​ ​​ ​  

Artificial​ ​Intelligence:  Practice​ ​and  Implications​ ​for  Journalism

September​ ​2017 

  Mark​ ​Hansen*   Meritxell​ ​Roca-Sales**  Jon​ ​Keegan**  George​ ​King**    *​ ​Brown​ ​Institute​ ​for​ ​Media​ ​Innovation   **​ ​Tow​ ​Center​ ​for​ ​Digital​ ​Journalism         

Platforms​ ​and​ ​Publishers:​ ​Policy​ ​Exchange​ ​Forum​ ​I  June​ ​13,​ ​2017​ ​|​ ​Columbia​ ​Journalism​ ​School 

Organized​ ​by​ ​the​ ​Tow​ ​Center​ ​for​ ​Digital​ ​Journalism   and​ ​the​ ​Brown​ ​Institute​ ​for​ ​Media​ ​Innovation 

Artificial​ ​Intelligence:​ ​Practice​ ​and​ ​Implications​ ​for​ ​Journalism ​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​1 

Executive​ ​Summary 2 

Introduction 4 

Discussion​ ​I:​ ​Al​ ​in​ ​the​ ​Newsroom 7 

Case​ ​Studies:​ ​‘A​ ​Spectrum​ ​of​ ​Autonomy’ 8 

Data 9 

Challenges​ ​for​ ​Publishers:​ ​Large​ ​Newsrooms​ ​and​ ​Small 9 

Discussion​ ​II:​ ​Technology 10 

Automation​ ​and​ ​Personalization​ ​of​ ​Stories 10 

Commenting​ ​Systems​ ​and​ ​Audience​ ​Engagement 12 

Proprietary​ ​Versus​ ​Open​ ​Algorithms 13 

Challenges​ ​and​ ​Limitations 13 

Discussion​ ​III:​ ​Algorithms​ ​and​ ​Ethics 14 

Transparency​ ​and​ ​Accountability 14 

Editorial​ ​Decisions​ ​and​ ​Bias 15 

Ethical​ ​Use​ ​of​ ​Data 16 

Concluding​ ​Remarks 17 

    The​ ​Policy​ ​Exchange​ ​Forums​ ​are​ ​a​ ​critical​ ​component​ ​of​ ​the​ ​Tow​ ​Center’s​ ​Platforms​ ​and  Publishers​ ​research​ ​project.​ ​In​ ​these​ ​sessions,​ ​participants​ ​representing​ ​both​ ​the​ ​platforms​ ​and  publishing​ ​sides​ ​of​ ​the​ ​news​ ​industry​ ​can​ ​engage​ ​on​ ​issues​ ​related​ ​to​ ​the​ ​ethical​ ​and​ ​civic  values​ ​of​ ​journalism.​ ​The​ ​forum​ ​focuses​ ​on​ ​the​ ​relationships​ ​between​ ​technology,​ ​business,  journalism,​ ​and​ ​ethics,​ ​and​ ​brings​ ​together​ ​diverse​ ​stakeholders​ ​to​ ​discuss​ ​current​ ​issues​ ​and  surface​ ​potential​ ​new​ ​ones.    The​ ​project​ ​is​ ​underwritten​ ​by​ ​the​ ​John​ ​D.​ ​and​ ​Catherine​ ​T.​ ​MacArthur​ ​Foundation,​ ​with  additional​ ​support​ ​by​ ​the​ ​John​ ​S.​ ​and​ ​James​ ​L.​ ​Knight​ ​Foundation,​ ​the​ ​Foundation​ ​to​ ​Promote  Open​ ​Society,​ ​and​ ​The​ ​Abrams​ ​Foundation.         

Artificial​ ​Intelligence:​ ​Practice​ ​and​ ​Implications​ ​for​ ​Journalism ​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​2 

Executive​ ​Summary    The​ ​increasing​ ​presence​ ​of​ ​artificial​ ​intelligence​ ​and​ ​automated​ ​technology​ ​is​ ​changing  journalism.​ ​While​ ​the​ ​term​​ ​​artificial​ ​intelligence​ ​dates​ ​back​ ​to​ ​the​ ​1950s,​ ​and​ ​has​ ​since​ ​acquired  several​ ​meanings,​ ​there​ ​is​ ​a​ ​general​ ​consensus​ ​around​ ​the​ ​nature​ ​of​ ​AI​ ​as​ ​the​ ​theory​ ​and  development​ ​of​ ​computer​ ​systems​ ​able​ ​to​ ​perform​ ​tasks​ ​normally​ ​requiring​ ​human​ ​intelligence.  Since​ ​many​ ​of​ ​the​ ​AI​ ​tools​ ​journalists​ ​are​ ​now​ ​using​ ​come​ ​from​ ​other​ ​disciplines—computer  science,​ ​statistics,​ ​and​ ​engineering,​ ​for​ ​example—they​ ​tend​ ​to​ ​be​ ​general​ ​purpose.     Now​ ​that​ ​journalists​ ​are​ ​using​ ​AI​ ​in​ ​the​ ​newsroom,​ ​what​ ​must​ ​they​ ​know​ ​about​ ​these  technologies,​ ​and​ ​what​ ​must​ ​technologists​ ​know​ ​about​ ​journalistic​ ​standards​ ​when​ ​building  them?    On​ ​June​ ​13,​ ​2017,​ ​the​ ​Tow​ ​Center​ ​for​ ​Digital​ ​Journalism​ ​and​ ​the​ ​Brown​ ​Institute​ ​for​ ​Media  Innovation​ ​convened​ ​a​ ​policy​ ​exchange​ ​forum​ ​of​ ​technologists​ ​and​ ​journalists​ ​to​ ​consider​ ​how  artificial​ ​intelligence​ ​is​ ​impacting​ ​newsrooms​ ​and​ ​how​ ​it​ ​can​ ​be​ ​better​ ​adapted​ ​to​ ​the​ ​field​ ​of  journalism.​ ​The​ ​gathering​ ​explored​ ​questions​ ​like:​ ​How​ ​can​ ​journalists​ ​use​ ​AI​ ​to​ ​assist​ ​the  reporting​ ​process?​ ​Which​ ​newsroom​ ​roles​ ​might​ ​AI​ ​replace?​ ​What​ ​are​ ​some​ ​areas​ ​of​ ​AI​ ​that  news​ ​organizations​ ​have​ ​yet​ ​to​ ​capitalize​ ​on?​ ​Will​ ​AI​ ​eventually​ ​be​ ​a​ ​part​ ​of​ ​the​ ​presentation​ ​of  every​ ​news​ ​story?    Findings   

– AI​ ​tools​ ​can​ ​help​ ​journalists​ ​tell​ ​new​ ​kinds​ ​of​ ​stories​ ​that​ ​were​ ​previously​ ​too  resource-impractical​ ​or​ ​technically​ ​out​ ​of​ ​reach.​ ​While​ ​AI​ ​may​ ​transform​ ​the​ ​journalism  profession,​ ​it​ ​will​ ​enhance,​ ​rather​ ​than​ ​replace,​ ​journalists’​ ​work.​ ​In​ ​fact,​ ​for​ ​AI​ ​to​ ​be​ ​used  properly,​ ​it​ ​is​ ​essential​ ​that​ ​humans​ ​stay​ ​in​ ​the​ ​loop.  

– There​ ​is​ ​both​ ​a​ ​knowledge​ ​gap​ ​and​ ​communication​ ​gap​ ​between​ ​technologists​ ​designing  AI​ ​and​ ​journalists​ ​using​ ​it​ ​that​ ​may​ ​lead​ ​to​ ​journalistic​ ​malpractice. 

– Readers​ ​deserve​ ​to​ ​be​ ​given​ ​a​ ​transparent​ ​methodology​ ​of​ ​how​ ​AI​ ​tools​ ​were​ ​used​ ​to  perform​ ​an​ ​analysis,​ ​identify​ ​a​ ​pattern,​ ​or​ ​report​ ​a​ ​finding​ ​in​ ​a​ ​story. 

– While​ ​the​ ​intersection​ ​of​ ​AI​ ​and​ ​data​ ​offers​ ​new​ ​kinds​ ​of​ ​opportunities​ ​for​ ​reader  engagement,​ ​monetization,​ ​and​ ​news​ ​feed​ ​personalization,​ ​with​ ​this​ ​comes​ ​the​ ​challenge  of​ ​finding​ ​a​ ​balance​ ​between​ ​creating​ ​echo​ ​chambers​ ​and​ ​remaining​ ​committed​ ​to  journalism’s​ ​public​ ​service​ ​mission.  

– Ethical​ ​use​ ​and​ ​disclosure​ ​of​ ​data​ ​(how​ ​information​ ​from​ ​users​ ​is​ ​collected,​ ​stored,​ ​used,  analyzed,​ ​and​ ​shared)​ ​is​ ​a​ ​fundamental​ ​issue​ ​that​ ​journalists​ ​need​ ​to​ ​confront. 

Artificial​ ​Intelligence:​ ​Practice​ ​and​ ​Implications​ ​for​ ​Journalism ​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​3 

– The​ ​potential​ ​for​ ​AI​ ​to​ ​augment​ ​the​ ​work​ ​of​ ​the​ ​human​ ​data​ ​journalist​ ​holds​ ​great  promise,​ ​but​ ​open​ ​access​ ​to​ ​data​ ​remains​ ​a​ ​challenge.  

-​ ​​ ​​ ​​ ​Artificial​ ​intelligence​ ​is​ ​unpredictable;​ ​we​ ​don’t​ ​feel​ ​that​ ​confident​ ​predicting​ ​where​ ​the  biggest​ ​problems​ ​will​ ​crop​ ​up.​ ​Vigilance​ ​on​ ​the​ ​part​ ​of​ ​both​ ​technologists​ ​and​ ​journalists  is​ ​necessary​ ​to​ ​keep​ ​these​ ​systems​ ​in​ ​check. 

  Recommendations   

– Investment​ ​in​ ​training​ ​editors​ ​and​ ​reporters​ ​is​ ​crucial.​ ​As​ ​AI​ ​tools​ ​enter​ ​newsrooms,  journalists​ ​need​ ​to​ ​understand​ ​how​ ​to​ ​use​ ​new​ ​resources​ ​for​ ​storytelling—not​ ​only  ethically,​ ​but​ ​also​ ​efficiently. 

– Developing​ ​and​ ​promoting​ ​the​ ​use​ ​of​ ​shared​ ​guidelines​ ​among​ ​journalists​ ​and  technologists​ ​around​ ​ethical​ ​use​ ​of​ ​data​ ​and​ ​public​ ​disclosure​ ​of​ ​methodology​ ​is​ ​a​ ​must.  Existing​ ​AI​ ​tools,​ ​like​ ​chatbots​ ​and​ ​commenting​ ​systems,​ ​should​ ​be​ ​used​ ​as​ ​opportunities  for​ ​thinking​ ​about​ ​how​ ​to​ ​apply​ ​editorial​ ​values​ ​and​ ​standards​ ​to​ ​the​ ​early​ ​stages​ ​of​ ​new  journalistic-specific​ ​technology. 

– For​ ​custom-built​ ​AI,​ ​which​ ​is​ ​too​ ​expensive​ ​for​ ​smaller​ ​operations​ ​to​ ​afford,​ ​newsrooms  should​ ​consider​ ​investing​ ​time​ ​in​ ​partnerships​ ​with​ ​academic​ ​institutions. 

– There​ ​needs​ ​to​ ​be​ ​a​ ​concerted​ ​and​ ​continued​ ​effort​ ​to​ ​fight​ ​hidden​ ​bias​ ​in​ ​AI,​ ​often  unacknowledged​ ​but​ ​always​ ​present,​ ​since​ ​tools​ ​are​ ​programmed​ ​by​ ​humans.​ ​Journalists  must​ ​strive​ ​to​ ​insert​ ​transparency​ ​into​ ​their​ ​stories,​ ​noting​ ​in​ ​familiar​ ​and​ ​non-technical  terms​ ​how​ ​AI​ ​was​ ​used​ ​to​ ​help​ ​their​ ​reporting​ ​or​ ​production.   

Artificial​ ​Intelligence:​ ​Practice​ ​and​ ​Implications​ ​for​ ​Journalism ​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​4 

Introduction  By​ ​Mark​ ​Hansen,​​ ​​director​ ​of​ ​Columbia’s​ ​Brown​ ​Institute​ ​for​ ​Media​ ​Innovation    Our​ ​conversation​ ​at​ ​June’s​ ​forum​ ​began​ ​where​ ​these​ ​discussions​ ​often​ ​do:​ ​with​ ​the​ ​idea​ ​that​ ​we  can​ ​enhance​ ​human​ ​ability​ ​through​ ​computation.​ ​Our​ ​specific​ ​focus​ ​was​ ​on​ ​journalism​ ​and​ ​tasks  associated​ ​with​ ​reporting,​ ​writing,​ ​and​ ​designing​ ​impactful​ ​visualizations​ ​and​ ​other​ ​journalistic  “experiences.”      First​ ​and​ ​foremost,​ ​computation,​ ​as​ ​a​ ​tool,​ ​extends​ ​our​ ​ability​ ​to​ ​perform​ ​basic​ ​calculations—that’s  the​ ​old​ ​magic​ ​of​ ​spreadsheets​ ​and​ ​the​ ​success​ ​of​ ​computer-assisted​ ​reporting.​ ​But​ ​advances​ ​in  computation​ ​also​ ​bring​ ​the​ ​ability​ ​to​ ​recognize​ ​new​ ​data​ ​types,​ ​new​ ​digital​ ​objects​ ​that​ ​are​ ​open  to​ ​computational​ ​techniques​ ​of​ ​analysis.​ ​And​ ​with​ ​new​ ​data​ ​types​ ​come​ ​new​ ​kinds​ ​of​ ​questions  about​ ​the​ ​world​ ​around​ ​us.​ ​More​ ​and​ ​more​ ​of​ ​our​ ​world​ ​is​ ​being​ ​rendered​ ​in​ ​digital​ ​data,​ ​so​ ​that  (in​ ​journalistic​ ​terms)​ ​our​ ​data​ ​sources​ ​are​ ​becoming​ ​more​ ​diverse—and​ ​the​ ​information​ ​we​ ​can  draw​ ​from​ ​them,​ ​deeper​ ​and​ ​more​ ​interesting.​ ​It​ ​almost​ ​begs​ ​for​ ​a​ ​kind​ ​of​ ​aesthetic​ ​that​ ​prizes  new​ ​computational​ ​voices​ ​in​ ​the​ ​same​ ​way​ ​we​ ​value​ ​a​ ​new​ ​human​ ​source​ ​with​ ​a​ ​unique  perspective​ ​on​ ​a​ ​story.     To​ ​ground​ ​what​ ​we​ ​mean​ ​by​ ​“enhancing​ ​our​ ​abilities”​ ​and​ ​the​ ​shift​ ​to​ ​new​ ​data​ ​types,​ ​let’s  consider​ ​how​ ​standard​ ​journalistic​ ​practice​ ​has​ ​changed​ ​when​ ​it​ ​comes​ ​to​ ​wading​ ​through​ ​piles  of​ ​documents,​ ​perhaps​ ​returned​ ​by​ ​a​ ​FOIA​ ​request.​ ​With​ ​machine​ ​learning,​ ​we​ ​can​ ​pore​ ​over  thousands​ ​upon​ ​thousands​ ​of​ ​documents​ ​in​ ​a​ ​kind​ ​of​ ​mechanistic​ ​reading.​ ​“Reading”​ ​at​ ​this​ ​scale  was​ ​not​ ​possible​ ​a​ ​couple​ ​decades​ ​ago,​ ​not​ ​without​ ​a​ ​lot​ ​of​ ​human​ ​effort.​ ​Now,​ ​instead​ ​of​ ​taking  in​ ​text​ ​line-by-line​ ​and​ ​word-by-word—as​ ​you​ ​may​ ​now​ ​be​ ​doing​ ​with​ ​this​ ​text—machine​ ​learning,  or​ ​more​ ​specifically​ ​Natural​ ​Language​ ​Processing,​ ​helps​ ​us​ ​to​ ​create​ ​summaries​ ​of​ ​texts​ ​or  divides​ ​them​ ​into​ ​groups​ ​with​ ​common​ ​features​ ​(called​ ​clusters).     Italo​ ​Calvino​ ​provides​ ​a​ ​simplified​ ​view​ ​of​ ​this​ ​in​ ​​If​ ​on​ ​a​ ​Winter’s​ ​Night​ ​a​ ​Traveler​.​ ​A​ ​character  from​ ​the​ ​book​ ​named​ ​Ludmilla​ ​explains​ ​that​ ​she​ ​has​ ​a​ ​computer​ ​program​ ​that​ ​reduces​ ​a​ ​text​ ​to  individual​ ​words​ ​and​ ​their​ ​frequencies.​ ​From​ ​here,​ ​she​ ​can​ ​much​ ​more​ ​easily​ ​“read”:     

What​ ​is​ ​the​ ​reading​ ​of​ ​the​ ​text,​ ​in​ ​fact,​ ​except​ ​the​ ​recording​ ​of​ ​certain​ ​thematic​ ​re-occurrences,  certain​ ​insistences​ ​of​ ​forms​ ​and​ ​meanings?     In​ ​a​ ​novel​ ​of​ ​fifty​ ​to​ ​a​ ​hundred​ ​thousand​ ​words​ ​.​ ​.​ ​.​ ​I​ ​advise​ ​you​ ​to​ ​observe​ ​immediately​ ​the​ ​words  that​ ​are​ ​repeated​ ​about​ ​twenty​ ​times.​ ​Look​ ​here​ ​.​ ​.​ ​.  

   blood,​ ​cartridge​ ​belt,​ ​commander,​ ​do,​ ​have,​ ​immediately,​ ​it,​ ​life,​ ​seen,​ ​sentry,​ ​shots,​ ​spider,  teeth,​ ​together,​ ​you​ ​.​ ​.​ ​.  

Artificial​ ​Intelligence:​ ​Practice​ ​and​ ​Implications​ ​for​ ​Journalism ​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​​ ​5 

Don’t​ ​you​ ​already​ ​have​ ​a​ ​clear​ ​idea​ ​what​ ​it’s​ ​about?     With​ ​computation,​ ​we​ ​extend​ ​our​ ​abilities​ ​to​ ​“read”​ ​thousands​ ​or​ ​millions​ ​of​ ​documents.​ ​(Franco  Moretti​ ​at​ ​Stanford​ ​formalizes​ ​this​ ​difference,​ ​contrasting​ ​“distant,”​ ​or​ ​machine-mediated​ ​reading,  with​ ​“close,”​ ​or​ ​line-by-line,​ ​reading.)​ ​These​ ​new​ ​abilities,​ ​however,​ ​necessarily​ ​change​ ​how​ ​we  think​ ​about​ ​collections​ ​of​ ​documents​ ​and​ ​the​ ​knowledge​ ​we​ ​pull​ ​from​ ​them—our​ ​abilities​ ​extend,  but​ ​also​ ​our​ ​perspective​ ​changes.     As​ ​with​ ​text​ ​sources,​ ​digital​ ​images,​ ​audio,​ ​and​ ​video​ ​are​ ​also​ ​all​ ​now​ ​open​ ​to​ ​computation.​ ​In​ ​the  same​ ​way,​ ​ou