Analytics and AI: The Good, the Bad and the Ugly
2 likes•1,773 views
The document discusses the evolution and current state of artificial intelligence (AI), highlighting its potential benefits and significant challenges. While AI could revolutionize industries, enhance humans' quality of life, and create new job opportunities, there are concerns regarding bias, job displacement, and the ethical implications of AI ownership and control. The author emphasizes the need for proactive discussions about the future of work, emphasizing that the economic and political implications of AI are pressing issues that require attention.
Analytics and AI: The Good, the Bad and the Ugly
- 1. Srinath Perera VP Research, WSO2 [email protected] Analytics and AI: The Good, the Bad and the Ugly
- 2. Predictive Analytics 2 Machine learning • Given examples build a program that matches those examples • We call that program a “model” • Major improvements in last few years (e.g. deeplearning) Can you “Write a program to drive a Car?” 2
- 3. AI: History • Perception Tasks • Voice recognition ( Siri, Google Agent, Amazon Echo) • Language Understanding ( ?? ) • Video recognition ( ?? ) • Cognitive Tasks • 1999 IBM Deepblue beats Chess Champion Kasparov • 2011 IBM Watson beats human champion in the quiz show Jeopardy • 2016 Deepmind’s AlphaGo beats Go Champion Lee Sedol 3
- 4. AI: Status Quo • Hitting Human performance in voice, translation • Auto Driving ( Google, Tesla) • Chatbot Lawyer • AI generated Art 4
- 5. Good • AI may be the next revolution to propel human’s to next level • Replace manual Jobs • Age of abundance ( better food, cloths, health, education to all) • Universal Basic Income • We might be able to work only 4 hours a day 5 Compared to pre-industrial revolution, most of us eat better, have better cloths, live better and longer than the royalty in 18th Century
- 6. To understand, we need to pick a time • Short Term (5 years) - Concerns right now • Medium (10-20 years) - Concerns of wider application • Long term - Where does everything leads 6
- 7. Bad>Short Term: Bias 7 It is very hard to debug the models and know what it will do (e.g. Stock Market flash crashes) Example: Model to decide who to hire • Our models inherit our biases • Models will enhance current status quo. • Not enough data means low rank
- 8. Bad>Short Term: Other 0.01% • AI tells 99.99% of the time you will • Commit a crime • Does not pay back the loan • Does not do well in life • What if you are the 0.01% • Also 0.01% is pushed to be 99.99% • Filter bubble • Suggestive advertisements ( what if they show you more cigarettes because you are in the profile) • Idea that you are not in control ( Where is free will?) 8
- 9. Bad: Medium Term: Jobs • Autonomous diving will wipe out lot of Jobs • Amazon Go will wipe out more • Even professionals seem to be in trouble ( Lawyers, Doctors) • So are Artists 9 Is there anything useful left for us to do?
- 10. Bad: Medium Term>Possible Futures 10 Case 2: AI is owned by all (e.g. governments) • Use wealth to give a UBI • Focus on larger goals to give humans direction ( explore the universe) Case 3: We can out-law AI and try to contain it. Unlikely to work long term Case 1: What if there is N companies doing all AI, robots • How can other humans pay for services? • Resources will run out • A war will come sooner or later
- 11. Bad: Medium Term>There is Hope • Humans might start thinking beyond day to day running earth • Lot of work is interfacing: finding the data, asking questions • Human interactions, human made will have a premium (e.g. hand loom) 11 • Best work can be done by human + AI • Most decisions are made without full information => leadership • New Jobs ( e.g. AI phycologists)
- 12. Bad: Medium Term>Injustice of Status Quo • Same time Status quo is not great. Many do not have • Healthy food ( AI and automation might do it) • Best health care (AI can do it) • Best education (AI can do it) • Best legal support (AI can do it) 12 Is it fair to hold the status Quo and let them suffer?
- 13. 13 Evil AI: Why should I do what humans tell me? Parental AI: these humans do not know what is good for them, I will protect them from themselves! Ugly: AI Armageddon Unconcerned AI: I need to explore the universe and to do that I need all metals and energy in earth
- 14. Can’t we build a Safe AI? • Asimov’s three Laws • Keep power with humans • Problem is that it is very hard formulate the requirement • Safety of humans might means AI keep us in cages • Happiness of Humans might mean put us in our own matrix • Peace might mean wipe out everything 14 For more detail https:// www.ted.com/talks/ stuart_russell_how_ai_might_make _us_better_people
- 15. Possible Futures 15 AI and Humans will merge (e.g. Brain computer interfaces) We figure out a limited use of AI (e.g. Advisory role or a servant role such as Asimov three laws) We keep AI out AI takes over ( that will be the end)
- 16. Conclusion 16 We need to start think about Jobs changes and transitions However, it is a economic and political problem, not a technical problem!! Don’t panic yet, we are not even close to general AI or self aware AI With luck there is enough time before a threat is eminent Bias and other immediate problems are real
- 17. Questions?