What does human flourishing have to do with human-machine teams? And how do we meaningfully engage stakeholders in consultations about some of the most challenging problems of our time? Listen in as we explore some of these questions with Kate Devitt, co-founder and CEO of BetterBeliefs – a platform for evidence-based stakeholder engagement and decision-making – who also happens to be an internationally recognized leader in ethical robotics, autonomous systems and AI.
Listen and subscribe on Apple Podcasts, iHeartRadio, PocketCasts, Spotify and more. Five star ratings and positive reviews on Apple Podcasts help us get the word out, so please, if you enjoy this episode, please share with others and consider leaving us a rating!
Credits:
Guest: Kate Devitt
Hosts: Zena Assaad and Liz Williams
Producers: Zena Assaad, Liz Williams, Martin Franklin (East Coast Studios)
Theme music: Coma-Media
Content notes:
We have chosen to list this episode as explicit because of some discussion of warfare.
Transcript:
Liz:
Hi everyone, I’m Liz Williams.
Zena:
And I’m Zena Assaad.
And this is season two of the Algorithmic Futures Podcast.
Liz:
Join us as we talk to technology creators, regulators and dreamers from around the world to learn how complex technologies may shape our environment and societies in the years to come.
…
Zena:
In this episode, we’re joined by Kate Devitt. Kate is an internationally recognised leader in ethical robotics, autonomous systems and AI. She has held many leadership positions over the years across organisations such as the Queensland University of Technology, Defence, Science and Technology group, and the University of Queensland. Her most recent role is Chief Scientist of Trusted Autonomous Systems, which is Australia’s first Defence Cooperative Research Centre. Kate is also the co-founder and CEO of BetterBeliefs – a platform for evidence-based stakeholder engagement and decision-making.
Liz:
We talk to Kate about her thoughts on human flourishing in the context of artificial intelligence and human-machine teams. We explore the importance of considering the implications of AI and social media on human well-being, knowledge, and communication, as well as the need to engage diverse stakeholders in the development and regulation of emerging technologies.
Zena:
Kate stresses the importance of incorporating human values into the design and development of technologies to ensure human flourishing. She also highlights the role of democratic participation in technology development, as well as the need for new regulatory bodies to bridge the gap between policy development and implementation in the industry.
Liz:
We are so pleased to have Kate join us on what is an incredibly thought provoking and enlightening episode. We hope you enjoy our conversation as much as we did.
Zena:
Hello, Kate. Thank you so much for joining us today.
Kate:
Great to be here. Thank you for inviting me.
Zena:
How have you been?
Kate:
Yeah, great. Winter is coming in Brisbane so it’s starting to get a bit cold and a bit snugly, so we love it in Queensland, an opportunity to wear an enormous puffer jacket when it’s just under 20 degrees Centigrade.
Zena:
I’m actually wearing a jumper today, which is so surprising. And Liz is in a t-shirt, which is usually, it’s the opposite.
Liz:
Well, it’s Canberra, right? It’s cold here, but we’re used to it. Anyway, we should get started, I guess.
Zena:
Yes, we should. Kate, Liz and I, obviously we both know you, but we still did a bit of a stalk on your LinkedIn and your socials about your background, and you have an enormous background. You’ve done quite a few different things, and part of those things has been your affiliation with QUT Center for Robotics and also the University of Queensland. Can you share with us a little bit about what your current research interests are and where you hope that they’re going to go in the future?
Kate:
Yeah, for sure. It is a really monumental time in human history with the launch of ChatGPT last November. It was really the moment where all of human knowledge to date, as a high fidelity version of human perspective, ended, and every piece of content on the internet thereafter needs to be thought of as potentially being a manipulated content made potentially by a human but contorted around their use of an AI assistant. So, while GPT-3 has been available since 2020 (and it’s something that I was looking at at Trusted Autonomous Systems in that time), because we knew it was quite a risky technology to be developed, but it was quite contained, and most of the public didn’t really know about it. But the launch of ChatGPT, which brings that user interface that allows humans to easily involve themselves in the process is a great equalizer. Which is great from one perspective, but of course has now introduced enormous risk into our information environment, even greater than all of the deep fake concerns we’ve had and the information warfare campaigns that we’ve seen over the last 10 years in particular.
So my research at the moment is twofold, really. I’m very, very passionate about human-autonomy teams or human-AI teams and how do we ensure the wellbeing and flourishing of the human operator, but also the human decision-maker or the humans responsible for using technologies? How do we make sure the humans flourish in the decision-making, and also those who are affected by the decisions made by those human autonomy teams — those humans are being treated in high levels of human dignity and wellbeing at the other end? So that’s one research question I’m really interested in.
The other is actually about democratic participation in the development, deployment, evaluation and regulation of technologies. So, in responsible research and innovation, there is an understanding and awareness that we need to consult widely with diverse stakeholders around emerging technologies, but what there isn’t, is very good methodologies to actually do this systematically. We see on social media people post ideas and then there’s comments and there’s likes and dislikes, but there’s not a systematic way to actually figure out, well, whose opinions do we need to pay attention to? There’s a bit of a death of the expert going on amongst the public. They don’t like experts very much. They think their voice is as good as an expert in a lot of cases or that their confidence in their own belief is sufficient to warrant or justify actions.
So there’s a bit of a death of the expert, but simultaneously, we desperately need experts from a very wide range of expert areas to help us manage this quickly changing environment of technology. And the old systems of regulation of technology are really not fit for purpose, and a lot of people are aware of this, so the next generation management of new technologies really couldn’t be more urgent right now.
Zena:
You mentioned social media and experts, kind of, I guess voicing opinions and things but I feel that there’s a fleetingness to things on social media. I know you’re very active on your LinkedIn, you’re constantly sharing lots of things on there. So what’s your experience with the engagement of those ideas?
Kate:
Yeah, it’s a really good point. It’s important to recognize the ephemera of it.
being cognisant of any use of these — LinkedIn, Twitter, Facebook. Those organisations don’t care about preserving the archive of your ideas at all, and so you are in a walled garden using these platforms. You have to recognise that if you are communicating on those platforms, you might as well treat it as an oral conversation. Oral conversations are not recorded in general terms, so you must not believe in any way that the thing you said this morning at 7am goes for any length of time beyond that flashpoint of attention. If you want those ideas to carry on in any substantive fashion, it is going to be your obligation to capture them and port them over to your own personal website or your own archive.
Luckily for the last 20 years, databases like Wikipedia or in philosophy, the Stanford Encyclopedia of Philosophy, there’s a lot of places on the internet where facts reside, or best expert opinion exists and persists and has good archival qualities to them. And the paid databases, even though they’re controversial in academia, the paid databases that index and keep hold of academic publications and then the newer pre-print archives like arXiv or the social sciences pre-print servers, these are absolutely critical infrastructure to try and hold onto ideas beyond that moment on social media.
So, I think everyone should use social media, but probably have that archive document plan because the self will dissolve, the self will dissolve. If you only exist on social media, then your imprint upon the world will dissolve, possibly end up in an LLM, potentially end up in AI because it’ll suck it up. So don’t worry, you won’t be lost forever, but you’ll be a statistical reverberation and you won’t have your identity preserved unless you are much more conscious about your information management.
Liz:
It’s interesting to hear you talking about creating that archive, etc., but connecting people to that archive, particularly when you’re trying to have a conversation in social media is a challenging thing. And I’m wondering, how do you approach that?
Kate:
Well, okay, so it’s probably a good moment to think about tiered communication. So, science communication is a whole field that looks at how is scientific information best disseminated? And ultimately science communication, let’s say the COVID example should exist on many levels. There should be this deep science level. There should be this government policy advisory expert panel level. There should be an opportunity for the general public to have a democratic participatory opportunity to voice their perspectives to try particularly to affect biases that may exist by the scientists themselves or the policymakers, particularly for marginalized communities and so forth.
What we lack at the moment is a good methodology for participation into government decision-making that is genuinely connected to the broader community the way they are connected to social media, and has systematic influence into government decision-making where people feel like, if they have a voice in social media, that that can actually be captured somehow. And at the moment, we have this really nasty social media culture for politicians in particular. It’s very aggressive–angry. So one premier or prime minister will put out a social media comment about something, probably sometimes innocuous, maybe they talk about a horse race or something. And then there’s a litany of very angry people talking about whatever it is that they’re passionate about or concerned about–often legitimate things to be worried about, but the acerbic tone of social media is so counterproductive.
No one is capturing that data and saying, “We want to preserve that”—because it’s so nasty. But there’s something important that needs to be captured about people’s experiences in their own countries, to enable them to channel what is very emotive and say, “Okay, let’s unpack that. You’ve got a lot of emotion. Now, what are the concerns you have? What is going on in your life? What do you think the government needs to be doing? And how can we start to package that in a way that the government can actually start listening to it?” Because at the moment, the public is throwing out a lot of energy in a very unpleasant fashion, and the government doesn’t know how to actually deal with it. So often, it gets ignored and then the public doesn’t feel like the government is connected to them. So this is a real puzzle for us to solve as a nation, I think, or as a global community.
Liz:
So I keep hearing the word flourishing and human flourishing in your research and in the discussion that we’re having here. And I’m curious if you could share, what does that look like for you? How do you define human flourishing and how do you bring that into to the work that you do?
Kate:
Oh my gosh, what a lovely thing to ask of me all about my PhD. So, I was very taken by Aristotle’s concept of eudaimonia. Now I’m probably pronouncing that incorrectly, but Aristotle had this idea of humanity, which is we start off as these little acorn seeds and that we have this potential to become great oak trees, but there’s so many things that stand in the way of us becoming a great oak tree and becoming I guess the best version of ourselves that we could be.
So Aristotle had this idea of a flourishing life. If you imagine a flourishing tree, which is that you have the limitations of yourself, you have the things for which you are not able to change your genetic background, some of those phenotypic things that have occurred to you in childhood that have changed your gene expression. You have all these things that you can’t change, and then you have all of the levers that you can change both internally within your internal controls and in the external environments that you are either able to take yourself into or that are imposed upon you by governments and societal organizations and so forth.
So, flourishing I think is the navigation of your own personal agency and autonomy to try and optimize your life to be the best kind of oak tree you can be. Now, the shape of that oak tree, the kind of oak tree you can be, well, that of course is the reflective process that philosophers think is very healthy for a human being — that we should go through a process of, we should have meditation and reflection upon our situation in life and to try and determine what we think would be what we would consider the flourishing self, what is for us the flourishing self, and to allow there to be great diversity between each of us in terms of what is the manifestation of that flourishing.
Liz:
Interesting. I’m wondering how you take that concept, which you’ve described very beautifully, and apply it to the concept of a human-machine team. How are you thinking about the humans in a human machine team growing into the oak trees that you’re hoping they’ll be able to grow into, while dealing with the fact that they are part of a team and they’re probably going to have some shared goal if they’re a team and they’re going to have to figure out how to thrive within that kind of environment while achieving that goal together? How are you applying that?
Kate:
Again, really kind question and a really, this is the core. For me, this is the core question. I always think about this. I love examples like massage robots for cows. I love that.
Zena:
I love those videos.
Kate:
I think the cow and the massager and the more advanced version of that, which is the automated milking machines, are some of the best examples to me of beautiful teammate arrangements. These automated milkers, they do measurements of the cow and they understand the udder situation, but the cow gets to make a decision about how … when the cow goes into the milking machine and then the lasers go, boo, then the automatic utter things go, boop, and then the milk comes out and the cow is like, ah, that feels great, and then the cow can leave. And having had children and been a breastfeeding mammal, being able to breastfeed at any time of the day is a very comforting thing for a mammal. And it’s a wonderful change for a cow to not have to wait for the humans to get up at 4am to be milked. That’s a terrible life for a cow and a terrible life for a human.
So the cow-milker team is beautiful because it allows agency and autonomy for the cow to get their needs achieved, which of course is a kind of biological wellbeing, and the massager serves the cow in that frame. So, while we don’t have conscious beings that are artificial, at this stage, let’s not get diverted to the philosophy of all of that but let’s just say for a fact that our current AI, robotics, autonomous systems are not conscious beings that deserve their own wellbeing metrics, they are there to serve the wellbeing of the humans or animals that they’re working with. And that alignment, that value alignment and goal alignment and process alignment even more importantly, it’s just something to be always front of mind in my view when we’re developing these systems.
And too often engineering gets paid for and optimization around the easily quantifiable factors is paid for, but the more difficult translation from human values into measurable metrics and then the actual money to do the right experiments or the right trials to evaluate how the human feels and behaves with those systems is chronically underfunded and often doesn’t come in right to the end of a technology project. It would just be infinitely better for us as a society to build that work in from the beginning, because it’s a process. There is no academic article out there that perfectly allows application in this emerging technology field. You have to be agile and adaptive through the development and design of technologies these days. It’s just not possible to implement a process that has worked for some previous technology anymore. It has to be done really every time with new technologies.
Liz:
I wonder if you have a sense, even an anecdotal one, of how do you think these things are going to play out? What do you think is going to eventually float to the top in terms of what we end up creating?
Kate:
Yeah, there’s a marvelous example where the invention of the vacuum cleaner didn’t reduce women’s work by the number of hours a day. It made the standard of cleanliness higher. This to me is what I worry about and don’t know what the answer is because I think with the increased productivity with computers and the ability to have the internet has increased the expectation of the amount of work you can do in a single day. And that COVID accelerated that hugely because people are working 24 hours a day now, and then they’re getting that standard of work on their resumes and so people apply for jobs. And if you haven’t been performing that 24-hour-a-day job, then you don’t rise to the top.
And therefore with these new tools, instead of it offering that wonderful opportunity to play with the cat, what a good idea, instead, people are like, great, and now they’re doing this overwork process where they’re setting up … Some people are taking multiple jobs at the same time because they say, “Well, I could actually do this job with this amount of my day and I can do this job over here with this amount of my day.”
Now, in some cases, of course, human agency and autonomy, you can choose to do that and try to make a lot of money, and I don’t want to begrudge anyone trying to optimize their lives. But as the standard for expectation for a normal human being, and I get back to this idea of, well, how much do we expect humans to be able to read deeply? Don’t be overly expectant of people. Allow them to watch the cat videos. It’s not necessarily a terrible thing.
What do we hope we’ll achieve in the balance, I guess, in terms of listening to these lovely podcasts and having that deeper dive and the affordances of technology that enable us to be closer to deep content, which has been accelerated under COVID where we have more access to good content on YouTube where we can’t travel to conferences. Sometimes you can listen to keynotes from those conferences and not have to travel and save those climate miles. So then it comes around human choice both in terms of when humans have a choice of what they do in the hours of the day that they have, when do they choose that content that is that deeper dive content or that more nourishing content, and when do they be manipulated by social media and surveillance capitalism to be staying on the page? Because those metrics, those mechanisms on the social media sites are specifically designed to keep your eyeballs there and to distract you from your bigger goals and your life goals.
I don’t think we’ve done a very good job as a society in bolstering ourselves to fight back, to find our own values, to find our own goals, and then to push our lives into the shape of those goals as opposing having the shape of our lives manipulated and determined by both those social media environments and then our employers and our general society that starts to think, oh, Zena can do six different jobs basically in her week. Well, that’s the expectation now, Zena. That’s the baseline. There’s no going back from that.
I think universal basic income becomes the next logical resurgence of a discussion, because we now have a lot of humans and a huge amount of the white collar class is going to be wiped out by ChatGPT, which to be honest–and it’s unfair and unfortunate–but it’s usually when the more wealthy humans, when they start to lose something, suddenly society starts to take notice. So, we might see some political change because those who used to think that they were valuable in society are starting to not think so anymore. And you might see protest and activism around, well, we just need to take care of the humans somehow and we need to form a model of economic productivity that allows humans to focus more on the things for which they can contribute that are unique and help them flourish. We can potentially offload and outsource some of the more pedestrian work that humans don’t find particularly meaningful to these devices. And if we get that right, it’ll be amazing, but the worry is it’s hard to get the mechanisms in place to really make that happen.
Liz:
Is this a point at which you were talking about democratic processes and their influence on technology and technology’s role in society, is this a point at which we might begin to see some work around that, feeling like it actually really needs to be done. Right? I’m wondering if you can comment on what you’re seeing now in terms of that and where you think it might go.
Kate:
I’m seeing a lot of action by people making the question mark around saying, “We need a future of regulation framework plan,” but I don’t see a lot of action from those who could fund it or support it in actually manifesting it. So, governments seem to be extremely shy around genuine regulatory reform. I think regulation feels … because it’s to do with safety for the large part and not to do with democratic participation, it’s extremely hard for governments to do something innovative that has a great change because the risk feels so high to change the way safety is conducted. And that’s unfortunate.
So there probably needs to be new regulators. And I haven’t seen a lot of really forward momentum by any nation, to be honest, around genuinely changed modes of regulation such as democratic participation in the dialogue around emerging technologies. There are still very old-fashioned mechanisms. We still have working groups, which is an idea that goes back to the Victorian times. We have committees, panels, working groups. We have boards. These are all things for which there was furniture created in the Victorian era and before, around pieces of paper, pens, quills, humans. They’re so old-fashioned. It is boggling that the only global participatory platforms we have are dominated by surveillance capitalism around eyeballs on the screen and cat videos, and has absolutely nothing to do with democratic participation or even organizational participation in a systematic way to lead to an evidence-based decision.
Such a massive gap in our digital infrastructure, massive gap in our digital infrastructure. And no government around the world, all the governments around the world that are leading on policy matters so, for example, I’m just going to use Australian and ally options because I know them well. If we take the US and the UK, they are really good at making policy. They’re actually really, really good at it and they actually put a lot of resources into it. They have the population and the wealth to do so. And if you make the policy, then you can affect the global dialogue to be around … If you set policy standards, then the rest of the world either has to come up with their own to fight back or they have to adopt yours. Australia is much smaller, so we’re always affected by the US and the UK policy positions. And unless we come up with our own, we’re stuck with aligning ourselves with theirs.
So that’s the way policy is done, but then what’s the infrastructure that enables that policy to have the change that they’re hoping to achieve? The reality is vast amounts of policy is never really enacted. There is no good audit trail really around compliance with policy. It’s very hand wavy. And actually, I find using ChatGPT for making policy hilarious because, boy, is it good at making policy. Making policy is one of its best attributes because it doesn’t require any references. You don’t need to cite anybody. You’re grabbing the statistical aggregation from the internet, which is actually quite useful for policy. And it can do listed policy. It can do in-depth policy recommendations.
So the fact of the matter is making policy is trivial now, absolutely trivial. Anybody who has a job in policy, you’re one of the white collar workers–you’re at risk. Because the challenge is not the policy. The challenge is action, behaviour and manifestation in technologies themselves where you can look at a technology and say, “This is how this technology has complied with your ethical AI framework or responsible AI framework.” It’s not just lip service. This is embedded in the process of development from the beginning throughout. And the people writing policy at the moment are over there in their own complete universe and have no direct relationship with industry. They don’t work side by side with the industry saying, “Okay, here’s the policy. How are actually building this week? What are we doing this week on the database? What are we doing this week on the algorithms?” No. Policy is over there on the website that GPT could have written yesterday and industry is over there chuffing along. Massive gaps in those mechanisms that don’t have any leadership that I’ve seen.
***
Zena:
So BetterBeliefs is your company where you are the CEO. And from the tagline that we’ve, again, we did a little bit of a social media stalk, and from the tagline on the website, BetterBeliefs is described as an evidence-based stakeholder engagement platform for actionable and justified decision-making, which is building on what you were just talking about. So can you share with us how you formed BetterBeliefs and what the motivations were behind this organization?
Kate:
Yeah, for sure. It actually came about because I had finished my PhD and I was trying to become an academic at Queensland University of Technology. And Bronwyn Harch at the time, she was the director of the Institute for Future Environments, she had all of these little Catapult grants of $50,000 for researchers who did transdisciplinary research with impact for industry. That was the buzz line for those $50,000 grants. And so I went home on the train, my little philosopher PhD, and I thought, “Well, how do I take my PhD, which is called Homeostatic Epistemology” …
Zena:
Love it.
Kate:
Love it, love it. Yeah, it is amazing.
Zena:
It just rolls off the tongue.
Kate:
Rolls off the tongue. How do I take this really obscure philosophy and make something that I work with others and actually is useful for industry? And I loved that challenge. I thought, wow, that’s a good one. So I wrote down on a scrap of paper and I was like, well actually, what we need is like social media, but instead of just having posts and comments, there’s actually something evidence-based about that. And my PhD had a lot of what we call Bayesian epistemology based on Bayesian statistics, which is all about using information under uncertainty. So, I was like, “There should be a way that we can have the ability to have a post and ‘like it’ but also dislike it so that we can actually get a measure of belief in a hypothesis or an idea. And then instead of just having comments, what if we had supporting evidence or refuting evidence for those ideas?” So, you’re still getting the vibe of social media like Reddit or Twitter or Facebook, but now you’re starting to get data in the service of something that you could actually make decisions around.
So BetterBeliefs is designed to be like a social media platform, but in the backend, we have what we call the Evidence Engine. And what the Evidence Engine does is it produces two different metrics. One is the degree of belief metric, which is based on the thumbs up and thumbs down of hypotheses or ideas. And then each evidence item is either supporting or refuting, and you can actually rank each evidence item out of five stars, a bit like a Goodreads book. And so as a user, you can rank items of evidence. You can like or dislike hypotheses. You can add evidence. You can add hypotheses.
So there’s a lot of different interactions on the platform, but you don’t have to teach people hard philosophy for them to use it. It doesn’t require users to become philosophers. You can take humans where they are, right now, which is that they can use social media, but in the backend, you end up with this groovy ability to say: this is the degree to which people believe in this idea and this is the amount of evidence in favor of this idea. Now if you chart that on two axes, you get degree of belief on one axis, let’s say on the X axis, and then you get weight of evidence on the Y axis. Now you have actually got this wonderful mapping of hypotheses.
So when we did this in Defense for Ethical AI, we ended up with 82 hypotheses. They’re all scattered around this chart, but there was this little sweet spot in this top corner over here, what we call where high degree of belief and high degree of evidence. And those hypotheses get green lit for evidence because you’ve met that data-driven, evidence-based requirement, but you’ve also met people where they are in their belief, which from a Bayesian perspective is the sweet spot. You should act … You should have beliefs that are concordant with your level of evidence. That’s what the Reverend Thomas Bayes said we should do.
So, I’m like we need to make a platform that actually delivers this abstract philosophical idea in a way that’s usable for decision-makers. And so when we worked with Defence, we actually had about 20, 25 hypotheses that were able to be in that green lit area, and we also put an additional requirement, which was that they had to be diversely interacted with. So at least 11 different people out of the hundred people that were at the workshop had to have interacted on that hypothesis for it to make it. And that meant we met a diversity and inclusion requirement, evidence-based requirement, as well as belief.
It was great because it took something which had a lot of divergence around it and a lot of difference where you could keep the divergence. You don’t have to change diversity. You allow that to happen. But then there’s a lot of things for which people do have agreements about and that becomes something you can progress with confidence. So, the number one hypothesis in that environment was that it was really important for AI ethics that we had military education around artificial intelligence. This was the no-brainer. Everyone felt like that’s the first, most important thing to do because with education around artificial intelligence, from there, you can become more ethical across any level of value.
Everyone agreed with this; industry, government, the Attorney General’s Department, the International Committee of the Red Cross, so the nonprofits, the people concerned about civilian harms, the academics. And if you have that cohesion across diversity, that’s very powerful.
So the platform does something that there is actually no other platform we know about that is able to capture the evidence around ideas as well as the belief, but also allows in a safe environment disconfirming evidence and some skepticism with regards to ideas, which is so valuable. If you only have the ability to like stuff or give it a love heart, there’s actually no ability to say, “Look, I’m concerned about this.” I think LinkedIn has a thinky emoji. I think a thinky emoji.
Zena:
Yeah. And it’s got the insightful one, which is the little light bulb as well.
Kate:
But the light bulb is too complementary still. Right? So it’s really hard to be critical in a productive way in our society. I think it’s a leak from the US, to be honest, a US positive culture hangover that the whole of the global society is now stuck with. And when we use BetterBeliefs’ inside Expedia Global on their Hackathon 7, we actually worked across their offices across the world and we actually did an evaluation of how skeptical each office was. And we found that certain parts of the world, like Montreal and India had a more skeptical attitude with regards to hypotheses and evidence. Whereas, American cities like Seattle, which is where Expedia, their central American energy, they were very positive. So they were highly positive about everyone’s ideas, like good job, go, that was a great idea. Whereas, in Montreal, Quebecois, a little bit of French there, and in India, there was a lot more ability to say, “Let’s question this. Are we sure this is the right way to go? This may be a terrible idea.”
The cities that were more skeptical around evidence and ideas, actually their ideas were more likely to win a trophy in the Hackathon because it has enabled them to not just accept everything. They just was able to prioritize ideas with some critical facility. And that was exciting to see that, and that’s where I think some of the potential future can be to actually allow dissent and to allow critique in a safe way.
Liz:
Oh, that’s really important and you’re giving me flashbacks to writing grants for US grant institutions.
Kate:
Right.
Liz:
No, I find that quite interesting. It’s almost as if you need to give people, particularly in places where feedback is generally meant, feedback generally needs to be framed in some very positive way. It’s almost like you need to train people to see feedback as something that is going to provide value, that is going to allow for them to do better. It’s interesting to hear this come through in this platform. I’m wondering how quality of evidence is assessed within what BetterBeliefs is doing?
Kate:
Yeah, great can of worms. So when we developed the platform originally, because I came from a library background and I was very thick with all this literature from nursing and medicine around systematic literature reviews, I had a hierarchy of evidence very clearly in my mind from a scientific perspective. And so I had this idea that we needed two different types of evaluation of information inside the platform. We wanted an objective measure, which actually took quality of information from some form of objective standard against which the information could be assessed, and then a subjective measure, which was around how individuals using the platform who were invited to interact could provide their star ratings basically. I felt like that was the right match, and then you could weight those different types of evidence differently, depending on the context. So that was the original vision.
And in fact, the platform today still has those slots in there for that objective insertion, and it’s still my dream to include that. But the funny thing is when we had our first client, which was actually we built the first version of the platform for a purpose inside Expedia, when we did an audit, an information audit around the way they used evidence to evaluate their ideas, well, of course they’re not using systematic literature reviews to evaluate their ideas. So the best software engineers inside Expedia are more likely to use information sources like Stack Overflowor other dot-com websites and things like that.
So every context of human endeavor actually has quite a different epistemology. Hey, it’s maybe the third time epistemology has come up today. So it’s really important that BetterBeliefs doesn’t become some form of ‘biased to one particular epistemology’ kind of place. So we actually have not included in the current open form of BetterBeliefs an objective measure of information quality. What does that mean? Well, it means it’s really important that you have the right stakeholders to participate in an event. So when you start an event, open up an event in BetterBeliefs, you say, “Do we have the right people invited who have both the right expertise as well as the right diversity?” Because then you’re likely to get not only good information suggested on the platform, but also good evaluations of each other’s information or genuine or reasonable evaluations.
Of course, if you invite a very narrow range of people to use the platform, then you will probably get confirmation bias around a narrow version of the information you’re considering. So, at the moment, the most important thing is that there’s someone responsible for using the platform for an event around trying to address questions, and that you are able to invite the right people who can provide legitimate counters to some of the ideas on the platform. And through that process, we think that you can really triangulate on the right way to make a decision. But of course, if you use the platform, you have results and you say, “These are the hypotheses we think we should progress.” If someone in your stakeholder group says, once you’ve disseminated this, says, “Hang on a second. We disagree. We don’t like those results.” So don’t like them, don’t believe in them. You say, “No problem. Let’s run another event, another workshop, another working group, and let’s see, who did we miss the first time around? Let’s bring them on to the platform.”
And this is not ephemera. We hold on to all of the data over time and we actually measure changes in degree of belief over time on the platform. So you can actually have multiple rounds of stakeholder engagement on the platform and fill the gaps wherever those gaps are. So if you have different democratic groups that you didn’t involve in one stage of your stakeholder engagement, governments can use the platform, do one round, send out their preliminary report results, ask for feedback on them, and anyone that flags a boo, this is terrible because blah, you invite them. You say, “Okay, you’ve got a feeling that this is no good for you. Come onto the platform and start to talk to us about, well, what is important about your perspective that we haven’t captured?”
And in that way, you can actually build up a case, a knowledge case, through interaction, respectful of dialogue, enable people to have emotions, but say to them, “Your emotions are not sufficient. You can’t just have a belief about this stuff. In order for it to be successful, you actually need to make a case.” And that case needs to be evidence-based. The evidence can be diverse. What I love about Bayesian statistics and Bayesian epistemology is it allows even low credibility evidence to play a role in the evaluation of an idea. But it needs to play a proportionate role.
So if you link to a newspaper article about an incident that can contribute to the evidence stack, but hopefully will not contribute as much as a more fulsome research report that’s been done by a responsible organisation. And if you have the right people interacting on the platform, they will adjudicate pieces of evidence proportionately to the amount of influence they should have on the evaluation of an idea.
Zena:
So Kate, I first engaged with the BetterBeliefs’ platform for the Responsible AI in the Military Domain Summit or the REAIM Summit, and that was earlier this year. Can you explain what the role of BetterBeliefs was in this summit and how that role has continued beyond that event?
Kate:
So this event at The Hague in the Netherlands was the first time that about 3,000 experts in thinking about responsibility of using AI in the military domain, were going to meet physically in person. And initially, they were going to just use Mentimeter or one of those typical kinds of survey tools to have digital engagement. And I suggested, because they were open to suggestions, I suggested, “Look, if we use BetterBeliefs as part of your workshop process, you don’t just have interaction that’s a survey. Instead, you actually get something that’s actionable potentially at the end of the event that provides further research that needs to be done or immediate decision-making that could occur.” So they were very receptive to this and I ended up using BetterBeliefs. Well, they said they’d make it accessible to all participants of REAIM, which was wonderful.
They were also careful not to prioritize any particular subgroup, so they weren’t able to put BetterBeliefs on all of their promotional materials, but they did promote BetterBeliefs in their newsletters leading up to the event. And then in individual workshops, we used it. And the way we managed it, because it was actually very little time that was allotted for a lot of these little sub events, we had to be very careful because the truth is if you’ve got only one hour to talk about an idea, it’s very difficult to get a deep dive on that content. The cognitive overload is extremely high if you are expecting people to find evidence for ideas from different evidence sources.
So before the event, we actually sat down with each group that were running an event and we developed good hypotheses to seed the platform with, because if you come into the platform and you actually see high-quality content and ideas, it’s actually really easy to immediately vote on those ideas. That’s a very low hanging fruit, so easy to get interaction on those.
And then during the course of an event, the ideas that are presented on the platform are introduced and emphasized in the actual face-to-face experience itself. So, there’s a correspondence between what’s on the platform and what’s being discussed and an opportunity for people to interact. It actually helped the presenters themselves in the lead up to their panels to get them to hone what they were really interested in progressing. Because I forced them–I’m like, “Well, what do you want to get out of this? Do you want this just to be a talk-fest and it’s one hour and then it’s done or do you have an agenda?” And once they realized that I could help them progress their agenda, it started to click for them. We’re not just doing a survey for the sake of interaction. I actually want to get some political weight behind this.
So I’ll give you an example. One of the sessions we had was on whether using AI to reduce civilian harms was something that should be progressed. Because a lot of the time, the discussion around AI in the military context is to do with force applications, so obviously the use in weapons and things, which is really important to think about, but there’s an opportunity potentially to use AI for humanitarian benefit. So, they were trying to shift the policy conversation around that type of discussion. That was their political desire. All right, good.
So then they developed these hypotheses that were basically assertions from a policy perspective of what they wanted to see progressed, but also they provided some alternate positions as well because obviously, you want to try to show a lot of different diversity in these ideas and then see what the audience actually believes. And we did this, the audience was extremely active, and the data that came out from BetterBeliefs around the use of AI to help with reduction of civilian harms and humanitarian aims had some things around which everyone coalesced on, which was very useful for the policy agenda because there was a lot of disagreement in the room. There was some very angry people in the room who were very vociferous and said, “AI is not trustworthy. It’s going to be used by powerful organisations and entities to try and paper over terrible problems.” They were very concerned about it.
So, if we incorporate that perspective, we also found that even though there was disagreement, for some parts, they thought, no, that would be good. And so therefore that became those green lit hypotheses, those ones in the quadrant that’s easier to progress. And then the controversial ideas remain controversial in the platform and then that also provides an action. You can say this is what we determined was our current belief around these controversial ideas from the REAIM Summit, and now we’ve got next things we’re going to do this year. So those organizers around that workshop are now able to take that data and to say this is what emerged from the REAIM Summit. This is what we felt there was agreement about. You should be aware of that. And these are the things for which people are still very unsure.
One of my favorite controversial ones was around using the same platform for the delivery of humanitarian supplies and the use of kinetic effects. People were very concerned that the sound of a drone that had been weaponized and used in a conflict would then be repurposed with first aid supplies and that this would be traumatic for the civilian population to hear the sound of a drone that was used for different types of purposes. That was something that the military people had not thought about because from their perspective, they thought it would be really good if we can use drones for lots of different purposes and have little different payloads for them and to serve different functions. And they hadn’t thought about something like that that affects the human populations in the environment.
So it was a really valuable capture of that data that is still present in its qualitative form, has not been quantified down to a statistic. Instead, it’s present on the platform and in the report from the event, and then that becomes a really interesting policy conversation around if you develop drones for use in a conflict, do they need to be really substantially different in a morphological sense in terms of how they look but potentially also in the way they sound, and should there be regulation around that humanitarian drones are cordoned off – are very different to drones that might be used with a weaponized effect? And that might be something that the globe decides is a standard that has to be adhered to for the sake of humanity. But it would be such an amazing evidence-based result if that was to happen because it would come from diverse stakeholder engagement in places of safety where people could disagree around hypotheses.
And that to me is a wonderful story from the use of the platform at that major international event that I think is a good blueprint for the way we should think about using it in a lot of different policy areas where there’s controversy.
Liz:
So Kate, a couple of episodes ago, we were talking with Sue Keay about the National Robotics Strategy — which they’ve just (as we record this) opened up consultations for. Do you have any thoughts on open consultations like this at a national level?
Kate:
Certainly, the national conversation around robotics is a great opportunity I think to try and provide the government better mechanisms for consultation. Even the process itself of a government coming out with a consultation paper and receiving consultations is problematic and it’s actually really hard for organizations to process the data they receive.
So wouldn’t it be great if we had a better mechanism for consultation? So that’s actually something that Sue Keay and I have been talking about it and she’s invited me to add a section in the Australian Robotics Network submission to the federal government around the process of responsible innovation and research in terms of the engaged stakeholders itself. So it’s not just a one consultation process. I don’t think that’s good use of everyone’s time. I think there could be a more efficient way for the government to understand the complexities of the decision environment that they face from a policy level.
Liz:
Basically, you’re putting in a recommendation that goes beyond the robotics sector.
Kate:
Yeah.
Liz:
You’re basically trying to say, look, there’s a better way to do these things in general.
Kate:
I know, but that’s such a philosopher thing to do, right? We’re doing it like this but if we stand six steps backwards, because I am concerned about people’s time and I know how much time people put into these consultations, and it can feel like your consultation goes in and your voice is never heard. I know that for me, I put in these submissions to all kinds of things and it’s crickets. You don’t hear anything back from government and it makes you feel fatigued as a citizen and as an expert. You feel fatigued and as though there’s no point. There’s no point to being an expert with an expert opinion about how we should change the way things are done because at the end of the day, the government is too under-resourced to actually process the information in an efficient way to get the results that would represent that diverse stakeholder engagement. So, let’s try and help the government. Let’s be more inclusive. Let’s do it all.
Liz:
We need some more transparent and actually useful feedback loops with regards to that. So you are incentivizing that participation and you’re actually showing that it makes a difference.
Kate:
And it’s doable. It’s not an impossible task.
Zena:
It is doable.
Kate:
It is doable. It’s not hard actually.
Zena:
It just requires change and people don’t like that.
Kate:
No.
Liz:
Which makes it hard, right?
Kate:
It’s the cheese grater of innovation, I call it.
Kate:
Whenever you’re in an innovation space and you’re frustrated, it’s because you’re literally having yourself grated away as you try and change the shape of the environment you’re in.
Liz:
Yeah. Well, but good changes usually involves some sacrifice and pain.
Zena:
It does.
Kate:
Blood, skin, yeah, bone. Yeah.
Liz:
I really have appreciated this conversation. It’s been fascinating. Are there any last words, anything that you want to leave with our listeners in terms of, I don’t know, final thoughts or imagery for the day?
Kate:
Yeah, something more positive. I think people should feel very empowered right now. What is amazing about the world right now is that if you make the decision within yourself to make a change, the tools are now available for most people who do not have a technical background to bring their humanity to the front. So, if you’re someone who has come from a background of not being a technical specialist, but believing or having evidence of a better way to live in our society, you have an opportunity right now. They talk about with alcoholics; you can’t force someone to give up alcohol if they’re committed to continuing to drink. But once an alcoholic decides that actually they know that it’s bad for them and they want to make a change, from that moment, that energetic moment, the support structures and your community and those who love you can help support you to find this new future.
So go and get yourself a mood board and some pens and some pictures and get analog and start to put what is a priority for you in your life and start using those priorities to then sculpt the world that you can because we have this opportunity today. And the only thing stopping us is being in the grip of manipulative mechanisms that keep us glued to these social media platforms and short junk food information snacks that prevent us from living the deeper fuller life of flourishing and wellbeing that we can live.
Liz:
What a wonderful place to end on. Thank you so much, Kate, for joining us today. We’ve had an excellent time talking with you and we hope to have you back another time.
Kate:
Thank you so much for your time and great questions. I appreciate it a lot.
***
Liz: Thank you for joining us today on the Algorithmic Futures Podcast. To learn more about the podcast and our guests, you can visit our website, algorithmicfutures.org.
Disclaimer: This episode is for your education and entertainment only. None of this is meant to be taken as advice specific to your situation.
