synergies between symbolic and sub-symbolic artificial...
TRANSCRIPT
![Page 1: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/1.jpg)
Synergies Between Symbolic and Sub-symbolicArtificial Intelligence
Thomas Bolander, Associate Professor, DTU Compute
Current Trends in AI, 23 November 2016
(c_e)L[^GA=f]2 (F[_E_B])L[=A,_Ac]L[=E,_B,_E]- [E,B,E]2L[F,=B,=E]2 L[^F,C=F]
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 1/17
![Page 2: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/2.jpg)
A bit about myself
Thomas Bolander
• Associate professor in logic andartificial intelligence at DTUCompute (since 2007).
• Member of the newly establishedSiri Commission (future of AI inDK. Est. by Ida Auken and IDA).
• Current research: How to equipAI systems with a Theory ofMind (ToM)?
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 2/17
![Page 3: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/3.jpg)
Content
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 3/17
![Page 4: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/4.jpg)
Symbolic vs sub-symbolic AIThe symbolic paradigm (50s untiltoday): Simulates human symbolic,conscious reasoning. Explicit/symbolicworld models. Search, planning, logicalreasoning.
↑robust, predictable
flexible, learning
↓The sub-symbolic paradigm (80s untiltoday): Simulates the fundamentalphysical (neural) processes in the brain.Artificial neural networks.
symbolic
↓
↑sub-symbolic
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 4/17
![Page 5: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/5.jpg)
Symbolic or sub-symbolic: both!
• Combined symbolic and sub-symbolic methods (00s untiltoday): Integration of the (so far) competing paradigms. E.g. GoogleDeepmind (learning arcade game) and AlphaGo.
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 5/17
![Page 6: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/6.jpg)
Aspects of human cognition... and of AI• Perception (sub-symbolic)• (Knowledge representation and) reasoning (symbolic)• Learning (sub-symbolic)• Planning (symbolic)• Interaction (multi-agent systems) (symbolic)
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 6/17
![Page 7: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/7.jpg)
Combining approaches and paradigms
Last sentence of the 2016 report of the One Hundred Year Study onArtificial Intelligence (Stanford University, September 2016):
Although the separation of AI into subfields has enabled deeptechnical progress along several different fronts, synthesizingintelligence at any reasonable scale invariably requires manydifferent ideas to be integrated. For example, the AlphaGoprogram that recently defeated the current human champion atthe game of Go used multiple machine learning algorithms fortraining itself, and also used a sophisticated search procedurewhile playing the game.
Another recent succesful example of integrating symbolic AI(reinforcement learning) and sub-symbolic AI (deep neural networks):Google DeepMind learning to play Artari arcade games (Nature,2015)...
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 7/17
![Page 8: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/8.jpg)
DeepMind playing Breakout
http://www2.compute.dtu.dk/~tobo/DeepMind.MP4
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 8/17
![Page 9: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/9.jpg)
DeepMind playing Pac-Man
http://www2.compute.dtu.dk/~tobo/DeepMindPacman.mp4
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 9/17
![Page 10: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/10.jpg)
Automated planning
Automated planning:• Computing plans (seq. of actions) leading to desired outcomes.• More precisely: Given a goal formula, an initial state and a set of
possible actions, an automated planner outputs a plan that leadsfrom the initial state to a state satisfying the goal formula.
• Actions are represented compactly via action models (actionschemas).
Example.Goal: Get A on B on C .
C
B
A
initial state
B
C A
C A B
B
C A
A
B
C
goal
· · ·
Put(C,table)
Put(B,table)
Put(B,C) Put(A,B)
Put(B,C)
· · ·
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 10/17
![Page 11: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/11.jpg)
Learning action models
Action models are traditionally manually crafted (no learning).
Learning approaches to learning action models:• Machine learning approaches (e.g. kernel perceptrons). (Mourao et al.,
2012)
• Logical/qualitative approaches. (Walsh & Littman, 2008; Amir & Chang,
2008; Bolander & Gierasimczuk, 2015)
Learning the right action model for Pac-Man generalises to all levels ofthe game, opposite the reinforcement learning of DeepMind!
Example of an action model (in ADL). (Pacman with one ghost and nomagic pills)
Action: Move(startpos, dir)Precondition: Neighbour(startpos, resultpos, dir) ∧ ¬Dead(pacman)Effect:At(pacman, resultpos) ∧ ¬At(pacman, startpos) ∧ ¬At(resultpos, pill) ∧when At(ghost, resultpos): Dead(pacman)
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 11/17
![Page 12: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/12.jpg)
Social robot bartender
(Petrick & Foster, ICAPS 2013)Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 12/17
![Page 13: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/13.jpg)
Robustness, trust, explanations, generalisability,predictability...
... are arguments in favour of symbolic and logical methods.
CoBot, Carnegie Mellon U.
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 13/17
![Page 14: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/14.jpg)
Proposal: A multi-paradigmatic competition in AI
http:
//www2.compute.dtu.dk/~tobo/AImuffins_on_MAsojourner.mov
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 14/17
![Page 15: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/15.jpg)
What does it take to build a socially intelligentrobot?
Social intelligence: The ability tounderstand others and the socialcontext effectively and thus to interactwith other agents successfully.
Theory of Mind (ToM): The ability ofattributing mental states—beliefs,intentions, desires, etc.—to otheragents.
Theory of Mind (ToM) is essential tohuman social intelligence (Baron-Cohen,
1997).
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 15/17
![Page 16: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/16.jpg)
Anti-social TUG hospital robots (2009)
Frustrated users of hospital robots in USA:
• “TUG was a hospital worker, and itscolleagues expected it to have somesocial smarts, the absence of which ledto frustration—for example, when italways spoke in the same way in bothquiet and busy situations.”
• “I’m on the phone! If you say ’TUGhas arrived’ one more time I’m goingto kick you in your camera.”
• “It doesn’t have the manners we teachour children. I find it insulting that Istand out of the way for patients... butit just barrels right on.” TUG hospital robot
(Barras, 2009)Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 16/17
![Page 17: Synergies Between Symbolic and Sub-symbolic Artificial Intelligencetobo/trends-in-ai1/Bolander.pdf · 2016. 12. 9. · \I’m on the phone! If you say ’TUG has arrived’ one more](https://reader034.vdocuments.mx/reader034/viewer/2022051806/60010a038cfb2803f037435b/html5/thumbnails/17.jpg)
Helpful robots = Automated Planning +Theory of Mind
Thomas Bolander, Current Trends in AI, 23 November 2016 – p. 17/17