Karen Hao is the Senior AI Reporter at MIT Technology Review, where she demystifies AI and explores its complex ethics and social impact. Previously, she was a tech reporter and data scientist at Quartz, and before that, an application engineer at the first startup to spin out of Google X. Follow Karen on Twitter @_KarenHao.
Karen studied mechanical engineering at MIT and worked as an application engineer in Silicon Valley before becoming a reporter. Enticed by the particularly far-reaching impact of AI on society, Karen turned to journalism to explore how the technology shapes everyday lives and in her writing aims to uncover its long-term implications.
Karen points out that tech media serves as the bridge between technologists and the public, and AI reporters have the important task of translating technical concepts for the average reader. With this in mind, accuracy in AI reporting is of utmost importance.
With the current hype around AI, however, more and more reporters are pursuing the beat, and some writers are under pressure to produce content on the topic before they fully understand it. This can lead to inaccuracies, misinformation and general misunderstanding.
This is complicated, too, by the emergence of seemingly credible, but inconsistently informed, public figures who claim authority over AI topics. For example, Karen says Elon Musk makes statements about AI that seem accurate to the everyday reader, when researchers in the field would never actually turn to him for direction.
The news cycle was already fast. Right now, it’s even faster. But Karen says this quickened pace hasn’t changed her primary goals as an AI reporter: to educate the public on how really AI works and how it’s shaping our lives.
Karen has always been focused on trying to understand artificial intelligence as deeply and accurately as possible, so she can best evaluate the importance of developments and news in the space. When she started out as a reporter, she went deep into AI research and even spoke directly with some researchers about their work. Karen leans on this base of knowledge when evaluating the numerous PR pitches she receives daily—notes from companies that hope she will write about their AI initiatives and products—and today, she uses that knowledge to evaluate the many pitches she gets about AI in the context of coronavirus.
Karen frequently comes across the assumption that AI offers a comprehensive solution to a specific problem—this, she points out, is impossible. In the context of the coronavirus, for example, she says AI will not “solve” the pandemic and we can’t expect it to, because the problems at the heart of the severity of the pandemic aren’t technology problems.
Karen mentions that the biggest issue that we’re seeing right now in the US, at least, is the lack of testing and the severe personal protective equipment (PPE) shortages. These issues can’t necessarily be solved by AI. Sure, you could use AI to optimize the manufacturing of PPE, but that’s not really the crux of the problem with the PPE shortage.
Another example: Karen believes that AI could play a role in getting vaccines produced, as researchers increasingly apply machine learning to rapidly identify vaccine candidates. But this would only be one piece in a far more complex solution. The labs that develop this work would also need to get funding and there would need to be an entire pipeline of production and distribution—logistics that are not ultimately about AI itself.
With readers hyper-focused on the coronavirus crisis, reporters are working to produce content that addresses relevant questions and concerns, but they don’t want to burn readers out with coverage of one single topic (and a grim one at that).
Karen notes that as much as journalists respond to reader interest, they shape reader interest, too. Today, news organizations are trying to establish a balance between pandemic and non-pandemic content—it’s something the entire industry is grappling with.
Jim Freeze Hi! And welcome. I’m Jim Freeze and this is The ConversAItion, a podcast airing viewpoints on the impact of artificial intelligence on business and society.
We’re excited to be back for a second season and look forward to sharing episodes on everything from how robots can adhere to social norms, to how AI is redefining the idea of a smart home.
Today, I’m reporting to you from my home office in Boston’s Seaport District. I’m one of the tens of millions of people working remotely during the coronavirus pandemic, and I’ve set up an at-home recording studio to keep The ConversAItion going.
The ConversAItion is presented by Interactions, a conversational AI company that builds intelligent virtual assistants capable of human level communication and understanding. In this episode, we’ll explore the important role of the media in educating the public about AI. In particular, we’ll unpack how our guest, Karen Hao, a senior AI reporter at MIT’s Technology Review, is approaching her coverage of AI’s role in the coronavirus pandemic.
Karen, welcome to The ConversAItion.
Karen Hao Thank you so much for having me.
Jim Freeze Yeah, this discussion is very timely, obviously. I think it might be interesting just to start off with a little bit about your background. My understanding is you used to work on the technical side of AI as an application engineer. What inspired you to make the transition or the pivot to media?
Karen Hao Yeah, so I actually was a mechanical engineer by training, so that’s what I did my undergrad degree in, and when I first joined the workforce, application engineer was a term for the startup that I was working on where we worked with clients to develop different apps for architecture design. I’ve pivoted many times in my very short career.
But when I was in Silicon Valley and I was working at this startup, one thing that frustrated me about the experience is I really wanted to be engaging in longer term problems and longer term thinking. But I wasn’t really finding an environment to do that in a startup, because with a startup you’re always moving so quickly and you rise and fall based on what your VCs want, how much funding you get and short term gains.
So, at some point I decided to pivot into journalism to start engaging more in thinking about how we build technology and how that technology then shapes our lives and society. And AI is one of the perfect technologies for doing this because it’s so expansive and covers so many different industries, so many aspects of our life. So it’s a really exciting lens through which to examine all the ways that technology is shaping the world.
Jim Freeze That’s terrific. I’m a big fan of pivots. I did the same thing as an undergrad and in grad school I did my undergrad master’s degree in mathematics and then I also have a law degree and am a licensed attorney, yet I’m a marketing exec.
Karen Hao That’s awesome.
Jim Freeze Yeah, I’m a marketing executive so go figure. I’m a big fan of pivots. So I listened to another podcast you were on recently where you talked about the role of the media in shaping the public’s perception of AI. Obviously, the media has or is a primary source of information about the technology, but you know like all media, has potential to generate some misinformation, as well. I’m curious to hear about your perspective on the responsibility of the media to educate and inform people about AI and its impact.
Karen Hao I think there’s a huge responsibility because I think that we are the middle layer between the technologists and the lay person. Because technologists, they’re really focused on doing their research and collaborating and innovating so they don’t necessarily have the time or the skill sets to keep a pulse on the public’s misconceptions of their work. So that’s really, I think, the job of journalists is to keep a pulse on the questions that people have, the misconceptions that they have, and be able to then answer those in ways that are understandable and accurate.
I do think that the media also struggles sometimes to cover AI in a constructive way because I think AI is such a sexy topic right now that there are so many journalists who start covering it or are forced to cover it, essentially, without necessarily having the background or the time to really deeply learn the subject material. So it can be challenging to be under pressure, be under deadline, trying to write a story about a technology that you only loosely understand and that’s sort of how errors can get introduced.
Jim Freeze Yeah. I completely relate to that based on certainly some of the stories I read. And it’s not only sexy, it also can be controversial and that sometimes lends itself to interesting angles on stories which have potential to be somewhat misleading or confusing to people, as well.
Karen Hao Yeah. And to be fair, I think the other challenge is that there are a lot of people who seem like big experts that say things that are kind of, if you’re in the field, are kind of off the mark. Like Elon Musk, he says a lot of things about AI that appear to the public to be a very authoritative opinion or authoritative source of information, when most people in the field would never look to him for trends about where the technology is going or what they should be focusing on. So, it can also be hard when you’re not embedded in those discussions to really filter out who are the authoritative sources and who are people that are kind of just talking.
Jim Freeze Actually that kind of really rolls right into my next question. I think it’s probably a very interesting time to be an AI reporter. Can you talk a little bit about how you’re approaching reporting in this current environment? Are there unique challenges? Are your goals different now than when you’re normally covering AI?
Karen Hao No, I don’t think my goals have changed. I think before the pandemic, the news cycle was already pretty fast. The pandemic has made it even faster. And it’s a noisy environment, before and after. It’s just, I think the pandemic has kind of amplified and exacerbated what used to be the case.
When I first started on this beat, the thing that was really important to me was first getting a really grounded understanding in the fundamentals of the technology. So I spent a lot of time reading research papers every week, not only reading research papers, but also getting in touch with the researchers who were writing them and talking to them about their work so that I could get a really clear understanding of the different concepts and also where this field is going.
So I guess I sort of see it as you pick pillars of your coverage. So, there were different concepts that became my pillars that I would work on understanding really well. And then I would use that to guide how I cut through the hype when I received pitches from companies or when I read other papers that were not necessarily written by well-established researchers.
And because of that foundation, when I cover things related to AI applied to the coronavirus pandemic, it’s just drawing from that base of knowledge to examine. Anytime I receive—I receive so many pitches in my email every day now where every company somehow is doing something to produce content related to coronavirus and you’ll see something like, “Hey, we made this computer vision system and trained it on all these CT scans of patient’s lungs, so now we can diagnose coronavirus.”
So what I do is I then think, “Okay, well what do I know about computer vision? If they’re using a deep learning technique, then they need to have a lot of images. So, how many images are they using to train? Oh, they’re only using a thousand images? That doesn’t seem right to me.” So, either follow up with the company and be like, “Hey, I have these questions. Explain to me how you’re able to get away with a system trained on only a thousand images,” or just delete the email. So it is about always going back to what is the core capabilities of the technology and does what I’m hearing about AI doing X, Y, Z thing actually align with that.
Jim Freeze Based on your recent conversations in this environment and some of those pitches you’ve been getting, what have you found the most surprising or misunderstood elements of AI are as it relates to the current situation, the coronavirus outbreak?
Karen Hao I think the biggest thing that’s misunderstood, and this is always the case, is people act like AI is the solution to everything. And techno solutionism is like a particular pet peeve of mine because anytime someone says, “Oh, X technology is the solution to this problem,” they cannot be correct. Like problems are so much more complicated, so much more nuanced and it requires many technologies, requires policy, requires social change. It requires a lot of different actors working together to create something that is comprehensive enough to actually mitigate a problem.
And it’s the same thing with the pandemic. Even if we had hundreds of thousands of images of CT scans that were training these amazing computer vision systems, it’s not going to solve the pandemic.
Jim Freeze No.
Karen Hao So, and I don’t think that practitioners necessarily believe this, although I do think they fall into a trap sometimes where they get overly excited about a small solution and make it seem like a much bigger solution than it actually is. But I do think that also the media plays a role into this, because I see a lot of headlines often that kind of take the language of the company trying to push their technology and end up writing the story that says, “AI is a tool we need to fight coronavirus,” or something dramatic like that. So yeah, I think that’s the biggest one.
Jim Freeze Yeah. And actually, you’ve kind of noted in some of your writing that AI has yet to prove its impact in the current pandemic. Why do you think that is? Are there shortcomings of AI or is there just an unrealistic expectation or a misunderstanding? I mean, what is it because you’re correct that it hasn’t yet proven its impact.
Karen Hao Well, I think that it’s because of the nature of the problem for the pandemic. So the biggest issue that we’re seeing right now in the US, at least, is the lack of testing and the lack of, or the severe personal protective equipment shortages, PPE shortages. And those are not things that can necessarily be solved by AI. Maybe you could say, “Oh, we can use AI to optimize the manufacturing of PPE,” but that’s not really the crux of the problem with the PPE shortage. It’s that we currently have a kind of a crisis of leadership at the federal level that’s not really aligning all of the different actors in our country to come together and fix this problem. And you could say, “Oh, maybe AI can help us maintain social distancing by having surveillance cameras look at whether or not we’re six feet apart at all times.” But that’s again, not really the solution for these problems.
So I think that’s kind of the reason why AI hasn’t really shown its use yet because the problems that are the heart of the severity of the pandemic are not technology problems.
The one thing that I do think AI could have a bigger role in is in getting vaccines produced. Because one of the ways that the pandemic can end is if we actually find vaccines for the novel coronavirus and there has been some work with researchers to use machine learning to try and more rapidly identify vaccine candidates so that is a way that it could accelerate that process. But again, it would not be the only solution that needs to happen. These labs that are developing this work also would need to get funding and they would need an entire pipeline of production and distribution and all of these things that are not ultimately about AI itself.
Jim Freeze Yeah, that’s just a really good point. Also, kind of in the current environment, thinking back to an article that you published recently talking about the trade-off between public health and privacy during the coronavirus. Consumer privacy is always a hot topic. Can you talk a little bit more about the biggest challenges AI researchers in the public health space are up against right now?
Karen Hao Yeah. I think there was this Guardian op-ed that came out like just a couple of days ago where an AI researcher was arguing that the number one issue with vaccine development is pharma companies aren’t giving up all of their data. And it’s not necessarily a new thing that AI researchers have been saying. I think we will perpetually hear some say as a community we want more data to do the work that we need to do. I do think that one of the things that, how we can actually manage this trade-off between privacy and public health is there are a lot of techniques that have emerged in the last few years that help preserve privacy, like privacy preserving machine learning techniques, where you don’t necessarily actually have to get the data into a centralized server hosted on Google Cloud or Amazon Cloud or whatever to actually train machine learning algorithms on that data. And so there have been some proposals from the AI research community about how to use these types of privacy preserving techniques to actually do the work that they need to do so that it no longer becomes a trade-off between privacy and public health.
Jim Freeze Oh, that would be ideal to get to the point where there isn’t a trade-off. You get the best of both worlds.
One last question for you. Obviously, whenever you have a situation like what we’re in, where one story seems to just dominate the news cycle and certainly with the pandemic, it’s center stage, there’s just no way around it. That being said, there definitely are other newsworthy topics out there. How are you balancing coverage of the outbreak with other interesting AI stories that you’re coming across?
Karen Hao Yeah, it’s really challenging. This is actually something that I think all journalists are really struggling with right now because, well, part of it is personal because when you’re writing so many coronavirus stories, it’s pretty toxic for your mental health. But then at the same time, it is the story that people clearly want to be reading. And in the first couple of weeks when this started hitting the US, I would propose other stories to my editor because I have to produce a newsletter every week about AI and I needed to fill it with content. And he always worried that no one would read the story because everyone’s mind is so focused and so attentive on this huge other looming crisis. And I don’t know. I haven’t actually figured out how to balance coronavirus stories and AI stories.
Jim Freeze I guess maybe it comes down to trying to figure out and forecast when people are just so mentally drained by just reading and hearing about nothing other than the coronavirus that they just need a break and they’re interested in kind of focusing on other things. I think that’s probably a very difficult balance.
Karen Hao Yeah. I also think that the news, like journalists, not only do we respond to reader interest but we also shape reader interest. So, I’ve kind of taken a bit of the approach of maybe people aren’t reading these stories yet, but I’m just going to keep covering them because if I didn’t then there wouldn’t be any non-coronavirus stories to read. But I think every journalist and every news organization is really trying to figure out that balance. You don’t want to burn out readers with perpetual coverage of a single thing. But it is also the most important, somewhat existential threat to us right now, and we need to be getting really important authoritative journalism out there about it.
Jim Freeze Absolutely. Well, thank you for the work you do. There’s probably not a more important time to have people who truly understand and have the background to comprehend artificial intelligence, be the same ones who are actually writing about it. So, thank you very much. This has been terrific and I really appreciate you being on The ConversAItion.
Karen Hao Thank you so much for having me.
Jim Freeze On the next episode of The ConversAItion, we’ll unpack the ways in which AI can amplify human intelligence to make predictions. We’ll be joined by Dr. Louis Rosenberg, the founder and CEO of Unanimous AI, a company that combines the power of AI with real-time human knowledge, wisdom, insights and intuition.
This episode of The ConversAItion was produced by Interactions, a Boston-area conversational AI company.
Well, that’s it for today, folks. I’m Jim Freeze, and we’ll see you next time.