Powerful But Limited: A DARPA Perspective on AI. Arati Prabhakar Director, DARPA

Transcription

1 Powerful But Limited: A DARPA Perspective on AI Arati Prabhakar Director, DARPA

2 Artificial intelligence Three waves of AI technology (so far) Handcrafted knowledge Statistical learning Contextual adaptation The first wave is still advancing and solving hard problems The second wave is amazingly effective, but it has fundamental limitations New research is shaping the third wave

3 Intelligence is an ability to process information perceive rich, complex and subtle information learn within an environment abstract to create new meanings reason to plan and to decide perceiving learning abstracting reasoning Intelligence scale Artificial intelligence is a programmed ability to process information

4 Handcrafted knowledge Engineers create sets of rules to represent knowledge in well defined domains First-wave AI technologies AI systems reason over narrowly defined problems Planning tools Command Post of the Future No learning capability and poor handling of uncertainty Perceiving Learning Abstracting Reasoning Cybersecurity Expert systems

5 First wave stumbles on natural data DARPA Autonomous Vehicle Grand Challenge 140 miles of dirt tracks in California and Nevada 2004 # completed: # completed: 5 The problem in 2004 Vehicles were able to follow the GPS waypoints very accurately but either missed or hallucinated obstacles ahead The difference in 2005 A probabilistic algorithm interpreted the flood of incoming sensor data to learn the terrain and map out an optimal driving surface

6 Second-wave AI technologies Handcrafted knowledge Engineers create sets of rules to represent knowledge in well defined domains Statistical learning Engineers create statistical models for specific problem domains and train them on big data Apple Siri Text analysis AI systems reason over narrowly defined problems No learning capability and poor handling of uncertainty AI systems have nuanced classification and prediction capabilities No contextual capability and minimal reasoning ability Farfade, Saberian, and Li 2015 Image recognition Perceiving Perceiving Learning Abstracting Learning Abstracting AlphaGo Reasoning Reasoning

7 Key enablers of second-wave AI Better machine-learning algorithms Big data Neural Nets Statistical Models Deep Learning AOGs SVMs Bayesian Belief Nets CRFs Graphical Models SRL MLNs Markov Models HBNs Ensemble Methods Random Forests Decision Trees , ,91 0 4,400 44,000 exabytes Many applications and big markets Power-efficient processing 1000 KW for 1Bparameter deep learning network CPU GPU Patents

8 Today s artificial intelligence is powerful Operate highly autonomous platforms Search the deep web Manage cybersecurity in real time Overcome spectrum scarcity

9 but limited Construction worker in orange safety vest is working on road A young boy is holding a baseball bat Andrej Karpathy, Li Fei-Fei

10 Internet trolls cause the AI bot, Tay, to act offensively

11 Future third-wave AI technologies Handcrafted knowledge Engineers create sets of rules to represent knowledge in well defined domains AI systems reason over narrowly defined problems No learning capability and poor handling of uncertainty Perceiving Learning Abstracting Reasoning Statistical learning Engineers create statistical models for specific problem domains and train them on big data AI systems have nuanced classification and prediction capabilities No contextual capability and minimal reasoning ability Perceiving Learning Abstracting Reasoning Contextual adaptation Engineers create systems that construct explanatory models for classes of realworld phenomena AI systems learn and reason as they encounter new tasks and situations Natural communication among machines and people Perceiving Learning Abstracting Reasoning

12 Develop explainable AI New learning process This is a cat: It has fur, whiskers, and claws. It has this feature: Training data Explainable model Explanation interface I understand why/why not I know when it will succeed/fail

13 Uncover causal relationships among 1,000,000s of scientific observations Early result: An automatically generated web of influences from 1000 research papers

14 Some third-wave AI technologies Human-machine symbiosis Automatic whole-system causal models Continuous learning Embedded machine learning Explainable AI

15 R&D investments in foundational technology and applications Domain: Natural language processing Dragon (Nuance NaturallySpeaking) Speech Understanding Research (1971) Machine Reading (2005) Artificial Neural Nets (1991) Global Autonomous Language Exploitation (2005) Personalized Assistant that Learns (2005) USA CPoF, USSTRATCOM, NMCI DTRA, NASIC Broad Operational Language Translation (2012) GALE: Over 35 systems deployed worldwide Apple Siri IBM Watson XDATA (2011) Military Foreign language Translation System POR, 16 DoD organizations, FBI Low Resource Languages for Emergent Incidents (2015) Communicating with Computers (2015) Deep Exploration and Filtering of Text (2012) Memex (2013) Big Mechanism (2015) DTRA terrorism indicators Law Enforcement, ATF, FBI, NSA DNI, CIA, Treasury, FinCEN, FBI, DEA DARPA programs Direct commercial impact DoD transition Deep Learning (2010) Robust Automatic Transcription of Speech (2010) Facebook AI (LeCun) Baidu Research (Ng) AF Rivet Joint, SOCOM PUMA UAV, IC, JSOC Traveler, JIATF-S, Navy EP3

16