Innovation report: Artificial Intelligence
1 like1,992 views
This document provides an overview of artificial intelligence including definitions, issues, and applications. It defines AI as the study of intelligent agents that can perceive their environment and take actions to maximize success. Some key issues discussed are predictive recommendation systems and development of smarter objects like home assistants. Applications highlighted include IBM's Watson for health and education, Google Photos for image processing, Tesla's Autopilot, and MIT's Deepmoji for understanding emotions.
Innovation report: Artificial Intelligence
- 1. Innovation Report: Artificial Intelligence Daniel Voignac • 08.2017
- 2. Overview I. Definition II. Issues III. Applications f
- 3. Motivation
- 4. Artificial Intelligence Deep Learning Self Driving cars Cognitive Computing Deep Neural Networks Pattern Recognition Machine Learning Diagnostic assistanceChatbots Intuition Algorithms Virtual assistant Machine Translation Recommendation systems Robots Search Engine Spam Detection Cancer detection Games Turing test
- 6. ● “The study and design of intelligent agents, where an intelligent agent is a system that perceives its environment and takes actions that maximize its chances of success.” Russel and Norvig ● “Just as the Industrial Revolution freed up a lot of humanity from physical drudgery, I think AI has the potential to free up humanity from a lot of the mental drudgery.” Andrew Ng
- 7. Definition Thinking humanly Thinking rationally “The exciting new effort to make computers think... machines with minds, in the full and literal sense.” (Haugeland, 1985) “The study of mental faculties through the use of computational models.” (Charniak and McDermott, 1985 Acting humanly Acting rationally “The study of how to make computers do things which, at the moment, people are better.” (Rich and Knight, 1991) “Computational Intelligence is the study of the design of intelligent agents.” (Poole et al., 1998)
- 8. Definition ● AI is a combination of: ○ Philosophy ○ Mathematics ○ Economics ○ Neuroscience ○ Psychology ○ Computer Engineering ○ Control theory and cybernetics ○ Linguistics
- 9. Robotic Process Automation (RPA) ● High-Volume repeatable tasks ● Different from IT automation because adaptable ● “Just as the Industrial Revolution freed up a lot of humanity from physical drudgery, I think AI has the potential to free up humanity from a lot of the mental drudgery.” Andrew Ng
- 10. Machine Learning ● Computer to act without programing ● Deep learning: automation of predictive analytics ● Three types: ○ Supervised learning ○ Unsupervised learning ○ Reinforcement learning ● Good read on current unefficiency of ML: https://techcrunch.com/2017/08/08/the-evolution-of-machine-learni ng ○ Some “old” AI algorithms are very good and suffice most applications (“linear/logistic regression, random forests and boosted decision trees”) ○ Deep Neural Networks is the real (“Getting deep learning to work is hard “)
- 11. Machine Learning ● Supervised learning ○ Sets are labeled → pattern detection → label new data sets ● Unsupervised learning ○ Unlabelled sets sortes according to = or ≠ ● Reinforcement learning ○ Markov decision process - good for processes that include long-term versus short-term reward trade-off (agent performs action(s) and is given feedback) - AlphaGo
- 13. Computer Vision ● Analog-to-digital signal processing ● Compares digital processing to database ● YOLO (You Only Look Once) open source real-time object detection ○ https://pjreddie.com/darknet/yolo/ ○ https://www.ted.com/talks/joseph_redmon_how_a_computer_le arns_to_recognize_objects_instantly/up-next ○ 20ms/image ● Google Image, self-driving cars ...
- 14. Natural Language Processing ● Processing of human language by a computer program ● Used very early in spam detection ● Current approach is based on machine learning ○ https://www.technologyreview.com/s/608382/to -build-a-smarter-chatbot-first-teach-it-a-second-l anguage/
- 15. Deep Neural Networks ● Algorithms designed after human brain ● Recognise patterns ● “Clustering and classification data on top of data you store and manage”1 ● Help group unclassified data ● Classify when given labelled data to train on ● “Deep” ⇔ stacked neural networks (multiple layers, more than one hidden layer) ○ Layers are made of nodes 1 source: https://deeplearning4j.org/neuralnet-overview#define (date of access 14/08/17))
- 16. Deep Neural Networks ● Deeper the net, more complex features can nodes recognize = feature hierarchy ● Google photos, Salesforce Einstein ● Three steps to neural networks ○ scoring input, calculating loss and applying an update to the model Starts with linear regression model for binary output but used logistic regression (sigmoïds, Hyperbolic tangents) 1 source: https://deeplearning4j.org/neuralnet-overview#define (date of access 14/08/17))
- 17. Deep Neural Networks ● 1957: Rosenblatt’s perceptron - 1st artificial neuron ● 1969: Minsky’s doubts ● 1989: LeCun proposes convoluted network ● 1996: Banks start using them for cheque reading ● 2011: Watson wins Jeopardy ● 2012: 15.3% error percentage ● 2016: Uber’s autonomous vehicles in Pittsburgh ● 2017: Alpha Go beats Ke Jie Source: Science & Vie 07/17
- 18. II. Issues
- 19. Prediction ● Predictive recommendation ● To simplify: ○ Consider all past events occurred at random ○ Group your data ○ Find a mathematical model that suits best ○ Iterate
- 20. Smarter objects ● Smart Home ○ Smart fridge ■ Samsung FamilyHub fridge (You can go have a look at it at Boulanger down the street) ○ Apple HomePod, Amazon Alexa, Google Home, Nest etc. ○ iRobot vacuum cleaner
- 21. Self-driving (cars?) ● Computer Vision ● Behavioural Prediction ● Influential companies include: ○ Waymo ○ Apple ○ Mobileye ○ Uber ○ Tesla Level Name 0 No Driving Automation 1 Driver Assistance 2 Partial Driving Automation 3 Conditional Driving Automation 4 High Driving Automation 5 Full Driving Automation
- 22. (Chat)bots ● Human agent interaction platform ● Can be included in ready existing chats ● Or embedded in a website ● Most common uses: ○ Personalised customer experience ○ Virtual Buying assistant ○ Included in a CRM ○ News ○ Productivity tools (slack bots)
- 23. Chatbots ● A few useful links: ○ Designing a chatbot (no coding) ■ https://chatfuel.com/ or https://botsify.com/ (also https://meya.ai/ ) ○ https://chatbotsmagazine.com/ ○ https://botlist.co/
- 25. Salesforce - Einstein ● Inbuilt AI module in the Salesforce CRM platform ● Combines and processes data from calendar, email, social, news to make predictions on the future ● Figures out competitor and recommends how to interact with new opportunities ● Writes email ● https://www.salesforce.com/products/einstein/f eatures/ See live demo
- 26. IBM - Deep Blue ● Chess playing computer ● Loses against Kasparov in 1996 ● Wins (Deeper Blue) against Kasparov in 1997 ● Kasparov’s latest book - Deep Thinking
- 27. Google DeepMind - Alpha GO ● Developed by DeepMind Technologies in 2010 ● Bought by Google in 2014 (628 M$) ● Beats World Champion Ki Jie
- 28. IBM - Watson ● https://www.ibm.com/watson/ ● First designed as a Question Answering computing system ● Designed to play the Jeopardy TV game ● Today applications include: ○ Health: cancer detection ○ Éducation: TJ Bot ○ Recipe generation ○ Concierge ○ Strategy ○ Weather forecast
- 29. IBM - Watson ● Salesforce Integration ○ “Integrate IBM Watson APIs into Salesforce to bring predictive insights from unstructured data inside or outside an enterprise, together with predictive insights from customer data delivered by Salesforce Einstein, enabling smarter, faster decisions across sales, service, marketing, commerce and more.” ● Hilton hotel Concierge robot
- 30. IBM - Watson ● Conversation ○ Conversation ○ Virtual Agent ● Vision ○ Visual Recognition ● Speech ○ Speech to text ○ Text to speech ● Empathy ○ Personality Insights ○ Tone Analyzer ● Discovery ○ NLP ○ Discovery News ○ Knowledge Studio Document Conversion ● Language ○ Translator ○ Naturall Language Classifier ○ Retrieve and Rank
- 31. Google - Photos ● Smart Classification ● Image Processing But also: ● Picture Enhancing - Street View ○ https://petapixel.com/2017/07/14/google-uses-ai-create-professional -photos-street-view-shots/ ● Image Generation ○ http://www.businessinsider.fr/us/these-trippy-images-show-how-googles-ai-sees-the-world-2015-6 /#the-engineers-found-that-the-ai-tended-to-populate-specific-features-with-the-same-object-for-e xample-horizons-tended-to-get-filled-with-towers-and-pagodas-androcks-and-trees-turn-into-buildi ngs-11110
- 32. Mobileye ● Not self driving car ● But device that makes any car smarter ● Even equipped a “self-pulling” suitcase ● Bought by Intel for 15.3 B$
- 33. Tesla - Autopilot ● Autopilot 2.0 - Level 4 ready ● Sadly famous accident involving white truck on bright weather ● Towards level 5 with rumors of Autopilot 2.5 https://electrek.co/2017/08/09/tesla-autopilot- 2-5-hardware-computer-autonomous-driving/
- 34. Google - Brain ● https://research.google.com/teams/brain/ ● Research team that sets its own research goals ● Vastly focused on “Making machines intelligent”
- 35. OpenAI ● “OpenAI is a non-profit AI research company, discovering and enacting the path to safe artificial general intelligence.” ● Recently announced developping an agent capable of beating world class champions at Dota 2 (e-game)
- 36. Google - AI ● https://ai.google/ ● Google assistant ● Cloud TPU (tensor processing unit) ● Platform to play with AI ○ https://aiexperiments.withgoogle.com/ ○ https://quickdraw.withgoogle.com/ ○ https://aiexperiments.withgoogle.com/auto draw
- 37. Tensor Flow ● https://www.tensorflow.org/ ● Developed by Google Brain teams ● Open Source ML library ● Helps deploying Neural Networks
- 38. MIT Media Lab - Deepmoji ● Understanding emotions with the use of emoji ● Countering e-bullying ● Monitoring social media ● Facebook announced 9000 job opening in social monitoring early 2017
- 39. Other AI research teams ● https://labs.pinterest.com/projects/ ● Uber Michelangelo
- 40. Appendices
- 41. References and useful links
- 42. References ● Patrick Winston. 6.034 Artificial Intelligence. Fall 2010. Massachusetts Institute of Technology: MIT OpenCourseWare, https://ocw.mit.edu. License: Creative Commons BY-NC-SA. ● S. Russell, P.Norvig, Artificial Intelligence: A Modern Approach, Third Edition, 2010, Pearson Education, New Jersey
- 43. Useful links - Top 5 ● https://www.coursera.org/learn/machine-learning ● https://deeplearning4j.org/neuralnet-overview#define ● https://chatbotsmagazine.com/chatbots-the-beginners-guide- 618e72599b55