AI in Journalism

Interviewer: Recording and as an introductory question first we would like to know if you could introduce yourself a little bit, talk later maybe about your, career maybe the responsibilities you have?

Expert: Yes, Katina Michael from the school for the future of innovation in society at Arizona State University and I'm also a joint hire in the school of computing, informatics and decision systems engineering. I've worked in telecommunications before my part in academia and also have been affiliated with the University of Wollongong for about 18 years in my main academic role. But in my research I look at society, policy and engineering and the interplay between all of these things. We don't put engineering first, we put the societal needs first but often if you reverse this paradigm and you say its engineering policy in society, often the engineering can lead to benefits but also costs and risks and potential harms which is why we are flipping this paradox and if I can say this challenge and saying it is more about society first, then policy, then engineering, in this order, rather than saying it begins with engineering and maybe there are policy and social implications. So we know for a fact that technology put him to solve one problem can create new problems and we know for a fact that with new emerging technologies there are also policy responses and societal impacts.

Interviewer: It's perfect, it's really matching our topic kind of.

Expert: Yes

Interviewer: And in general at first what do you think which values or ethical principles are particularly important in journalism?

Expert: Definitely in journalism having had experience in the past with maybe doing between 300 and 600 interviews. Definitely 300 published outcomes in journalism in print media, in online medium, in TV, in radio, I've had a lot of experience there. But also as investigatory work, a lot of journalists will ring me to ask me questions and through this experience I can talk quite confidently that trust is very important and so is evidence. So good report putting mechanisms ensure that facts are broadcast in some format to the wider public and made accessible and you know your trusted media sources. Usually it's the broadcasting Corporation, the incumbents, like in Australia it's the ABC the Australian broadcasting Corporation, again it's ABC in America. And you also have print media and there are also these, if I can call them fringe fringe media outlets that people may go to and navigate to. There may not be the mainstream interpretation of a problem or a societal fact but they do have in essence some important evidence that is not mentioned in the mainstream, either due to space or to advertising rights or other things. But then there is also now the rise of citizen journalism through social media. And citizen journalism really is not evidence based, it's very much opinion based. It's very much not about facts and it's very much about persuasion. Having said that I think there are many citizen reporting outlets that are very very important and are factual, but they may not agree with mainstream media. Behind mainstream media are magnates who have particular orientations in political orientations, philosophical orientations, and many times they are run by governments, depending on which markets you're talking about. So if you want freedom of the press, we usually have this expression in English, 'freedom of the press', who is behind the publication outlets, what kind of orientation do they have? Will they be for the government, against the government, will they be for the people, will there be for big business? You know, these are all very subtle things that the average person does not really understand but they're looking for comprehensive reporting, investigative reporting, and factual, but a lot of people, as one author in the Transactions in Technology and Society has said recently, well increasingly people don't care about facts, they want to feel secure in what's called fiction and sometimes, what is happening now, is we're seeing the publication and the reading of fiction in the guise of fact and people can't differentiate between the two. And the other thing that happens is, the fiction starts to spill so good that it becomes the default belief and so we have polarised societies as a result of the fact versus fiction approaches.

Interviewer: There are really interesting aspects. And according to that if we now put artificial intelligence to it, so have you ever experienced artificial intelligence in connection with journalism?

Expert: Yes, both as a reader and as an editor. And I just want to go back to one point, Mara. I think trust, trusting your sources means bidirectional trust between the key informants and the reporter is also very important. Sometimes the discussion is so sensitive, that anonymity is increasing the importance as the press is under great surveillance by state actors. And so when you say something as somebody who's reporting Turick (?) is a keen format to a story. You want to know that the journalist journalist who's asking you the questions is trustworthy. But also that behind them the network or the publication outlet is also trustworthy and will not take your words and change them to mean something different than what you said, that the reporters are technically competent. How often have I given interviews and I've requested to the Reporter 'just let me see what you've written when the story is close to ready, so I can tell you not to change anything but rather whether you have interpreted what I've said is correct' and often the technical capability of the Reporter is not there. They are trying to latch on to terms, they're not fluent in the terms but they realised it is important for the reporting. And so they can make a fundamental error like a technology constraint. And if you are acquainted as saying something and there is an error in the quotation or the innuendo or the suggestive nature of the quotation, then your academic reputation is put on the line. So to your next question with regards to AI, I often realise that we failed the Turing test; we don't know when content has been created by an AI, that is looking at factual so called factual sources but, also increasingly I'm seeing editing services using AI to clean up text that has been written either by authors, by editors or others. And some of the very innocent mistakes that I see often is gender representation. You know, the he and she. What do you do with transgender, for example. You know, they or their... and I'm seeing errors in subtle things like that when it comes to auto engines going through automated and editing. You know, once I asked one publisher, who I will not name, 'I've already gone through this twice, how come every time you give it back to me it has different errors' and of course they did not want to admit that the reason, there are errors, is because he has gone to a bot to actually cross checking. This is like frustrating to the human. How come a clean document all of a sudden has these injection errors, let's call him injection errors, in a sudden SQL injection area, but let me call them injection errors, where the publication that was clean keeps coming back with errors, that were never there to begin with, and so you are arguing the point, something has mucked up my document, not someone. Something. I remember as well another time we changed suppliers. And the suppliers did not tell the publication outlets that they were using an automated tool. This was around 2014, 2015, and we had given clean text and all of a sudden we're seeing these characters injected or spacing that's wrong. And it was like again every time it comes back to cross check the proof you're saying 'how come the document has new errors' and how come these areas which do not look like traditional formatting errors like a percent sign is injected into a paragraph for what reason. None. So, often when we read content we cannot distinguish between that which is bot driven and that which has been written by human. We can't, we've lost our capacity to do that, especially when it's small paragraphs of information. And on the other hand, when we are on the editing site we picked them up because there is an, what I called, non traditional errors that humans do not make. For example, if I gave a document to a sub editor, they would not go through putting percentage signs in every paragraph topic sentence beginning, right? It doesn't make sense. So you can detect, as an editor, AI mistakes faster than if you can detect as a reader, that's what.

Interviewer: You talked a little bit about the transparency of those AI written articles. What do you think, yeah... it sounds like you think that an AI generated article should be marked as so, both for the editors and both for the readers?

Expert: I think authorship is becoming increasingly important. You know, when people do historical criticism on documentation, my husband is a textual criticism expert, right, and he looks at manuscripts dating back to 2000 years old. You think to yourself 'authorship is such a big thing because it authenticates the document'. Okay. So if I know 'Katina Michael wrote the document', Katina Michael is accountable for what is written in the document. If the reported from ABC decides to make up some information, that has not been transmitted with trusted sources, then they should be held accountable. It should be transparent that that person has fabricated evidence. It's like when scientists fabricate results in experiments, right, the same thing. And so, yes, I think it should be marked. Do I think bots should not generate contents? That's not what I'm saying. Because I think in the future, although machines cannot really be held accountable, their owners can be held accountable. And they could be liable for, if I can say, fake news. And what I want to see is a block chain of evidence. What was the source of this quotation? What was the original point of the data beats? Whatever it was. So for example, there are many outlets that are like syndicated and they copy so they have the menu newspaper source and then they will take fragments of this main newspaper and select a subsection of the document and then they will syndicate to about 90 to 100 local other publications. Like in Australia, for example if an article appears in the Sydney Morning Herald. It will likely appear also in another outlet possibly, not a competing outlet. So let's just say the Age, the Australian and then it filters through to the rest of the media unit that owns the publication outlet. So if you have grabbed 2 lines from this, then you should allow the reader to track back and it should also allow the publisher to say, using the blockchain: 'The first primary also was... the person who conducted the interview was... and this was the date and time.' Now if I decide to create this information angles, and really, what is this information? It is like taking a cup of water and putting vinegar in it, just one drop of vinegar. It's still water but it's been mixed with vinegar. And what I'm arguing is: It is not now pure water, it is something else. I can still drink it but something tastes a little bit funny. Now with taste, our sensors in our body can detect the change in the purification of the water. But when we are reading something, we're not so good at this. Even when we are looking with our eyes in visual TV, it could be a snippet of multimedia and I say something and I say 'oh, how terrible was that' whatever the incident was. I cannot any longer trust. Is it real footage? Is it real people? Is it real sentences? All of these things can be like the vinegar in the water and all it takes is for one word to be changed in the new ones of that quotation has changed. But as a reader, you think 'oh this looks interesting, this sounds okay', but I'm fooled because I don't have the authentic trail of evidence, what we call in legal speak 'the chain of custody', right, 'the chain of command, the chain of custody'. When, for example, an enforcement officer takes DNA evidence from a somebody, right, like a hair follicle, and the evidence can only tempered with, it must remain pure in the right conditions, it must be placed so that it is not exposed to heat and other things, so that the reading of the hair follicle for the DNA is accurate. Right? But if I took someone else's hair and said 'here' and I mixed the evidence up, it's not a proper chain of evidence. There has been some temporary and increasingly we will not know, we will not perceive and we will not be able to detect with the human eyes and a brain, that this is not real, this is fake. And unfortunately, some people love fake news and it becomes the de facto news. This is their fact. This is their life board, this is the evidence expect. You can't convince me. Now see, I told you, I told you COVID-19 is about vaccinations and chip implants. This is one of the recent kinds of memes going around on the Internet. So is that true? Well, yes, they are saying vaccinations are important for overcoming COVID-19. Is there a pattern with chip implants? Yes, we can find that on the US government patent website. Does Bill Gates really want to do this? I don't think so. Right? So this is the kind of messaging that can filter through echo chambers in social media and convince readers without reading in most of the time now; people, because of the information overload, are not reading what they are re-sharing. 'Ah, the headline looks good, it came from a good person, press re-share'.

Interviewer: So do you think there are some ways to ensure that an article that is written by an Ai is perceived as trustworthy?

Expert: Again, yes. I think we could have trusted marks as one way. You know, we have secure websites with the HTTPS. Right? The SSL functionality functionality. Why not have the same kind of approach with journalism and data and, you know, if it's not trusted, that's okay, it doesn't have the trusted mark, but if the writer and the publisher can authenticate, at least did you know that the chain of evidence is accurate, the people who are quoted are real people, the headline, you know, has been verified, so to speak, and it began with maybe a mention in the New York Times or it began with a mention somewhere that is respected. Now having said that, it does not mean that even if we have verified news that this system is infallible because that's what people on the blockchain want to tell you; that the blockchain is secure, it will never be fallible, no one can breach security. Well, it is quite possible that key informants can be lying. And it is up to the investigative reporter, the human can do further fact finding to actually figure out whether it is true or false. But I don't know if the machine will worry about that additional layer of scrutiny. It doesn't have nuances, it doesn't have human faculties to basically figure out deception, although some of our bots are actually very good at mimicking deception and interpersonal deceptive behaviour. But we don't know, right? So, I want to say to you, it will be better than what we have now, but who is verifying the article? What is verifying the article? Is it again back to the lobbyists in the background who owned the media units and if we want to be very honest, how many magnets are there in the world, in the western world. Media moguls, we used to call them, media magnets. Well there's a handful that controlled the media, and so citizen journalists say 'I don't care if my work is not verified because my opinion counts' and this is the David and Goliath story that we have seen. You know, especially during covert, we have seen what has happened with the print media. It's gone out of business. Advertising cannot support it anymore. And so, when we start to look at the online format, or what is truth, what's fact? Who says it's fact? Who's history is this? Whose interpretation is this? And this is when it becomes entirely complex. And I envisage perhaps, but wars, right, AI bot wars. You know, my AI? Hm, I think I wanted to have a tinge of philosophy in there. The other AI? Hmm, I think I build my AI for political gain, so I get money, right? It's all very interesting but it will not be perhaps hard coded, but the AI will have internal bias, it has to. Someone has to create the code. So we can't say just because the machine generates the news article, it is not biased. We can't say that. You know, there could be hidden bias. They could be misunderstood by us (Bias?) or just purely accidental, you know? I'm putting in the rules to create nice articles, and edit it, but then, for example, I make a mistake and I don't address an under represented minority group in the right way. People living with disability, people in other circumstances. So, I think we will learn by doing and we will get better and better at the AI, but I also think we should give it a go. I don't think it's wrong to say that machine learning can help us generate media. But then there is also a lot of sensitivity. For example, if i put in a public space a sensor that collects the dialogue of the day. Is that news? Even though it was in a public space? Was the recording overt? Did the person who was in the shopping mall, realised that what they were saying about a particular brand, might end up being pointed in the newspaper? You know, this is the kind of thing we have to know our rights as key informants, as generalists, and the complex chain in the background which is now online contents. You know, Google is currently in trouble. Almost over in Australia, for example, Google and Facebook are being told 'you manipulate the small players in publication because they then have your stealth, you take their content, you reproduce it in an article on your online news and you don't pay for it'. You know, there is a value to content. And sometimes we perceive news to be free on the Internet, but in actual fact it is not free. You know, there is a value, information comes with the value. And who has the right to access the information is also the flipside of this. For example, if I am an individual, I don't have a subscription to a service, and then every time I want to go to a new space I have a firewall or a paywall that says 'pay to access', then I could be a little bit sad, you know. Maybe the machine learning will break open this model of paper view, to say 'this is trusted and it's free and it was generated by citizen science person'. It's very very interesting, the business model is changing very rapidly at the moment.

Interviewer: That's true. You mentioned some aspects like the fact checking, for example, and for example in this context, what do you think is the role of AI independence? Should there... like who is responsible, for example, for the content or should there always be like a person who is checking every article before it is published by an AI?

Expert: It's very interesting. I think the publisher is definitely liable and accountable. And the more transparent they are, the less accountable they may be. If they can prove through the blockchain that the error was not made by them, then they will not be held accountable, especially if it's one of them machine driven programmes that is taking from a trusted source, allegedly a trusted source, and then there is a problem. So the machine looks at and find something by crawling through the web or through some syndication. It takes the information an re-publishes the information with some, if I can say, reference back to the original source. And the block chain can do that. For example, if I take my mouse and I click over Wikipedia 3 lines and I put it in an essay that's published, I should be able on the electronic format to see that those 3 lines were taken from Wikipedia. I shouldn't be questioned about it. But if there is other evidence where I'm found to be at fault and it really it is about being at fault, then the publisher or someone else must be accountable. A machine can't be accountable. I can switch off the machine, the machine doesn't care, right? It is about the publisher. So publishers who are seeking to go into this kind of business must be ready to be held accountable for authorship, because inadvertently they are in control.

Interviewer: So also for example, if there are like mistakes, for example, published, then you would say there's also the publisher who's accountable for this?

Expert: Yes, definitely. But if they're looking at a particular authentic source, they perceived to be authentic it's been verified but the mistake hasn't happened with the publisher, then the source is actually in trouble. And so this will negate many of the court cases saying he said, she said, because there is a trail of evidence going back enough. I'll give you an example: Let's say when a child is young and it is learning how to do a long division and it doesn't realise, but it makes the error in the first line of the long division. And then the next person comes along and they carry the long division error right through to the answer, they will still be marked correct. But they will lose half a month because in the first line they made a mistake, right? When you are examined it's not the result that counts, it's the trail of evidence on how is the working out. And right now we have never held anyone accountable for working out. right? In the future we will say, if the error happened in the first line and a publisher carried the error through, I get back to the first line and I said the person who made this mistake, made the mistake, I didn't make the mistake. But I can't, for example, deliberately put the blame on somebody else, I just can't. I've gotta also somehow cross check. There has to be the least level of cross check. I don't know what that would look like, but at least a human in the loop, as we say, some human to intervene to crosscheck. But increasingly, with the amount of information we have on the Internet,, it becomes too hard to moderate every single one. But I think, if we want to go this way, we have to invest in this. And I'll give you another example: Right now people are using YouTube live and they're uploading content, Facebook live there uploading content. Some of this content is not fit for purpose. This content is discriminatory, racist, evil, breaks criminal laws, and should not be on the Internet. And until recently Facebook and Google were saying this is not our problem. Actually it is their problem. If you have a platform and you are accepting content, you are responsible for his publication. And what they've done is they've unfortunately brought in a lot of content moderators who are not paid too much money, they are working long hours, they are waiting for people to flag and report problems and they're troubleshooting. After the fact, yes they have AI algorithms to denote, for example, nakedness of children, so pedophilia or otherwise, but they're also really left behind, because you cannot possibly expect one AI programme or several or dozens to actually trawl through and find errors in, for example, authenticity. When we look at, for instance, when a deep fake of Nancy policy in America occurred, nobody knew at the beginning with a Nancy Pelosi had actually said what she said. Or it was a deep fake and it was a deep fake. You know, but which which bot in the world can detect, this is the real Katina Michael or it is a fake one, or it looks like Katina Michael and she's saying the wrong things, or this is an image of someone who has her name but actually this image is a fake image, it is not a real person on the earth surface. Good luck! What are we going to do, take the facial image of every person in the in the world, verify that and then have a constant database that keeps looking through and says 'ah this visual recording is Katina and it's not a deep fake of Cortina or it actually is a real person with the wrong name'. Like this is going to be entirely, entirely complex, riddled with errors and it will have a lot to do with identity schemes.

Interviewer: You mentioned before for example biases who can be for example in articles or maybe racism, those kind of things. What do you think if artificial intelligence for example has a database of former articles or something like this, who maybe include those biases, what do you think how can I handle this? Can it be ensured that the article, that the AI is generated, is nevertheless avoiding for example those biases or racism or aspects like that?

Expert: Basically there are different ways to implement AI algorithms. And so traditionally we've used historical articles as a data set to learn from and so we give the AI bunch of data, a lot of data, as much as we can, and we say 'we train you on this data set' and as you say, this is when we have biases come up. Secondly how do we overcome these biases? Well, as our ability to go towards, perhaps unsupervised learning algorithms, we may be able to have more authentic depictions of articles. But we know the case of Microsoft stay many years ago that was unleashed onto the Internet and within 24 hours it was making anti Semitic comments, praising narcissism and much more. So there is something to be said about training a particular analysis that you told me 'this is clean, this is good, this is positive' versus the opposite where it is not good and you don't want it to be left to its own devices. I think there's going to have to be a compromise between clean data sets, if I can put it like, that but we will never be able to delete conscious bias and bias embedded. We can't, no matter what we do. And so when people say to me 'humans have an imperative and machines do not', I often say 'but who built the machine? And what is the machine being taught? What are they learning, what's their learning material?' So we will never be able to escape bias in AI if anything it may amplify the bias and even unconscious bias, bias that we don't even see, depending on our life world. You know, a lot of people in the western world will say 'I have the perfect AI', but that AI, if placed in the context of China, will not look perfect. It may even look racist, but to the westerner they are espousing values and principles in the western world which may not necessarily gel with those principles in other markets. And so you will always have a clash of cultures, that's why culture is really paramount when it comes to technology.

Interviewer: Yes, these are interesting aspects. Also we look at the ethic aspects and ethic is also so difficult to define because in the different parts of the world and they have different understandings of ethics, so it is a little bit difficult.

Expert: Yes, I had one... one researcher once said to me 'oh, this is easy, we can solve this problem. We have one global ethics'. And I said to the person 'Good luck. I don't agree'. You know, markets will not agree.

Interviewer: But what do you think, what rules would an artificial intelligence have to follow when writing a journalistic content or are there some rules which have to be established?

Expert: So I think AIs do things that are brilliant like structure, like grammar, like having a point to the article, right? They can mimic human behaviour and good human writing. That's not a problem. In terms of other things that they may have to follow, I think knowing the fine line between reporting. And sometimes humans have this capacity, for example sensitivity. Let's just say a crime has occurred. Some media outlets will be more graphic with the crime, others will be more reserved. You know, we don't know if the machine is looking at police records or legal case president. We don't have to know everything that happened. Sometimes this is not good for the human psyche. And I think, knowing the limits of the reporting is important. Just like i said before, just because it can, it doesn't mean it should. Because then, even though social media people are self reporting what they eat for breakfast, when they're out at a cafe, pictures of their kids, you know? Is the machine going to do the same. But instead of giving the responsibility to the human to do this, person by person, what if I said to you we go into a coffee shop and because you are there and the machine has seen that most people you know will take a selfie and I'll take a picture of the food, imagine a machine, which was embedded in the sensors of the lighting, starts to take all these pictures and then, as you leave, it says 'well, guess what, I've just taken pictures of you with your family and your friends, what you ate, when you were laughing. Would you like me to publish this on social media?'. Most people would find that creepy. But the machine does not understand the difference between privacy in the context of surveillance and socialities and data. So in a utopian world, every single image, instance of what you did, would be taken and some of that would be news. You know, maybe the way to go and this is not the way to go, but the Chinese, for example, have got classrooms where they have a real time camera in the classroom and they invite members of the public to watch the class. And one school said as a response to a journalist 'we do this because when the students know that other people are looking at them, they are more in line. They feel like celebrities. They feel like this is reality TV.' Now, if a machine does not know the difference between the reality TV and the average person's life, which is poor, personal, private and their own, and then we have a lot of problems and the way that media might be generated. For example, imagine the paparazzi. We used to talk about, you know, when we looked at lady Diana's passing, you know, allegedly there were paparazzi that were taking photos of her with Alfie it ??? (38:09) in the area in France. And many people believe that the car was being chased and then it hit a wall and she died. But what if the paparazzi were machines, little tiny machines with pinhole cameras. And the next time Prince Charles went out out of his Palace, the next time a singer ??? went out of their room, the hotel. We don't need to pay people $40,000 for the photo. We get the photo because there's a surveillance camera there and even better, there's a mobile surveillance camera and while we're rolling the footage and filming and recording everything that happens, we can using biometrics, identify who was there with accuracy. 'Oh, it was this pop star, it was this politician, they did this.' So, really it is about total visibility. I have a fear that if we open up the reporting mechanisms with AI without limitations, then we are trespassing on peoples personal privacy.

Interviewer: So you would say like the limitation is to teach the AI to notice the space of the personal space for example, something like this, if it's possible?

Expert: You have to change the policy, you can't do anything with this data without consent. That's what you gonna do. So if they want to use my image from today at the supermarket, they have to tell me. Verified tick. You know, we got consent, we can do that.

Interviewer: So, transparency one more time?

Expert: Hm. (yes)

Interviewer: Okay. Do you think that in journalism or in the companies itself have to be conditions in order to use AI for writing articles, so are there any preconditions that have to be guaranteed?

Expert: Yes, I think there has to be internal policies. There also has to be a balance between what I would call the human worker, Co working with the robot in this instance our machine a piece of software. Machine learning, for example, AI. And so I don't envisage a future where Fairfax media or the magnates have a company, just with bots, and all the journalists in the flesh are out of work. They would possibly love this because it means they could keep generating news and making lots of money and pay nobody. The way I see the future of work in the space of journalism is when humans are co-working with machines and the labour which is arduous, for example editing nuance, and not nuance, forgive me, editing structure, keeping to 900 words, blah blah blah. All of the things that are procedural, that take a lot of time. We can reinvest that brain power into stories, into real important media. But I think the human has to be in the loop and so imagine a visit to a future where journalism is just bots, that would be a sad future I think.

Interviewer: And how did you imagine the use of AI in journalism optimally? So are there somethings that, would you say, if those are guaranteed then this is an optimal use of AI?

Expert: I think the optimal use would be on the blockchain, figuring out whether a piece of evidence that you are reusing and have not received primary source for, is true, that would be awesome. Being able to use the power of AI to authenticate the story. You've done everything you can, so you're running the machines in the background to make make sure that the data gathering process is foolproof. I think as well getting information in a timely manner to the web or through different mediums, for example, what happens if I have a textual story is based in text and I want a news reader to read out and to offer this content in multimedia. Well I'm not opposed to avatars. I'm not opposed to something other than a human reading the story in a very engaging way, so that I can hear it when I'm walking in the morning. So I think we have to play with text-to-speech, speech-to-video and think about if I want to show a story rather than having the video production guys in the background, trying to find or generate payroll ??? (43:08) or go half an hour in the car to try and find film that is relevant to the area. (technical issue) I can engage in AI to do that, so I have a location seen automatically, it's two milliseconds on the TV or on the Internet, but it means I don't have to get somebody to drive 30 minutes in traffic to get 2 milliseconds of footage. So in stitching the story together we can become very ingenious, but we also have to say 'we weren't there', so you have the story that is a real story from a human but then the reading of the story doesn't necessarily have to be human and the stitching together of background images doesn't have to be human driven. But we have to be honest, how many times do we see bloopers on ABC in Australia. It used to be a programme, 5 minutes only every week, and it would show the lies that would happen in the media. For example, someone claiming they were in Queensland and it writes on the screen in a live from Queensland and they would have to say 'well, it wasn't life and the Reporter was really somewhere else'. Well I think in the future we won't have to lie about things like that, you know. Media watch and the authenticity of media and so we can still be authentic using AI, it just depends how we use it.

Interviewer: And do you see AI more like in those supporting areas, like for example taking those pictures, taking those videos, or do you also think that there are some areas where you would prefer using AI to a human journalist? Or are there some some tasks or subject areas who is an AI maybe more able to solve maybe than humans journalists?

Expert: Yeah, I I don't know, I'd have to think about that more carefully. I'm sure there are. I'm sure there are a, but this needs a great amount of dialogue and discussion. For example, let's say factual things that don't change, you know, 2 + 2 equals 4, this is never going to change. So if things that are unchangeable reported on, for example shareholder prices. You know, this is not changeable. And we're already seeing a lot of innovation in terms of how reporting happens in the financial world. You know, on the spot reporting. And so, things of that nature which are factual, you know, the day's trading price doesn't change unless it changes. And what's already using their capability to talk about training and some people have a great advantage when I have bought sat there utility to show them trends and other people don't. So I think opening this up to more people empowers them. So anywhere where it's about empowering people, I would say 'Go ahead and use the bots'. But as we so often see, some bots will empower some and disempower others, is always advantages.

Interviewer: Yeah, always two sides.

Expert: Yes, always, no matter what, no matter what.

Interviewer: I have one final question: What do you think what other rules or important aspects will be important in the future or in the development of AI in journalism?

Expert: That's incredible, you know, this questions, all of these questions blow my mind but I'm seeing them happen in front of us. It's going to be increasingly complex to denote a story which is fake. In 2004, long time ago, we had a group of MIT students who were trying to demonstrate that some international conferences where money making conferences and we're not really peer reviewed. And so they created an article that was submitted for peer review, allegedly submitted for peer review, and the article was fake. They used terminology from here, dictionary from here, structures of language and structures of articles and somehow they penetrated the peer review process. Now once as a technical editor, somebody submitted something on swarm intelligence, but this was back in 2005/6. And I remember reading about the 2004 story and thinking 'This paper looks, smells, everything looks great about it. Template is is being used, everything is fantastic. But something is not right.' And so I do further cross checking, I went to the author website, to my eyes it looked fake but to others it might have been real. They had a connection to an Amazon book. The book was worth $1. I thought to myself 'This looks strange.' So they knew that somebody would try to authenticate by looking at the web but they were already one step ahead. They had a website, they had an Amazon profile, you know. That was a book costing a dollar or self publication means you can publish a book for a dollar and you can sell it for a dollar. And then the next thing I did, because I was corresponding via email, they never lost their cool. They provided me with the University telephone number to call. 'This is my supervisor.' I rang. I went to voicemail. I said 'Look, I'm investigating the authenticity of this author, could you ring back.' This professor never rang me back. We did not put the article through peer review but this wasted so much time. So if a bot and one... I remember in another case I wasn't sure about an article, I gave it to an expert in the field of decision-making, and the person says to me 'You know what, and you know my credentials', I said 'yes', 'I can't tell.' If an expert cannot tell the difference between a fake and a real written article by human, that puts us at a very vulnerable moment in history, you know.

Interviewer: That's true.

Expert: And the thing is, is it garbage, in garbage out, well it looks good I just read it, it comes back to the deep fakes, you know, people latching onto fiction because it makes them comfortable. It doesn't challenge their personal beliefs. So I think we're at a very interesting precipice and I want our journalism students to be aware of the pros and cons of AI. I'm not saying 'Don't use AI', but I think it has a great function for process. Not so much the actual storytelling in terms of the getting the cane forming evidence. Definitely for radius of content and you will come up with back end models, business models. For example, let's say I'm the publisher and it's the city Morning Herald and I get paid by the Sydney Morning Herald. I take the two lines. I own them because I did the primary evidence searching, but if a competitor wants to quote my 2 lines, that will pay a fraction of a cent. They didn't do the work. It was authenticated with the Sydney Morning Herald but they pay a fraction of the cost in reimbursement for access to quote those two lines. Now imagine that. It's a different cost model, it doesn't rely on advertising, it relies on goodwill and also acknowledging the effort of the other person who went to to the point of doing that. And if we look at this from a thesis perspective, people doing a PhD, it's becoming harder and harder to detect where the sources are coming from. But if you knew that a book that you quoted from 2006, the original source was 1977, you should be quitting the original source, not 2006. And this is where we're getting at, like historical data is still important in this in this new age.

Citation: Katina Michael with Mara Ortmann, 29 May 2020, “AI in Journalism”, https://www.katinamichael.com/interviews/2020/5/29/ai-in-journalism