Machine Intelligence Landscape 2016
Download as PPTX, PDF
3 likes1,155 views
This document summarizes the machine intelligence landscape in 2016. It notes that compared to 2014 and 2015, there was a greater focus on companies trying to understand machine intelligence applications. It also discusses the growing enterprise opportunity around using agents and bots to support business functions like scheduling, customer service, and content management. Additionally, it examines challenges in developing and deploying machine intelligence models at scale, and how this is leading companies to rethink their technology stacks and processes for managing intelligent systems.
Machine Intelligence Landscape 2016
- 1. MACHINE INTELLIGENCE LANDSCAPE 2016 @SHIVON @JAMES CHAM NOVEMBER 8, 2016 // SAN FRANCISCO
- 2. 2014
- 4. 2015
- 6. 2016
- 8. WHATS CHANGED 8 2014 5 Groups 35 Categories 233 Companies Focused on academics and big tech companies 2015 8 Groups 34 Categories 265 Companies Focused on startups and new autonomous systems 2016 8 Groups 38 Categories 318 Companies Focused on companies trying to understand machine intelligence
- 11. GO 11
- 18. HOW THEY FAIL 18
- 20. CYBORGS SYNTHESIZERS COORDINATORS TRANSACTORS COMPANIONS DIAGNOSTICIANS AGENTS
- 21. THE ENTEPRISE OPPORTUNITY: SUPPORT STAFF FOR EVERYONE 21 > Schedulers: Clara and x.ai > Handlers: Google Now, Siri, and Cortana > Coordinators: Howdy, Standup Bot, Tatsu, and Geekbot > Notetakers: Gridspace Sift, Evernote and Pogo > Copy Editors: Textio and IBMs Watson Tone Analyzer.
- 23. THE NEW STACK 23 Questions from big companies about Machine Intelligence lead to a new stackand the real risk of AI
- 24. CODE DATA MI IS DIFFERENT FROM SOFTWARE
- 25. CODE DATA DATA DATALOTS of DATA THE CURRENT STATE OF IT
- 26. CODE DATAMODELS ADDING MI TO IT
- 27. CODE DATAMODELS Generated by code using data Different from traditional SW More flexible, too Harder (impossible?) to grok Harder (impossible?) to verify When to trust? LEADING TO NEW QUESTIONS
- 28. CODE DATAMODELS MODELS MODELSLOTS of MODELS Iterates faster than code Buy or build Can improve quickly or subtly go very, very wrong Where should you deploy? AND NEW ISSUES
- 29. MICRO-ECONOMIC MODELS FOR MACHINE INTELLIGENCE 29
- 30. WHAT HAPPENS WHEN COST OF PREDICTIONS GO DOWN? 30 Analysis from Agrawal, Gans, Goldfarb > (bit.ly/economics-of-AI) > Idiot savantsunderstand the limitations of models > What happened with MC(arithmetic) 0 > What happens when MC(prediction) 0 > Think: substitutes, complements, new business models
- 31. BUILDING THE MI STACK 31 > Not the same as SW devadd models to your IT inventory > Instead of systems of record, records of predictions > Requires new answerswhen do you trust the model? > New change managementfor organizations and as models change > Look at examples in other industries
- 32. INDUSTRIES 32
- 33. NEVER NEVER LAND
- 34. PROMINENT ACQUISITIONS 34 Nervana Intel Magic Pony Twitter Turi Apple Metamind Salesforce Otto Uber Cruise GM SalesPredict eBay Viv Samsung Strong leaders Broad technology platforms Scarce, valuable talent Often $100M+ But also a reminder of difficulty of building a independent startup
- 36. MACHINE INTELLIGENCE CAN SOLVE NEW PROBLEMS 36
- 38. MACHINE INTELLIGENCE LANDSCAPE 2016 @SHIVON @JAMES CHAM NOVEMBER 8, 2016 // SAN FRANCISCO
Editor's Notes
- Stats 1.0 Groups: 5 Categories: 35 Companies: 233 2.0 Groups: 8 Categories: 34 Companies: 265 3.0 Groups: 8 Categories: 38 Companies: 318 1.0 --> 2.0 = 14% more companies year over year 2.0 --> 3.0 = 20% more companies year over year 1.0 --> 3.0 = 36% more companies since year 1
- Not something new, but a pretty old idea that needs to be called out Rather than talking about algorithms or bots or engines, I suggest we call out modelscode that is generated by other code by adding in data