媒体测量中的人工智能机遇

Opportunities forArtificial IntelligenceIn MediaMeasurement November 2024 Introduction3 About CIMM4 About the Author The Building Blocks of AI 6 Best Fit Applications for AI and ML 10 Using AI for Media Measurement17 Putting AI to Work Challenges to Adoption 27 Conclusions30 CONTENTS Introduction There has been much excitement aboutthe application of AI to many fieldsover the last 18 months, and mediameasurement is no exception. In this paper,I have consulted across the industry andbeyond to understand the current useof AI technologies, both planned and inproductiontoday. I have found many exciting and innovativeapplications that are likely to continue todrive the evolution of media measurement.This paper will provide a practical,in-depth introduction to the array of AItechnologies for media researchers andan appraisal of the current state of theart, likely future innovations, and risks toconsider when deploying AI in your mediameasurementsolutions. – Tom Weiss, author AboutCIMM The Coalition for Innovative MediaMeasurement is a non-partisan,pan-industry subsidiary of the AdvertisingResearch Foundation, focused oncultivating and supporting improvements,best practices and innovations inmeasurement and currency, datacollaboration and enablement, and theuse of new metrics and approaches tounderstanding the value of media. CIMMembraces the entire media and advertisingecosystem and prioritizes effectivecollaboration to deliver meaningfulchange. To find out more,visitwww.cimm-us.orgor contactinfro@cimm-us.org. About the Author Tom Weiss is a technologist, data scientist, and serialentrepreneur with over 30 years of experience acrossmedia and technology. Currently serving as ChiefTechnology Officer at Run3TV and Board Member at MX8Inc., Tom has been at the forefront of AI-driven consumerresearch and smart broadcasting innovations. He hasfounded and led multiple global technology ventures,developed industry-leading data science teams, andholds several patents in media and advertising. With anMA in Physics from Oxford University, Tom blends deeptechnical expertise with a passion for helping brandsbetter understand and engage with their audiences. Youcan follow him on substack at www.thegeniehouse.com •JustinEvansfrom SamsungAds•KeithSmithfrom MSI•KenWilburfromUC San Diego Rady School ofManagement•KevinKohnfrom Kinetiq•MatthewMcGranaghanfrom the University ofDelaware•MichaelVinsonof Comscore•NickNorthfrom the BBC•PaulDonatofrom the ARF•YanLiufrom TVision Tom would like to thank the following who have allprovided input and inspiration for thispaper: •AndyBrownfrom the AttentionCouncil•AndyPrincepfrom MarketCast•AshwinNavinfrom Samba•BrianWestfrom NBCU•ChrisWilsonof Hyphametrics•EmilyMcReynoldsof Adobe•JimBisbeefrom VanderbiltUniversity•JohnBrauerfrom Comcast•JonWattsof CIMM The views and opinions expressed in this report aresolely those of the author, Tom Weiss, and do notnecessarily reflect the views or positions of any otherindividual or entity. This report is intended solelyfor informational purposes. Neither CIMM nor theauthors make any representations as to the accuracyor completeness of any information contained in thisreport or in any report or web site linked to in this report.Neither CIMM nor the authors will be liable for anyerrors or omissions in this information or for any losses,injuries, or damages incurred from the display or use ofthisinformation. TheBuildingBlocksofAI The history of AI dates to the mid-20thcentury. The concept of “artificial intelligence”was first coined in 1956 by John McCarthyat the Dartmouth Conference, markingthe birth of AI as an academic discipline.Key milestones include the developmentof early programming languages, creatingthe first neural network in the 1940s, andmachine learning in the 1980s. The 1990sand 2000s saw significant advancements,with the development of more sophisticatedalgorithms and increased computationalpower. Noteworthy new stories that capturedthe popular imagination include IBM’sDeep Blue defeating chess champion GarryKasparov in 1997, the introduction of deeplearning in the 2000s, and the victory of AIsystems in complex games like Jeopardy!and Go. More recently, AI from Google’sDeepMind has made breakthroughs in severalcomputationally expensive areas of scienceearning its creator a Nobel Prize in Chemistry,and OpenAI’s ChatGPT has wowed millionswith its human-like responses. As of the end of 2024, we now have AIs withnatural language processing abilities thatcan pass the bar exam in the USA and solveMaths and Physics problems at a comparablelevel to undergraduate students. We cancreate realistic videos based on a singleimage or a text description of the video andedit photos and videos simply by asking theAI what to do. None of these seemed feasible at thebeginning of 2021. Their development hasdriven the creation of “Transformer”-basedmodels, which were first postulated in 2017.They have massively increased the volume ofdata w