Somo AI Breakfast Briefing
6 likes1,830 views
- Alexa - Siri - Cortana Personal Finance: - Digit - Cleo Health: - Anthropic - Babylon Travel: - TripAdvisor - Expedia Retail: - Amazon Go - Anthropic Education: - Duolingo - Coursera Finance: - Wealthfront - Betterment Transportation: - Uber - Lyft Manufacturing: - Anthropic - Rethink Robotics Energy: - Opower - Anthropic Agriculture: - Indigo Ag - Granular Government
![Confidential and copyright of Somo Custom Ltd. October 16 7
We are seeing exponential growth where we are used to linear growth
We are in a period of change
"I think if we can effectively merge with AI by improving the neural link
between your cortex and your digital extension of yourself […] then effectively
you become an AI-human symbiote, and if that then is widespread, and
anyone who wants it can have it, then we solve the control problem as well. We
don't have to worry about some evil dictator AI because we are the AI
collectively. That seems like the best outcome I can think of.”
Elon Musk - CEO, and CTO of SpaceX; co-founder, CEO, and product
architect of Tesla Motors](https://image.slidesharecdn.com/aiworkshopfinal-oct20-161020150357/85/Somo-AI-Breakfast-Briefing-7-320.jpg)
Somo AI Breakfast Briefing
- 1. Artificial Intelligence Workshop 20th October
- 2. Confidential and copyright of Somo Custom Ltd. October 16 2 Somo accelerates mobile transformation through rapid innovation to create products and experiences your customers and employees will love.
- 4. Confidential and copyright of Somo Custom Ltd. October 16 4 Agenda 1. The relationship between man and machine Demo #1 – Alexa and Allo 2. A look under the bonnet 3. Machine Learning & Deep Learning Demo #2 – Image recognition with Cortexica 4. Applications of Artificial Intelligence by Industry Demo #3 – Pepper the robot with Emotion Robotics Coffee Break
- 5. The relationship between man and machine There is a change happening between the relationship between man and machine
- 6. Confidential and copyright of Somo Custom Ltd. October 16 6 We are seeing exponential growth where we are used to linear growth We are in a period of change
- 7. Confidential and copyright of Somo Custom Ltd. October 16 7 We are seeing exponential growth where we are used to linear growth We are in a period of change "I think if we can effectively merge with AI by improving the neural link between your cortex and your digital extension of yourself […] then effectively you become an AI-human symbiote, and if that then is widespread, and anyone who wants it can have it, then we solve the control problem as well. We don't have to worry about some evil dictator AI because we are the AI collectively. That seems like the best outcome I can think of.” Elon Musk - CEO, and CTO of SpaceX; co-founder, CEO, and product architect of Tesla Motors
- 8. Confidential and copyright of Somo Custom Ltd. October 16 8 We are in a period of change "The development of full artificial intelligence could spell the end of the human race." Stephen Hawkings
- 9. Confidential and copyright of Somo Custom Ltd. October 16 9 We are in a period of change "Ten years from now, it will have even more significant wealth creation, stockholder appreciation opportunities, than what I believe we saw in social networks in 2005 and internet investing in 1995.” Jim Breyer
- 10. Confidential and copyright of Somo Custom Ltd. October 16 10 AI software automation and robotics will fundamentally transform business models Businesses will need to embrace disruption Businesses that don’t leverage the technology will likely find it impossible to compete General motors 840,000 workers $11 billion in earnings Google 38,000 workers $14 billion in earnings AI = different business models • Increased efficiency • Growth through creativity AI solves high value recurring problems • Decrease in resource vs output • Increased efficiency = increased ouput 1979 2012
- 11. Confidential and copyright of Somo Custom Ltd. October 16 11 AI can be broken down into 3 main groups AI is a broad term Artificial Superintelligence (ASI) • An intellect that is much smarter than the best human brains in practically every field Artificial General Intelligence (AGI) • Refers to a computer that is as smart as a human across the board Artificial Narrow Intelligence (ANI) • Specialises in one area • Completes and solves one problem
- 12. Confidential and copyright of Somo Custom Ltd. October 16 12 AI is already all around us, with some key future trends How will it effect us in the near future? • Smarter Robots • Faster Analysis • More Natural Interactions • Hotter Competition • IOT – explosion of connected devices • Proactive assistant become the norm • Expansion of Deep learning moving more towards machine learning
- 13. Confidential and copyright of Somo Custom Ltd. October 16 13 Who are the main players? We have seen intense acquisition activity in AI market 2011-2016 Source: CB Insights
- 14. Confidential and copyright of Somo Custom Ltd. October 16 14 An open source software library for machine learning TensorFlow Google autonomous cars reached a total of 2.7 mn km Google launches Google Assistant in all Android phones and Google Home Facebook launches the Moments app based on facial recognition Apple launches Siri in iOS AI is experiencing similar tech advancements and cost reduction curves as we have seen on PCs and smartphones Rapid productisation of AI Amazon launches Alexa Facebook opens Messenger App to bots 2011 2015 2016 Facebook launches automatic facial recognition for photos 2010 2012 Google Now is launched on Android 2014 Tesla Motors announced its first version of AutoPilot on Model S cars Google launches Face Unlock in Android WeChat launches its bot platform 2013 2mn 500mnPeople who used an AI-powered service
- 15. Confidential and copyright of Somo Custom Ltd. October 16 15 Roles/Tasks for which using AI According to smartphone users WW Why do you use an AI assistant US AI assistant users Frequency of use for voice controlled personal assistant US mobile phone users Source: Ericson ConsumerLab, 2016 Source: AYTM Market Research, June 2016 Source: 451 Research March 2016 How has AI re-set customer expectations New behaviours require a change in product and services 73% 64% 57% 44% 43% 41% 35% Search Engine Travel guide Personal Assistant Teacher Medical advisor Financial advisor Companion to talk 64% 40% 25% 23% 17% 8% Asking general questions Getting directions while driving Making calls Dictating texts and emails Checking updates Other Don’t have opinion, 33% Have opinion but never tried, 17% Tried once or twice, 13% Monthly, 10% Weekly, 14% Daily, 13%
- 16. Confidential and copyright of Somo Custom Ltd. October 16 16 With changing customer expectations we are seeing a surge of group use of AI How is AI responding to customer expectations? Virtual Personal Assistant Automation Chatbots
- 17. Confidential and copyright of Somo Custom Ltd. October 16 17 Amazon Alexa The ecommerce giant is using the Echo speaker and the personal assistant Alexa to push the Prime service into all homes and create a bridge online- offline Google Assistant Google Assistant is now available on Google Home and on Pixel phones. Google was the last to launch its personal assistant but has significantly more data about consumer behaviours compared to its rivals Apple Siri Apple was the first to launch its voice assistant Siri integrated in iOS to provide a faster access to apps and features. In 2016 Siri has been extended to Car and TV OS and has been opened to third party apps Big tech firms have packaged AI as a Personal Assistant
- 18. Alexa and Allo Demo #1
- 19. Confidential and copyright of Somo Custom Ltd. October 16 19 Alexa, do you use AI?
- 20. A look under the bonnet
- 21. Confidential and copyright of Somo Custom Ltd. October 16 21 A system that can sense, analyse, learn and act Sense and Feel Learn and Act • Natural language processing • Sensors processing • Audio recognition • Image recognition • Gesture recognition • Machine learning • Deep learning • Expert systems • Decision support systems • Robotics • Contextual Automation
- 22. Confidential and copyright of Somo Custom Ltd. October 16 22 How are AI and Deep Learning related? Deep Learning Deep Learning breakthroughs drive AI boom Machine Learning Machine Learning begins to flourish Artificial Intelligence Early AI stirs excitement 1950’ 1960’ 1970’ 1980’ 1990’ 2000’ 2010’
- 23. Machine Learning & Deep Learning A very shallow introduction @Andrew_Wyld
- 24. Confidential and copyright of Somo Custom Ltd. October 16 24 What is deep learning? Deep learning gets a lot of headlines for super cool applications: • Image recognition • Speech recognition • Language processing It’s called “deep” because it’s a simple technique applied multiple times. • Each layer takes input from the last • Each layer gets feedback from the next • Layers can model more and more complex concepts as they get deeper
- 25. Confidential and copyright of Somo Custom Ltd. October 16 25 How do we get to that? We will define a “shallow” learning technique using several inputs to predict one output. Then we need to work out a way to apply it repeatedly. Since each layer is trained by comparing it with “correct” output, we need to find a way to get the training into the hidden layers. We will combine several of these into a layer. 1 32
- 26. Confidential and copyright of Somo Custom Ltd. October 16 26 Training a classifier How we get this: Classifiers are trained using what’s called maximum likelihood estimation. We find a form for the classifier that produces results closest to the ones we expect.
- 27. Confidential and copyright of Somo Custom Ltd. October 16 27 Training a classifier (1) A classifier at its simplest takes input data, combines it and provides a yes/ no prediction. Each input positively or negatively affects the prediction. In training, the positive and negative effects are summed and compared with a known value. 1 2 Inputs are strengthened or weakened or even reversed depending on their effect on the prediction. This is done till the prediction is correct. 3 prediction true value
- 28. Confidential and copyright of Somo Custom Ltd. October 16 28 Training a classifier (2) In practice, there will be a large number of different sets of input and output predictions. They will not tend to agree in their predictions all the time. Here, a direct relationship between input and output is present except in one case. The ideal mapping would be wrong here in some cases.
- 29. Confidential and copyright of Somo Custom Ltd. October 16 29 Training a classifier (3) For this reason, the weight of inputs that don’t predict the outputs well will be reduced. Even then, there will be some errors (not pictured, because I’m awesome). The aim is to reduce the error as much as possible while still having a model that generalises well to new data.
- 30. Confidential and copyright of Somo Custom Ltd. October 16 30 Deep learning Hugely powerful. Nobody knows what’s inside it. Yet. The “deep” in deep learning refers to the fact that several classifiers are stacked, one in front of the other. Each one can learn more sophisticated things by building on the previous layer. Nobody really knows what goes on in the middle layers (although we’re beginning to research it).
- 31. Confidential and copyright of Somo Custom Ltd. October 16 31 Stack high the classifiers! A neural network is just a sequence of classifiers in a stack. Each layer can use the output of the previous layer as input; thus, by the end, features can be very sophisticated, based on complex combinations of other, simpler features. The hidden layers make the technique powerful but inscrutable.
- 32. Confidential and copyright of Somo Custom Ltd. October 16 32 Back propagation Just as in simple classifiers, the last layer is trained by comparing predictions with reality. However, the inputs to the last layer are then scored on how “good” their predictions were. This score is used to train the next layer in … … and so on.
- 33. Confidential and copyright of Somo Custom Ltd. October 16 33 Transfer learning The early layers of (for example) a cat recognition system will probably pick up general image features— corners, colour transitions, diagonals and so on—that would be useful for any image recognition task. If you want to make a dog recogniser but don’t have a lot of data, you could simply cut off the last layer, steal these early features, keep them the same, and glue a simple classifier on the end in place of the old last layer. This works surprisingly well.
- 34. Confidential and copyright of Somo Custom Ltd. October 16 34 Cool stuff: a very non-exhaustive list Stanford machine learning course https://www.coursera.org/learn/machine-learning/ did it, loved it. University of Washington machine learning specialisation https://www.coursera.org/specializations/machine-learning/ doing it now. Tensorflow online neural network http://playground.tensorflow.org/ have fun! Google/Udacity deep learning course https://www.udacity.com/course/deep-learning--ud730 want to do it!
- 36. How your business can benefit from Artificial Intelligence
- 37. Confidential and copyright of Somo Custom Ltd. October 16 37 Actionable insights and new value creation Insights Usability Gap Deep learning! AI bridges the insights usability gap The data value chain Advanced analytics and data visualisation Exponential growth of data • Smartphone penetration and usage • Social Media • Internet of Things • Digital payments • Big Data • Dashboards • Balanced scorecards • Infographics • Data Visualisation • Machine Learning • Deep Learning • Artificial Intelligence • Personal assistant (Alexa, Google assistant) • Proactive and contextual recommendation engines (Google now) • Decisions in the moment • Tactical automation
- 38. Confidential and copyright of Somo Custom Ltd. October 16 38 AI allows companies to push customer value expectations and deliver a compelling customer experience AI can be used to re- invent the way you deliver traditional services and products AI helps leaders constantly seeking ways to reduce overhead costs and streamline business processes 3 ways to realize business value Reinventing business models Creating a compelling customer experience Optimising operations and efficiency
- 39. Confidential and copyright of Somo Custom Ltd. October 16 39 Creating a compelling customer experience easyJet is using Artificial Intelligence to analyse all different aspects of the 1.5bn hours flown each year to help identify ways to improve the customer experience, predict customer needs and avoid flight delays 1.5bn hours of flight BMWi Genius is a customer assistant chatbot which responds to questions on all BMW models 3,500 product features
- 40. Confidential and copyright of Somo Custom Ltd. October 16 40 Source: Bloomberg technology Optimize operations and efficiency Google cut its electricity bill by 15% using DeepMind- powered AI. Saving a few percentage points of electricity usage means major financial gains for companies like Google In Sweden, Swedbank launched a AI web assistant called Nina. In 2016 it achieved an average of 30,000 conversations per month and first-contact resolution of 78% in its first three months 30,000 conversations/month 15% energy saving
- 41. Confidential and copyright of Somo Custom Ltd. October 16 41 Reinventing business models AI-powered autononomous cars could be deployed by 2020 or sooner. Driverless cars are expected to completely disrutp the automotive industry radically changing the business models from ownership to mobility service 2020 or sooner The raise of chatbot banking and AI money-managing assistants is expected to change financial assistance business models. Companies like Digit and Cleo are taking advantage of conversational interface to deliver financial assistance $4 bn
- 42. Confidential and copyright of Somo Custom Ltd. October 16 42 Image recognition with Cortexica Demo #3
- 43. Applications of Artificial Intelligence by Industry
- 44. Confidential and copyright of Somo Custom Ltd. October 16 44 Almost USD 1 billion has been invested in AI startups since 2013 New deal activity by industry 2012 2013 2014 2015 2016 (Q2) Financial Services Health Advertising, Sales and Marketing Travel / Transports IoT / Wearables Education Retail / Ecommerce Source: CB Insights
- 45. Confidential and copyright of Somo Custom Ltd. October 16 45 Applications of AI by Industry Personal Assistants Customer Assistants Automated Robots Algorithms Deep Learn. Financial Services Health Advertising, Sales and Marketing Travel / Transports IoT / Wearables Education Retail / Ecommerce
- 46. Confidential and copyright of Somo Custom Ltd. October 16 46 Health > Personal Assistant • Training apps struggle to provide useful recommendations • UA integrated IBMs Watson to create a Cognitive Coaching System • Launched January 2016 • 10m+ downloads on iOS + Android to date Under Armour cognitive training assistant
- 47. Confidential and copyright of Somo Custom Ltd. October 16 47 Health > Algorithms and Deep Learning • Launched DeepMind Health division to work with NHS • Research how AI can support NHS to improve performance and become more cost efficient • One of the first projects announced is Streams, an app that detects cases of acute kidney injury • Will develop open and interoperable technology while protecting patient data confidentiality DeepMind to tackle clinician and hospital issues
- 48. Confidential and copyright of Somo Custom Ltd. October 16 48 Transports > Autonomous Vehicles • Launched in Pittsburgh, US on September 2016 • Fleet includes 15 vehicles • Cars still have human drivers behind the wheel “Developing an autonomous vehicle is basically existential for us” Travis Kalanick, Uber CEO Uber launched first US self-driving taxis
- 49. Confidential and copyright of Somo Custom Ltd. October 16 49 Logistics > Autonomous Vehicles • Civil Aviation Authority (CAA) has granted Amazon special permission to test its aerial vehicles in UK • Amazon is also implementing AI in warehouses spawning an ecosystem where supply chains link themselves with each other enabling seamless flow of products and information from one end to the other • According to Accenture, 85% of organisations have adopted or will adopt digital technologies in their supply chain within 1 year Amazon moving toward a nimble supply chain
- 50. Confidential and copyright of Somo Custom Ltd. October 16 50 Travel > Personal Assistant • AI-based travel concierge • Inspiration-based - Lets customers ask natural language questions • Recommendation engine based on IBM Watson • Connects interests with places, offers, social, cultural and economic data to provide targeted travel suggestions and offers • White label, Widget or APIs Wayblazer AI powered travel concierge
- 51. Confidential and copyright of Somo Custom Ltd. October 16 51 Finacial Services > Customer Assistant • Assistant for banking customers • Based on IBM Watson • Started as an internal support tool, now provides answers to end customer's questions • Performs simple banking tasks like money transfers RBS Luvo now responds to customer queries
- 52. Confidential and copyright of Somo Custom Ltd. October 16 52 Financial Services > Customer Assistant Spixii transforms insurance sales • Guides users through the process of purchasing insurance based on their needs • Helps compare options and features • Conversational interface • Context based notifications • Also promotes offers and pushes upselling opportunities
- 53. Confidential and copyright of Somo Custom Ltd. October 16 53 Sales and Marketing > Customer Assistant • Developed in collaboration with chatbot company Allia • The car does the talking, with a tone of voice and personality • It automates the initial conversations with a potential client • Direct, emotional connection with the product AI brings Jeep Renegade to life
- 54. Confidential and copyright of Somo Custom Ltd. October 16 54 Sales and Marketing > Algorithms and Deep Learning • Campaign was made up of 2 components: • A video ad with people sent on blind dates based on matching taste profiles • A ‘flavour profiler’ branded website • Based on the IBM Watson cognitive engine • The profiler uses flavour profiles to suggest tailor made recipes • 7 million views on YouTube in three days Unilever Knorr “Love at first taste” AI campaign
- 55. Confidential and copyright of Somo Custom Ltd. October 16 55 Sales and Marketing > Algorithms and Deep Learning • Mariana identifies new target customers on social media platforms based on a comprehensive segmentation methodology based on AI algorithms and machine learning • The AI engine uses 50K data points from social, web and proprietary sources to define personas and exact customer archetypes • It also supports conversion, automating contents and communication. The deep learning engine crafts suggested topic areas based on each persona’s interests AI-enhanced social targeting
- 56. Confidential and copyright of Somo Custom Ltd. October 16 56 Sales and Marketing > Algorithms and Deep Learning • Predictive lead scoring representing the likelihood each lead will convert into an opportunity • Forecasting sales and spending • Create predictive journeys • Proactive recommendations about what communication or product each lead may want • Sales and marketing automation Salesforce integrated an AI platform in its CRM
- 57. Confidential and copyright of Somo Custom Ltd. October 16 57 What next RobotsEmotions Zero UI Creativity Autonomous Vehicles Sense and Feel Learn and Act Smarter Assistants
- 58. Pepper the robot with Emotion Robotics Demo #3
- 59. Q&A