] mov a[si-1],al zz: loop yy dec bx jnz xx
- 4. How do we program computers? 40 N = N - 1 70 ISCAN = 1 IOK = 1 ISTOP = N IF(ISTART - ISTOP) 50, 110, 110 50 IF(NUMS(ISCAN) - NUMS(ISCAN+1)) 90,90,60 60 J = NUMS(ISCAN) NUMS(ISCAN) = NUMS(ISCAN+1) NUMS(ISCAN+1) = J IOK = 0 90 IF(ISCAN - (ISTOP - 1)) 80,100,100 80 ISCAN = ISCAN + 1 GOTO 50 100 IF(IOK) 105,105,110 105 ISTOP = ISTOP - 1 GOTO 70 110 DO 120 I=1, N 120 PRINT 130, NUMS(I) 130 FORMAT(I10) STOP END
- 5. How do we program computers? for (c = 0; c < n; c++) scanf("%d", &array[c]); for (c = 0 ; c < n - 1; c++) { for (d = 0 ; d < n - c - 1; d++) { if (array[d] > array[d+1]) /* For decreasing order use < */ { swap = array[d]; array[d] = array[d+1]; array[d+1] = swap; } } }
- 6. What are programs good for? • Easy to compile to machine code • Easy to execute • Easy to understand what it really does, in terms of machine code
- 7. What are programs not so good for? • Difficult to prove correct • Difficult to analyze and reason about • Difficult to use for other purposes than execute (This is why testing beats correctness proofs) • Difficult to understand (by programmer, other people)
- 8. Model-Based Approach to Software SW for managing electricity networks Model of a network: • power lines (topology) • switches • protection devices • transformers • customer connections Functionalities: • Monitoring and diagnosis • Control and reconfiguration • Management of maintenance crews • Medium and long-term planning of network maintenance, modifications, extensions
- 9. Model-Based Approach to Software SW for data-intensive applications Model of an application: • Data model • Rules governing change of data • Permissions to access, modify (who? when?) Functionalities: • Analyze the model • Implement the model (code) • Automate tasks (A.I.)
- 10. Our Goals • Make (some of) SW development engineering • Systematic process to develop SW specifications • “AI” to do (almost) all of the routine coding • Lift the abstraction level of code written by programmers • Shift focus from code to business/organizational processes • Result: • Decreased development effort • Easier testing and validation • More modifiable and extendible SW • Independence from execution environment (PL, OS, DBMS, UI)
- 11. Our Goals: Longer Term • Intelligent agents (A.I.) supporting (and replacing) humans in organizational roles • Networks of interacting and inter-operating SW systems • Fully automated companies, agencies, services, ... • Integration with physical assets and infrastructure • management of primary industries, energy, manufacturing, transportation and logistics, services, ...
- 12. EIAI / aitocode: Software Production users HTML/HTTP mobile apps email/messaging SW DBMS cloud storage external SW
- 13. How Is Software Produced? 1 Precisely specify what the software is to do 2 Analyze software specification (automatic) 3 Choose: • Programming language • Database system(s) • Operating system / execution environment 4 Synthesize software system (automatic) 5 Deploy software system (automatic)
- 14. Abstract View of Software 1 A university employee can establish a new course in a given faculty, if the employee is the director of education of the faculty. 2 A university employee can change the responsible teacher of a course, if the employee is the director of education for the faculty responsible for the course. 3 A university employee can change the requirements for a B.Sc. degree granted by a given faculty, if the employee is the director of education of the faculty. 4 A student can be granted a B.Sc. degree, if the student has passed all courses required for the degree. 5 A student can register for a course, if the student has the right to study at the university, and the course is part of the studies for a degree the student is studying for. 6 A student can register for an exam, if the student is registered for the course. 7 A university employee can record the exam grade for a student, if the student was registered for the exam, and the exam is for a course for which the employee is the responsible teacher. 8 A student is recorded as having passed a course, if the student has passed an exam for the course and fulfilled other requirements (for example, done obligatory home assignments and a course project.) 9 A university employee can mark a course as passed for a student, if the employee is the responsible teacher of the course, and the student was enrolled on the course.
- 15. Rules of an Application Each of the informal rules can formalized as three parts: • Inputs • What does the user supply when performing the action? • Each input has a type (int, string, set, list, ...) • Condition • What condition has to hold for the action to be possible? • Boolean combination of atomic conditions (and, or, not, ...) • Also limits the possible inputs! • Effect • How does the action modify the data (in the DB) • One or more pieces of “code”
- 16. Rules of an Application Each of the informal rules can formalized as three parts: • Inputs • What does the user supply when performing the action? • Each input has a type (int, string, set, list, ...) • Condition • What condition has to hold for the action to be possible? • Boolean combination of atomic conditions (and, or, not, ...) • Also limits the possible inputs! • Effect • How does the action modify the data (in the DB) • One or more pieces of “code”
- 17. Rules of an Application Each of the informal rules can formalized as three parts: • Inputs • What does the user supply when performing the action? • Each input has a type (int, string, set, list, ...) • Condition • What condition has to hold for the action to be possible? • Boolean combination of atomic conditions (and, or, not, ...) • Also limits the possible inputs! • Effect • How does the action modify the data (in the DB) • One or more pieces of “code”
- 18. Example A Facebook-style social media application Who can post on a wall? • The owner of the wall • Friends of the wall owner Result: Post appears on the wall and in all relevant feeds. action post(p : feedData, wallOwner : user) { wallOwner == currentuser or currentuser in wallOwner.friends ==> new ent : feedEntry [ t = now, entry = p, comments = [], auth = currentuser.auth, owner = currentuser ]; wallOwner.wall = concat(wallOwner.wall,[ent]); currentuser.posts = concat(wallOwner.wall,[ent]); forall (u in currentuser.friends) u.feed = concat(u.feed,[ent]); }
- 19. Example Peer-reviewing for an academic journal Some articles can be published as a new volume if • the current user is the editor of the journal, • the articles are unpublished and belong to the journal, • the set of articles is non-empty. action createNewVolume( p : article set, j : journal ) { currentuser == j.editor and subseteq(p,j.unpublishedarticles) and card(p) > 0 ==> j.lastvolumenumber = j.lastvolumenumber + 1; j.unpublishedarticles = j.unpublishedarticles p; new v : volume [ volumenumber = j.lastvolumenumber, publicationtime = now, articles = p ]; j.publishedvolumes = j.publishedvolumes U { v }; forall (a in p) a.volume = v; }
- 20. How Is A Rule Implemented? 1 Is current user able to take the action? (UI menus etc.) • Can precondition be true, for some inputs? • Conceptually, must go through all possible input values 2 What input options to offer user? (UI input forms) • Which input combinations can make precondition true? • Conceptually, must produce all input combinations 3 Execute the rule • Is precondition (with inputs) true? (easy) • Execute the “Effects” part (easy) • Implemented as a DBMS transaction First two are both NP-hard (or harder)!
- 21. How Is A Rule Implemented? 1 Is current user able to take the action? (UI menus etc.) • Can precondition be true, for some inputs? • Conceptually, must go through all possible input values 2 What input options to offer user? (UI input forms) • Which input combinations can make precondition true? • Conceptually, must produce all input combinations 3 Execute the rule • Is precondition (with inputs) true? (easy) • Execute the “Effects” part (easy) • Implemented as a DBMS transaction First two are both NP-hard (or harder)!
- 22. How Is A Rule Implemented? 1 Is current user able to take the action? (UI menus etc.) • Can precondition be true, for some inputs? • Conceptually, must go through all possible input values 2 What input options to offer user? (UI input forms) • Which input combinations can make precondition true? • Conceptually, must produce all input combinations 3 Execute the rule • Is precondition (with inputs) true? (easy) • Execute the “Effects” part (easy) • Implemented as a DBMS transaction First two are both NP-hard (or harder)!
- 23. EIAI / aitocode: What We Have Now Project since 2016, with 2018 funding from Business Finland • Core technology completed • Prototype implementation works • Demo applications running • peer-reviewing and publication for academic journals (10 rules) • university course management (5 rules) • social media (“Facebook”) (7 rules) • blogging website (3 rules) • web chat (2 rules) • Other applications under development
- 24. EIAI / aitocode: Next Steps • IDE integration (Q3/2019) • Alternative PL, OS, DBMS, UI (Q3/2019) • Mobile apps (2020) • Verification/validation (2020) • correctness proofs (w.r.t. stated properties) • spec anomalies: unreachable and redundant actions & data • Intelligence (2021) • optimal data input (w.r.t. evaluation of a condition) • suggest actions to human users (w.r.t. users’ goals) • take action sequences autonomously • minimization of a cost measure, maximization of a reward measure
- 25. Interested? We are looking for collaboration partners: • Companies to participate in technology trials • Software developers willing to try out • Everything else to transfer the technology to real use
- 26. Conclusion • Automation of software production • data and knowledge-intensive information systems • web applications • et cetera • Future: • Increased intelligence (automated embedding of “AI” in SW) • Highly automated analytics and validation Code is not interesting, systems and their features are!