SPEAKER_00 0:37

Hello, you're listening to the Philanthropisms Podcast. This is the podcast where we try to put philanthropy in context. I'm your host, as ever, Rod Davis, and this week it's just me. There was potential for having an interview episode this week, but I've had to juggle my schedule around for various reasons, so I've got a whole bunch of great interviews coming up in subsequent weeks. But I thought I'd take the opportunity this week to talk a little bit about philanthropy and artificial intelligence, or AI. It's something that I've worked on for quite a few years now. I think I first did some work on it back in about 2016 when I was still at Charities Aid Foundation and doing quite a lot of work on emerging tech and philanthropy. And I've had various sort of ideas about it, that some of which people have found interesting and taken seriously, and some of which have probably been absolute nonsense. But it's certainly a subject that has become ever more relevant, and I think particularly since the end of last year and the start of this year, AI seems to be, you know, a topic of conversation pretty much everywhere on the front pages and on TV and radio and even at middle class dinner parties. I guess, you know, why is that? Well, the obvious reason is the emergence of ChatGPT, which is the conversational interface uh developed by OpenAI based on their originally GPT-3 and now GPT-4 algorithm and large language model. And I think for the first time for a lot of people, this gave a sense of what artificial intelligence really is and what it could do, and has sparked all kinds of interest in harnessing the benefits, but also all kinds of moral panic and fear about what the impact is going to be on various different aspects of our life. I think as as a result of that public awareness, and probably it was already happening anyway, there's also, I think, you know, a big race on to own the technology and also to control and regulate it. So I think companies are all kind of uh circling around wanting to be able to snatch a lead on their competitors when it comes to developing uh tools that harness some of the new capabilities of AI, and governments and regulators and others are thinking about how they invest in AI and develop their own AI policies so that they don't fall behind in a kind of global race. I think there's also, I guess, quite a lot of talk increasingly about the more science fiction-y sounding aspects of artificial intelligence and the potential for the development of genuine human-level general artificial intelligence and whether this would be a good thing, or whether, as lots of people think, it poses a sort of existential risk to humanity as we know it. I think in a more prosaic sense, there are also questions about whether automation, even at a level below uh general intelligence, would have a huge impact on uh on our economies and on the nature of work and what that would mean for the for the future of our societies, where so much of it is kind of predicated on the idea of us all being useful economic units. So I want to talk about a bit of that stuff, but I want to focus also specifically on what the relationship is between uh artificial intelligence and philanthropy, both in terms of how philanthropy might be able to harness AI and also what role philanthropy needs to play perhaps in addressing some of the unintended consequences of AI. I think it's interesting to note, just as a historical curio that you might not be aware of, that there's a pretty strong philanthropic angle on the the origins of AI as a concept and certainly as a piece of terminology. In fact, the the first recorded usage of the phrase artificial intelligence was in a grant application to the Rockefeller Foundation in, I believe, 1955, by a group of computer scientists, obviously sort of relatively early computer scientists, who were looking for funding for what became known as the Dartmouth Conference, which is a sort of legendary computer science conference in 1956, um, which led to the development of a lot of the kind of um early thinking about artificial intelligence. And, you know, that's interesting in itself. I think it also gives a sense that actually, you know, philanthropy might have had a hand in one of the the kind of the problems with AI as a concept and as a terminology, which is the phrase itself, artificial intelligence, is perhaps not always helpful when we're talking about this technology. When we're talking about the potential for general human-level intelligence in a sort of speculative sense, yes, maybe artificial intelligence makes sense, but a lot of the time when we talk about AI or artificial intelligence as it is applied in the world at the moment, I think this gives a false sense often that what we are talking about is sort of human-level general intelligence or kind of the rope, you know, killer robots or that sort of thing that we see on television and in films and read about in books, whereas actually a lot of it is you know much more sort of prosaic. It's actually about you know analysis of large data sets and developing algorithms that are able to um spot patterns and offer predictions. And and so actually there's you know, kind of I think some more specific terminology that is useful to flag up. I don't want to do a kind of AI 101, I'm not sure this is the best place to do it, and also I think there are plenty of places that you can find that out. But just in terms of specifying a few bits of terminology that I'll be using throughout um the podcast, one is machine learning or deep learning, which is a specific form of machine learning. The the crucial point there is that when uh artificial intelligence first came about as a concept in the 1960s, the the way in which you would try and achieve uh intelligence within a computer system was to program in directly the rules for what you wanted it to do. Now, the challenge always there is there are quite a lot of things that humans do which we would see as crucial parts of being intelligent or being human that we're very good at doing, but we're absolutely terrible at being able to specify rules for, and that includes things like image recognition or being able to hold a conversational in in sort of normal natural language. Uh, and so actually the systems were very limited in terms of their ability to do this. The the shift came with um the idea that instead of trying to figure out all of the rules ahead of time and program them in in the form of a specific algorithm, why not develop algorithms that themselves start off relatively simple but are designed to be able, through a process of kind of um trial and error and constant self-um uh modification to learn in a sense or to kind of get better at a specified task. And and this is really, you know, without going into too much detail, where a lot of the kind of amazing um developments in the last decade or so in terms of things like image recognition and what they call natural language processing have come from. And then more recently, um you may have heard about large language models or LLMs. Well, these this is just a specific terminology for the particular application of machine learning that leads to things like ChatGPT, which are language-based systems where they're trained on very, very large data sets of language, um, and then they develop through this um process of iteration a model that allows them essentially what they're doing is not holding a conversation with you in the way that we would understand it. They are predicting what would be the best order of words that would meet the requirements of the task that has been set for them. And you know, there's a very deep question about whether or not they actually understand that in any meaningful sense and what the distinction is between being able to use language in a way that's perfectly recognizable and understandable, and actually understanding it, which anybody of a philosophical bent might have heard of John Sill's Chinese room argument, and it's essentially kind of a version of that, but that's going to take us down a rabbit hole and also get me back into my um university level uh maths and philosophy days, where I'll very quickly prove that I can't really remember most of the detail of these things, so I think maybe let's skirt away from it. But that's a little bit of the terminology. I guess the the other thing I just wanted to say before we dive into the the actual detail on AI and philanthropy is you know what why is it that AI has accelerated so much over recent years, even before the development of uh of Chat GPT? Well, one thing is, as I've kind of hinted at already, the development of new tools, so these new approaches like deep learning and forms of uh image recognition, natural language processing, and development of large language models that have kind of accelerated the capabilities of the technology pretty significantly. Another factor has been a huge explosion in the availability of data. Now, data is the kind of raw material that makes the particularly machine learning possible because you have to have these vast data sets over which um these algorithms can be trained essentially so that they can get good at whatever task it is that you want them to perform. And until I guess relatively recently that was a huge limiting factor. There were probably only you know a few areas of of life or industry where there were those kinds of data sets. But as a result of the internet and um the rise of social media and and other things, there are so many images being created, you know, video, photo, there's so much audio data, there's there's so much language in the form of social media posts and online uh material that all of a sudden data's not as much of a limiting factor in any case, you know, in any sense as it once was. There are still some pretty big questions about the ownership of that data and what it means to train an algorithm on the basis of you know data sets based on human endeavour and kind of who actually is able to make money off those, which we might touch on towards the end in some of the sort of ethical issues, but but the essential point is the data is now there. The the other kind of quite pragmatic point is that even if all that data was there and you developed the algorithms, that really wouldn't work unless you were able to run them on something. And a huge increase in the um the the degree of processing power has been a big factor as well, particularly um the development of ever more um powerful what they call graphical processor units or GPUs, which have been driven quite a lot by the the computer gaming industry because there's a huge demand for these GPUs to make um sort of large, powerful, visually um intense games work, and they are actually very important increasingly now for um for AI and particularly kind of machine learning systems. And then the final factor is is money, you know, as uh as people have seen the potential in this stuff and as some of it's become more of a reality and and it's clear that there is a market impact, money has piled into AI. Whether some of that is part of the classic technology hype cycle and uh you know there's uh a bubble developing, you know, that still remains to be seen. For my for my money, or my imaginary money, because I don't actually have any to invest in in AI in any meaningful sense, you know, there is some overhype, but it's not it doesn't feel anything like to me the sort of hype that we saw in, for instance, cryptocurrency and blockchain, which is another area I used to do a lot of work in where you know the hype never felt as if um it always felt as if it was out of kilter with the reality of what was being developed on the ground, particularly when it came to blockchain and cryptocurrency and philanthropy where there really weren't that many use cases, and uh you know that was out of whack with what was being claimed for the technology. You know, as we'll see in a moment, there there's still a a danger of overhype when it comes to the application of artificial intelligence to philanthropy, but on on the other hand, there are an increasing number of real life tools harnessing AI, or you know, even if that's not the right terminology, harnessing this tech this this kind of technology of machine learning to do really quite meaningful things that could have an impact at least on the world of work. So, you know, for all of those reasons I think it's something we really need to be thinking about. So, in terms of thinking about the relevance to philanthropy, I like to split it into kind of three broad areas. So the the first that I'm going to talk about first is about the direct application of philanthropy, uh of artificial intelligence to finding new ways of addressing social and environmental um problems or challenges. This is what often gets termed the sort of AI for good space. So I think the first thing to say here is, is this a real thing? Yes, it is, and I'll say a m a bit in a moment about um you know some of the examples that I'm aware of that I think are are really interesting. I think there is room for a note of caution, as with a lot of these things, that it it's easy for it to get a little bit overblown because you know I spend quite a lot of my time trying to look for examples of AI being applied in the non-profit or charity sphere, because I quite often get asked to talk about AI and philanthropy, and you know, one of the first things naturally people say is, Oh, could you give me an example of that? So I like to make sure that I've got some to hand. And I still struggle to some extent to find good ones because I think there is still a relatively small number of case studies out there. Now, it may be that we're approaching a tipping point and actually there's a vast number of organisations experimenting with this stuff, then we're just it hasn't quite filtered out into the wider consciousness, or I don't know about it yet. Um so you know, it it might be that all of a sudden this stuff will be ubiquitous, but I think there is a danger of of kind of uh overclaiming for the scale of AI for good as it stands. Um I guess the other thing to say is it might be that we're looking in the wrong place potentially, if what we are expecting is charities or sort of traditional non-profits to be engaging in AI projects or kind of developing their own use cases of AI. When in actual fact, as with a lot of other sort of tech for good applications, it tends often to be more startups or established tech companies that are doing this as you know part of their desire to show that they have a social purpose. That's interesting and it goes to an interesting wide question about the increasing prevalence of corporations and for-profit companies putting social purpose front and centre in what they do or kind of claiming that the work they they do has a social purpose. And tech companies, I think there's a particular danger that sometimes they present the work that they do commercially as if it is, you know, almost as important and from a social good aspect, because they they're so sort of messianically believe that they're improving the world through the provision of their you know widgets and apps and platforms that it's also you know f for the good of humanity and society, and so therefore it you know it should be seen as kind of social purpose or philanthropic. I think there are dangers to blurring those lines. Um not that I necessarily am entirely cynical about the idea that you can uh you know further social uh good through uh for-profit means, but I think you need to scrutinize those claims quite hard. I think the other thing is there's also a danger of what often gets called tech solutionism, which is essentially the problem that you know if you have a uh a hammer, everything looks like a nail. And from the point of view of people coming from the tech industry, they might their worldview is obviously likely to be shaped by the fact that they have uh applied technology effectively in their commercial background in order to sort of make money, uh, and so they might see technology, they might both see kind of problems as technology problems, but also they're likely to see solutions as lying in technology as well. And you know, there's absolutely a huge amount of scope to apply technology more to address social and environmental problems, and as I said, I'll give a couple of examples in a moment, but there is a danger of solutions going in search of problems, um, and that that is something that you do see time and time again when uh sort of people start out from the point of view of wanting to use a particular technology and struggles sort of casting around for for charitable use cases, rather than starting from well, what is the problem, and then thinking about what the the most uh appropriate way of addressing it is, which may involve little or no technology at all and certainly might not involve artificial intelligence. But in terms of there being actual examples of AI for good that we should you know kind of take seriously, um there are some really interesting ones out there. So as I say, I mean, at the moment they they tend to be clustered around areas in uh where the uh size of the data sets is sufficiently large that you can um kind of apply machine learning, and often they are developed in partnership with academic centres and also with the technology industry, because again, you know, the the skills and the know-how um are not necessarily or very unlikely to be found within the non-profit sphere, and that's probably something that we will continue to see that partnership is is a going to be a key part of harnessing this technology. But one area, for instance, is around uh medical research, um, particularly the use of um medical imaging data, so sort of um scans of things like um eyes, for instance. So there's an interesting example where Moorefield's Eye Hospital in the UK is part of a project where they've been applying uh machine learning to scans of people's um eyes, and then using that to develop more effective um pattern recognition and therefore sort of prediction mechanisms to try and spot the early onset signs of certain kind of uh degenerative eye conditions, and it's you know proven to be hugely effective, and this is something we've seen in quite a few other areas of medicine as well. Um, there's also quite a lot going on in the field of conservation, where again there's quite a lot of data because there's um you know images coming in from photos and camera traps and video and that kind of thing. And so there are quite a few projects using um remote camera trap data, for instance, and uh algorithms have been trained that are able to identify either images of animals or things like cryptic uh animal footprints and and signs, and become much more effective at identifying those than humans are, and certainly doing it at a scale that's not possible. And what that's meant in a few cases is where there's sort of very large amounts of camera trap data and things coming in, the people involved in these projects, the scientists and those working for conservation organizations, have been able to get much a much better sense of the actual scale of populations of sort of extremely rare uh animals that are sort of difficult to spot. And in some cases that's given them a sense that actually those populations are more healthy um than they might have otherwise thought. There's also more active uh applications, so there are things like um uh I know the Lindbergh Foundation uh had a project a few years ago and maybe still be going with um uh an AI company called Neurala, where they were uh applying AI to um data that was coming in from um uh video uh surveillance and drones, I think, um in around one um uh nature preserve in southern Africa. And they were then using that data to predict the movements of um herds of elephants and also data on uh the movements of poachers, and were able to develop a protection mechanism that sort of told them where there were particular uh areas of risk of poaching, and then they were able to kind of intervene earlier and move the elephants off to other areas to sort of prevent that happening. So I think that kind of thing's really interesting. And then another example that's been around recently that I think is really interesting, there's a project called Be My Eyes, which um was already around as a as an app, and essentially what it did was offer to people with a visual impairment that they could sign up to use this app, and then initially what would happen is they would they would sort of log a task that they needed doing, so a document reading or something identifying that they weren't able to do because of their visual impairment, and that was then given out to um visually able volunteers who were able to sort of do this as a a kind of digital micro volunteering um thing. But a new version of BMIs or a new kind of app has come out where they've automated some of that function using image recognition capabilities and um actually chat GPT, and so now there is a version where a visually impaired person can use this this app and they can talk to it essentially conversationally and ask it questions about their surrounding, which you know, using the image recognition functionality the the the app is able to then answer and sort of tell them what's around them or to be able to read documents and things and it's entirely automated. So I think you know, in terms of uh accessibility possibilities, which is something I'll talk a little bit more about in a moment, I think there's some really exciting things potentially going on. Okay, I think you know that gives you a sense that AI for good may not be as big as it you know sometimes is claimed to be, but it's a definite thing. Um in the next section I want to go on and talk about what I think is probably the the area in which more most non profits will feel the impact more uh immediately, um, because I think A lot of them will probably see AI for good as a relatively niche subset of wider sort of philanthropy in civil society, and possibly not that's something for them unless they're extremely kind of um you know tech-savvy early adopters. But the the the way in which I think more organizations will feel the impact is the impact that AI has on the wider operating environment in which charities and nonprofits work, both themselves as organizations and the systems and structures that they work within. So stay tuned for that. Okay, so in this section, yeah, I want to talk about the impact of artificial intelligence on non-profits themselves, as I say, both at the level of organizations, uh individual organizations, and then in terms of the wider operating environment within which they work. So, in terms of that organizational level impact, I think this is where you start to uh encounter the kind of uh a lot of the excitement and and you know to some extent the hype about the impact of Chat GPT. There's a lot being said at the moment about what does this mean for fundraisers, what does it mean for charities, how can grant makers use it? Um and this this I think it's you know worth kind of unpicking what some of the potential possibilities are. So not all of these are reliant on ChatGPT, some of these are different applications of of uh machine learning, um, but they're all kind of tied in to the capabilities of AI. So let's start, for instance, with a relatively basic one, which is the use of what's called optical character recognition um for data capture. So this is basically AI technology that uses image recognition, which you can then point at a document and it will recognize the the letters and the characters, as the name suggests, and then it can turn that into data essentially. Uh, and so there are increasingly there are tools you know um that allow you to do this to take a picture of a table or whatever, and then it will automatically you can do this in Excel now, actually, it'll automatically turn it into data in a spreadsheet, or similarly, you can do this, you could maybe have you know like paper grant applications and you could use your phone or another piece of technology to take a photo of that, and all of the information instead of having to be manually entered by a human being, would just be automatically taken into your your CRM or your grants database. And you know, potentially this is really exciting because a lot of that work is um quite boring. I would think most people do it, uh you know, don't necessarily enjoy doing it, and there's an easy, you know, there's a lot of potential for human error and making mistakes. And this this feels like one of those areas where automation, I mean, it might threaten some people's jobs, and that's a question we'll come back to in in a moment, but it feels as though it might actually free up a lot of people's time not to do those sorts of tasks and instead focus on you know more interesting, more rewarding, higher value work. Similarly, there are a whole range of uh you know, really kind of rapidly increasing range of efficiency tools that don't just automate those kind of data intensive repetitive tasks that we've we've been able to automate for quite a while, to be honest. They they manage to automate tasks that up until recently we would have said were exactly the tasks that human beings were still required to do. And some of that is still fairly rote stuff that most people don't enjoy. So it's things like managing your calendar uh or writing emails um or you know producing summaries um of long reports, but but also they're increasingly there are kind of creative tasks as well. So, you know, there are now um products, I mean things like Dectopus, for instance, which is a AI-powered presentation generator where you can essentially get it to do slide decks for you, um, or durable, which is an AI-powered website builder, and I mean it's really very quick, or Looker, which is an AI-powered um sort of uh graphic design platform, so you can build logos and things like that. And and also, you know, there are other things like you can uh if you need imagery nowadays, there is um there are things like uh Dawly, um which is another product um of OpenAI, or um stable diffusion and others, where you can generate art from text prompts. So essentially, if you want a picture of something, so you're like uh I mean I've done this particularly if you want a picture of uh Andrew Carnegie as a Marvel supervillain, for instance, you just type that in and it produces really a very good image of uh Andrew Carnegie in the style of a Marvel supervillain, um, in case that's something that you felt you needed. Um but you can also, you know, just say uh if you're producing marketing copy or website content, for instance, you put in a picture, you know, a photo of people standing around uh looking at a donation box or something like that. Um and instead of having to go and find that yourself on the internet or pay you know rights um for it, which often you you would through a sort of subscription service, you can just instantly get that that um that image that you want to use. And similarly, you know, you can produce written content obviously using uh things like uh Chat GPT and other tools, um, and you can combine that image generation and and uh text generation to produce you know social media posts or you know, website content or marketing content, and and even you know, in in kind of more practical things like grant writing, um I mean there's a lot of people I think starting to experiment with using Chat GPT to write write grants, or at least as the basis or as a starting point for writing grants, and and on the flip side, using it for things like prospect research, so sort of finding places to go to look for funding, again, using ChatGPT almost as a sort of search engine uh there. So there are all kinds of ways in which um what is called generative AI, it's probably a piece of terminology I should have clarified up front, which is artificial intelligence or AI that is able to uh create or generate um you know writing or images or video and that kind of thing, you know, ways in which that could potentially have a huge impact on lots of the different functions that exist within all kinds of organizations, including non-profits. There are some immediate challenges I think it's worth spelling out at this point. Um one is that um Chat GPT is you know is a hugely powerful tool, um, and you know, I've certainly played with it myself, and many of you listening might have done, and it is it is deeply impressive. There there is reason for quite a lot of caution because um it produces uh m material that, shall we say, is not always factually accurate or you know what is technically often known as utter bullshit, because it it doesn't understand in any meaningful sense, and it is a system designed to produce uh a text-based response that meets the requirements of the prompt that you've given it and in the best possible way. And unfortunately, if this means doing things like making up references or even kind of claiming that things happened in the real world that didn't happen, uh it it may well do that. And people are sort of finding this out to their cost. So there's a growing number of instances, uh examples, for instance, of lawyers who have clearly used Chat GPT to help them in preparing cases and to sort of find precedence and case law, and then they've they've gone and presented these cases, and somebody has pointed out that things that they're citing don't exist, they're just made-up cases. So, you know, there's a real need to be very careful when taking some of this information that is produced at at face value. I think there's also a kind of wider ethical uh imit uh issue as well, which I mentioned briefly earlier, which is, you know, yes, it's amazing to be able to create some of these, you know, um uh AI generated images or text, but we probably do need to be aware that the only reason that is possible is because they have harnessed very large uh data sets of human-created images and text from the internet as as training data. And there is a real question about what that means in terms of you know ownership and intellectual property and whether that that's kind of an appropriate thing. I think you know there there's also the the challenge of kind of leaping headfirst into into automation and seeing that as a way of you know replacing human beings, which is kind of the the bad version of of AI and and work, which is the you know the one that kind of gets raised as a potential scare story often. And I think it's important this point to say, you know, it that doesn't necessarily have to be the case. There's definitely a choice there. You know, the the optimistic or utopian vision that a lot of people have is that actually what we will see increasingly is people getting to grip these systems, and we will have humans and uh AI systems working together in in harmony, and actually what we'll get as a result is the best of both. There's an interesting precedent for for this from the world of chess, where obviously computer chess was uh developed in the the 1980s, and then it sort of hit a real turning point when um the chess computer Big Blue managed to beat Gary Kasparov, who was a um world champion at the time, and a chess grandmaster. And I think after that there was a point at which the chess world became quite despondent because they essentially sort of thought, well, have we reached a point where no human being will, you know, in in a few years' time being able to beat one of these chess supercomputers? So what what do we do? Do we just do we let the computers play each other and watch that? Do we ban computers and let humans play humans, but in the knowledge that they're not as good as you know these other chess playing um agents? Or do we allow humans and chess computers to work together and see what happens there? And and this was really interesting because what they developed was what was called centaur chess, so um obviously kind of drawing on the um classical um mythological idea of the centaur, so the sort of half-human, half-horse creature, and this was the combination of human and chess computer, and what they found that was really interesting was that although the best chess computers increasingly would always beat the humans, actually the combination of human and computer would beat the best other computers, so there was something in that mixture of sort of human ingenuity, lateral thinking, creativity, and the the sort of brute processing power of these computers that meant that the combination of them was more powerful than either um working alone. And you know, the optimistic view is maybe that's what we'll find happen, and we'll get a version of you know, sort of centaur philanthropy, for instance, or what um Beth Cantor and Alison Fine, who've written quite a lot about this sort of thing, have called cobotting. Unfortunately, you know, that is the utopian view. Uh, the danger in the short term is that we are seeing exactly the bad version play out, and this is kind of front of mind at the moment because there was um an example that hit the the headlines recently of a non-profit, um, the National Eating Disorders Association in the US, which is a kind of a paradigm example of how not to implement Chat GPT. So this organization basically um they they ran a uh an advice line for people with eating disorders, and they decided, uh, because it should be said, because their human workers decided to unionize, so that's you know another another element of this story that's pretty horrible, but they because they decided to unionize, the the nonprofit decided that instead they would develop a chat PTG, uh ChatGPT-based um automated AI-driven version uh of their advice service and implement that before uh and implement that. But a couple of days before it was due to go live, um, through some of the people who had been testing it pointed out that actually this system had had started giving out advice that was absolutely contrary to all of the kind of best practice on what you you know, the kind of advice you should give to people with eating disorders, and a lot of what it was telling people to do was actively harmful, and so they you know the the non-profit pulled the plug on their plan to uh to implement it. But I think it goes to show that the you know the dangers of kind of rushing to uh entirely replace human in uh involvement in in these systems, and uh when we're looking, I guess, in the non-profit world to implement things like Chat GPT, you need to do it very much with a sort of humans in the loop mindset so that we're not kind of rushing to some of that naive automation. I think that just in terms of some of the other challenges that come with generative AI, another um philanthropy and non-profit example that was pretty interesting uh a couple of months ago was that um one of the branches, the country branches of Amnesty, I think it was Amnesty Sweden, I'll have to double check, but um, they put out a post about police brutality in Colombia, and they decided to use an image to to um to go along with this for social media purposes, but it transpired that they they you know it was it was clear on the thing, it sort of said at the bottom this image has been generated by AI, and this caused a huge furrore. Now, I think the the the challenge here was that Amnesty had done it with the best of intentions. I think they wanted to do this because they felt it was inappropriate to have a picture that of a real person being essentially kind of brutalized by the police, so instead they would produce an AI-generated image that illustrated this same point. But I think a lot of people said that's really dangerous here because you know the whole point for campaigning organizations and and NGOs, you know, is about their authenticity when they're talking about some of these issues. And if people get a sense that actually they're kind of you know using fake images or you know what might be called deep fakes, which is something we'll talk about in a little bit, you know, that that is potentially really problematic because it undermines that authenticity and the trust people have in what they're saying. So I think that was a really kind of interesting example. And then if we sort of move beyond uh the impact on individual organizations, I guess the the other place where we'll see quite a lot of impact of AI that's relevant to philanthropy is in the impact that it has on the wider operating environment. And this is something I don't think there's been a huge amount said about at the moment, but I think is an area that merits a lot more thought and exploration. And this is where the application of AI in things like the financial system or in regulation, you know, that's already happening. Like there lots of financial companies are starting to use algorithmic decision-making systems, for instance, to determine eligibility for financial products or for bank accounts. And similarly, regulators are experimenting with using AI to develop more kind of um predictive uh mechanisms for regulation so that they can shift from kind of policing things after the event when they've gone wrong, to kind of predicting where they will go wrong and intervening much earlier in the stage. And I think you know, we'll start to see some of that happen in the worlds that um that charities and nonprofits work in as well. Um so I think that that kind of impact on the wider operating environment is is really important. But the one that I want to focus in on, because it's it's absolutely sort of front and centre relevant to uh philanthropy, is the impact in the wider operating environment in terms of how AI is going to affect how people give to charities and how they give philanthropically. And I think this is an area again where I've I've been banging on about it for a few years, but I feel as though there's a lot more that needs to be said about it. And I guess the reason that I think this is important is that we're we're increasingly kind of used to the idea of getting tailored personalized recommendations in all sorts of other areas of our lives. So increasingly, you know, we're you know, we use uh Netflix and we get recommendations for what to watch and Spotify tells us what to listen to, and and YouTube, you know, either tells us to watch children unboxing presents or invites us to watch ISIS recruitment videos. Um, you know, because the algorithms I think don't always work, although that's a slight sort of tongue-in-cheek thing, because increasingly they really are quite accurate. But the point is we're getting increasingly used to those sorts of tailored recommendations that are provided by algorithms, what are called recommender algorithms, and I think we will come to expect that in lots of other areas of our lives. Um, and the consumer expectation will extend to philanthropy and to giving. I think in a few years' time, people will think it's kind of weird when you come to make decisions about giving that you don't get the same sort of tailored recommendations. And, you know, people will move to meet that demand. Particularly, I think this is tied into something else that we've talked about on the podcast before, which is the platformization of philanthropy, and increasingly the fact that a lot of giving is is happening more and more through commercial platforms that are adding giving functionality to what they do rather than sort of traditional non-profit platforms. So you've got kind of uh commercial payment platforms um and sort of you know cash payment platforms like Venmo and Cash App, but also things like PayPal. And then you know, in China, the growth of uh of online giving has been really kind of driven by big platforms like um Tencent and WeChat and Alipay and others. And I think these commercial organizations, you know, their driver is to meet customer expectations and preferences in order to make them stay on the platform. If those customers want to give to charity, absolutely great, meet that need. But if what they want is also to get tailor recommendations on where they should give and how they should give, you know, these companies will want to meet that need too. And I think this then raises some really sort of quite interesting and potentially challenging questions about you know who gets to decide what the algorithms are that offer us this information and these choices, and how will that increasingly shape the decisions that we make about giving? Um there's already quite a lot that's been written about the idea of hypernudging, you know, using AI, which is the idea again of nudging, as in the sort of behavioral economics idea that was uh very a la mode um, you know, 10, 15 years ago, but a version of that where it's tiny little um sort of attempts to shape our behaviour through how information is presented, but in a way that it's constantly being adapted and iterated to our our kind of the way that we respond and the way we behave, so that it becomes ever better at kind of meeting our individual particular likes and dislikes and wants. And you know, I think this is something that we'll start to see in the context of philanthropy, sort of hyper-nudging philanthropy. And I I think we'll also see uh the application of some of these systems, not just you know, in a commercial mass market way, but also I think to specific philanthropy advice, which is currently something that is, you know, because of the cost involved, relatively restricted to high net worth or kind of high-end donors. I think it is possible through automation that that will extend further down into the mass market, and philanthropy advice will be something that is kind of more and more um a thing that the you know the average everyday giver is able to access. Um, again, you know, what that looks like and and what that means and who sort of shapes that advice, I think that will be uh pretty important in kind of determining whether this has a positive or negative impact. And then the the bit, the slightly more speculative bit that I'm gonna throw in here because I've talked about it quite a lot over the years. Um and I I do still, you know, I think that this is the direction of travel. And I'll outline first of all what I'm talking about, and then I'll say why I don't think it's ridiculous. So, you know, if we've got to the point where increasingly people are getting advice uh or recommendations on where to give and how to give from AI systems and recommender algorithms, it's only really one small step to say, well, why don't we kind of cut humans out of the picture altogether? And if we've got you know data on where you know the most pressing needs are at any one given time, and then we've got data on what are the most effective organizations kind of addressing those needs, why don't we just kind of have some algorithm that essentially, you know, on a kind of continuous basis matches resources to where they're most needed, and you know, and people can just kind of let that run and be happy that it's you know it's a rational and effective way of matching and kind of more rational and effective than a human being would be. And the reason why, you know, even though that sounds a bit science fiction-y and and mad, I mean, first of all, I would say I'm not necessarily saying I think that's a good idea, I'm just saying I think it is something that might well happen. I guess a couple of reasons are well, firstly, there's a historical context point, which is there's a very, very long history of people wanting to impose more rationality on philanthropy or on charity, and to kind of get beyond the fact that it is you know emotive or subjective or driven by the heart and make it more you know objective and rational than driven by the head. And I think in the context of that long tradition, the idea that people would want to harness this new technology to make philanthropy hyper-rational, I don't think is is actually too out there at all. The other point I think is that at the moment maybe it feels weird because we're still in the early stages of getting accustomed to the idea of reliance on algorithmic recommendations and on algorith and our artificial intelligence systems. But again, I think that consumer expectation and experience is going to change very quickly over the next few years. So in a few years' time, it won't seem at all weird to any of us. And then I guess the the other point is that. We might not be very far off the point where there are new types of economic transactions and contexts in which the only way to harness some of that money for philanthropy might be to automate it. And I'm thinking here of things like the um the potential for what's called a machine-to-machine economy. So another idea that's been around for a couple of years and is sort of tied into everything that's going on in emerging tech and has uh you know some crossover with AI is the idea of the Internet of Things. So this is where you have sort of smart objects, so objects that are that have embedded sensors and also have processors and are able to kind of sense and then respond to their their surroundings. And this might be a smart fridge that is able to tell when you're out of milk and then just kind of automatically orders more for you, or it could be you know much more complicated things and it goes up to things like sort of autonomous vehicles. But you know, one of the things that's sort of uh speculated about this is that increasingly as that network develops and that Internet of Things develops, it will make sense for uh smart objects to be able to kind of conduct their own economic transactions with each other, right without people having to kind of oversee all of that. And so you will get what is called a sort of machine-to-machine economy. And if that does happen, it will be very different to most sort of normal transactional economies in that it will probably be a very, very high volume of extremely fast, low-value transactions. And so actually, if you want to harness some of that money for philanthropy, it would be very difficult for a human being to kind of have oversight of you know any one of those transactions. So actually, what you would need to do is apply some sort of algorithm that would be able to express you know preferences that you had set out uh up front and do that in a sort of highly automated way. So again, you know, I think that's one reason why the idea of um filgorithms, as I've called them, these sort of um you know automated uh AI decision processes for you know that kind of automate the process of philanthropy is worth taking seriously. But anyway, I'm gonna pull us back there from the brink of uh speculation and um uh and kind of science fiction, uh, and just say that in the next section I want to come on to talk about some of the new challenges that artificial intelligence is going to pose for society uh and for the world and what that means for philanthropy and kind of what role philanthropy can play in addressing some of that. So back in a moment.

SPEAKER_02 46:34

Genius billionaire playboy philanthropist, genius billionaire playboy philanthropist, genius billionaire playboy philanthropist.

SPEAKER_00 47:05

Okay, so yeah, we're back now. Uh and just for the final section, as I say, I just want to talk a bit about some of the challenges of AI, the kind of the negative unintended consequences, and potentially some of the you know intended negative consequences and what that means for philanthropy and and civil society, and also what role uh civil society can potentially play in sort of addressing some of these challenges. So the first one is the challenge of uh that I've already mentioned around misinformation. Um, and this is misinformation is kind of not I'm thinking of this as not necessarily deliberately uh false information, but things that is that are factually wrong, and this is where the the kind of chat GPT bullshit uh problem comes in, which is that it's not trying to deceive you for any particular end, it's just giving you information that is wrong, and that might be problematic if we come to rely on on those systems. And I think this is obviously, as I've said, a challenge for charities and for funders and for philanthropic organizations, but it will also be a challenge for the people and communities that they work with. I think there is a linked challenge that goes beyond that, which is disinformation, which is where you are actively providing fake information for a particular purpose in order to um you know undermine somebody else's credibility or you know to destabilize an electoral process or this sort of thing. And there I think you know AI has a particular um role in that because it offers new capabilities to produce content that is sort of indistinguishable from genuine content, but that can show things happening that you know actually have never taken place. And this is where the problem of what I mentioned before about deepfakes uh comes in. And so deepfakes are it's audio or video content generated by artificial intelligence that is essentially kind of indistinguishable from the real thing, and you know, you can use this to produce um you know images of prominent people, politicians or whatever saying things that they never said, and obviously that could have a hugely kind of damaging uh and destabilizing effect if if kind of deployed in in the right way. And you know, the the challenge I think uh around some of this technology, particularly with deepfakes, is what do charities do about it? I mean, first of all, how do they get themselves savvy enough that they know what the problem is and aren't getting kind of caught out by uh deepfakes themselves? I think there's a another question about you know whether there are ethical questions about the use of this technology. Um and I think there's a blurry line in some ways between you know generative AI used in creative ways and you know potentially problematic applications of deepfakes. And we've already seen this a little bit in the charity world. There's been a few examples, uh, you know, even a few years back, of charities sort of producing video content of famous people, um, and then applied um deepfake technology to kind of show them, you know, for instance, talking in languages that they can't talk in. There was one I remember of David Beckham, where a charity had sort of used footage of him and then and then dubbed it uh using deepfake technology into sort of various different languages. So it looked like you know, David Beckham was able to, you know, as a sort of multilingual genius who was able to pass on this charity's message, which, you know, I with the best will in the world, I don't think David Beckham is a as an enormous polyglot. So so and you know, that's a kind of relatively harmless application in some ways, and in some ways it seems quite a sort of clever, canny bit of fundraising or sort of non-profit awareness raising. But in the context of these wider concerns about the application of deep fakes and the the erosive and corrosive impact that that might have on authenticity in the online world, I think that you know there's really kind of potential reason to be concerned about that. I suppose the big question there then is well, first of all, you know, if this erosion of trust and authenticity is a particular problem for for civil society organizations, what do they do about it? As I say, you know, there's an enormous uh role to play here for organizations that are getting involved in kind of fact-checking in the online uh arena and and kind of uh using using often you know technology to combat some of these problems of technology by pointing out how deep fakes can be spotted and this sort of thing. And I think being aware of initiatives doing that and potentially supporting them um you know th through funding um is part of the solution. I mean there may also be, I guess, in a weird way, an opportunity for civil society organizations in that if increasingly online information is difficult to authenticate and people's you know level of trust in it is reduced, actually uh civil society organizations might have a particularly important role almost as sort of oracles or kind of guardians of authenticity by being able to use the the kind of hard-won authenticity and legitimacy that they've built up over years offline to sort of point out where information is is false or incorrect and to be able to verify information that is correct. The challenge, obviously, in doing that is how do they then stop you know legitimacy theft if other people are able then to kind of create uh you know false websites or false social media content or videos that purport to be that civil society organization and and then use that in order to kind of push their own agenda? I guess the the other challenge with this is um you know how how do you cut through it all without becoming part of the problem? Again, you know, how is it possible to to sort of be the one that says, hang on, you know, we need to slow down here and not necessarily just rush ahead with this application, the technology, or do you risk looking like uh a Luddite there? And or or can you kind of harness the technology yourself in order to show positive ways that it could be used? And is it possible to do that with without perpetuating the problem? Um on that that kind of goes you know more widely, I think, to lots of questions about the uh the adoption of technology by um by uh civil society organizations and in the world of philanthropy. Another challenge of AI that I think that's that's important for philanthropic funders and for civil society organizations to be aware of is um bias. There's a big problem that people are increasingly aware of with what's known as algorithmic bias, which is where algorithms and algorithmic systems um produce decisions or take actions that are clearly biased against certain groups in society. And the reason they do this again is not because they're you know horrible and racist or sexist, it is because they are trained on large sets of data and you know they they then shape themselves uh on the basis of that data to uh to meet certain kind of targets that they have been set in terms of goals. Now, if the data on which they have been trained contains historic statistical biases for things like race or gender, which it quite often does because historically, statistically, society has been quite racist and quite sexist, the problem is that if we just use that data without correcting for some of those uh those issues, the um the algorithms that we develop off the back of them not only come to reflect those biases, they actually kind of double down and entrench them and make them even worse. And and it's the problem is exacerbated by the fact that often these algorithms then end up operating as as what's known as black boxes, which is essentially you know, kind of you can't really see what's going on inside them and nobody necessarily really understands it. So the the ability to kind of challenge them or to to work out where accountability lies when decisions are taken that are problematic or biased, you know, that raises real issues. There's also, I think, an issue about inequality. The problem when any new technology is developed is that access to the technology and and the ability to control it and to own it at a different level, that becomes potentially a new dimension of inequality. And given the the potential impact of AI and the sort of huge impact that it could have economically and societally, I think you know, at a at a national level between countries, but also sort of regionally within countries and between different groups in society, that inequality of access could be a huge issue. And you know, unfortunately for charities and civil society organizations, often they are not amongst the early adopters, and they often, you know, uh do have uh sort of deficit in terms of skills and ability to access this. And you know, there's already lots of concerns about there being a digital divide that prevents charities engaging with you know much more straightforward uh digital technology and the internet, let alone engaging with um AI and uh and the you know some of the tools coming uh out of that. So I think there is you know potentially a real kind of issue around access we need to be available, uh, need to be aware of. A couple of the two other things I think are sort of slightly bigger picture uh issues around uh AI that we hear a lot about. The one that is probably more realistic and short term and um but still you know uh not something that we're immediately worrying about in the in the kind of coming weeks and months, is the impact that AI and automation is going to have on the future of work. And again, I think this is, you know, as we were saying earlier, the the difference in the discussions in recent years has been, it's long been the case that kind of blue-collar manual work has been uh at risk of replacement through automation. The qualitative difference is it feels like quite a lot of traditional white-collar or knowledge-based jobs that were, you know, we probably would have thought were entirely safe from automation. It turns out we might be able to automate those too. Uh, to the extent where there are, you know, some people are offering predictions that kind of entire you know, sectors and industries will essentially become redundant because we'll be able to do all of the, you know, the current perform all of the functions they currently perform using uh automation. And I guess the role the relevance of that for civil society and philanthrop for philanthropy at the biggest picture level is if that is true and we're shifting to you know a kind of post-work future, or at least one that looks radically different in terms of our notions of work and our relationship with work, well, first of all, there's an interesting question about whether this will free up a lot more capacity for people to engage in things on a voluntary basis. And actually, you know, again, maybe there's a utopian vision that we have a sort of golden age of volunteering, and actually engagement through voluntary means becomes ever more important. I mean, I think for that to happen, when other plate pieces need to come into place, like potentially some form of basic income, which is you know a whole other question that people are grappling with at the moment. And I guess that makes the point that in order for that transition to happen in a positive way, we we need to radically rethink or come up with new ideas for the economic models that we have, because the one we have at the moment, where people's societal worth as individuals is largely based on the idea of them being productive economic units or taxpayers, that doesn't necessarily work in a future where increasingly people don't have the option of working because automation has made you know a lot of that work redundant. So we need to think radically differently. And so actually, civil society and philanthropy probably needs to put more time and effort and money into thinking through some of that that kind of those you know radical possibilities for the future. And then the other challenge in terms of of AI that's you know, we obviously do hear stuff about, and I mentioned at the beginning, is is the the much more speculative sci-fi stuff. And this is where the kind of existential risk warnings and the idea that kind of superhuman uh AI might emerge which will take over, and you get talk about you know a sort of technological singularity, a kind of point at which technology develops and you know, artificial intelligence develops, so that there's kind of a point of no return where we can't go back because it overtakes us and there's no you know possibility of putting the genie back in the bottle. I guess the questions around this are well, firstly, there's a big question about whether or not we should take it seriously. Um, you know, there's a lot of debate about that, and it's quite relevant for philanthropy because some of that debate is playing out sort of in the world of philanthropy, because a lot of money is going into looking at uh sort of existential risks from AI as a result of um the effective altruism movement and kind of linked long-termist thinking. So actually, it's kind of inherently kind of tied into this debate, and then on the the other side of that debate, increasingly there is a kind of vocal group of people saying, you know, actually, all of this stuff about you know the uh the risks or the existential risk of of superintelligent AI is is massively overblown, and it's also being driven by a particular type of you know relatively privileged white Western male who you know comes from a technological background and therefore naturally thinks that the biggest problems facing society are ones to do with the kind of technology that they've spent all of their careers with. Whereas in reality, they're saying you know, what's more important, at least in the short term, is the impact that AI is going to have at a much more practical level on the lives of you know marginalized communities all around the world. And you know, uh what we risk in putting lots of focus and spending lots of money on the sort of existential risks of AI that may be quite sort of low probability, is that there isn't there aren't resources going to the much higher probability or actually certain near-term impact of the technology on on uh communities you know around the world that don't necessarily have as as big a voice in some of these debates. So I think you know there's a real kind of issue there. And I guess that brings us to you know the final point I want to make, which is what what role should philanthropy and civil society be playing, you know, given all of these concerns about AI? Well, you know, one is I guess let's go with a positive one first, that there's a pot there's a potential for harnessing some of the capabilities of AI and showing how that can be done in a sort of positive and responsible way. Um and you know, that's where some of the AI for good uh stuff, when it's meaningful, is is really positive and important. I think there's also the um the opportunity, not necessarily just through kind of demonstration and adoption, but through the development of kind of positive alternative visions of technological development to sort of show different paths that we could take, and that that some of the fatalism that you hear in debates about AI, which is you know, people sort of acknowledging, oh yes, it is very worrying what's going to happen, but we know we can't stop it, can we? We need to just plow full speed ahead. Actually saying, you know, no, we don't we don't necessarily need to stop and not do this stuff, but there are different ways of doing it, and there are choices to be made. And you know, actually, if we got a broader and more diverse range of voices into these debates that wasn't just you know from the technology companies that are already making these products, maybe we should would have a different perspective on how some of this stuff could develop. And that's where things like I think the solar punk movement are really interesting because they're offering these kind of slightly more uh positive visions of what a technology-enabled future could could be like. I think the other role is you know challenging much more directly the negative impacts of um of AI, you know, first and foremost is just highlighting the challenges as they play out on on you know, in terms of the people and communities that uh civil society organizations and funders work with. Um and actually that's a really important role because you know, civil society organizations are often tapped into lots of sort of grassroots communities and able to speak up on their behalf, their behalf. And that's you know, that's a role they should really be playing when it comes to the impact of AI. And unfortunately, you know, that is a voice that in my experience certainly is sorely lacking in the debates over AI and AI ethics and the development of AI longer term, you know, at the moment is very largely government, the tech industry, and some academics talking about that. There is there is not a voice of civil society in that room. And, you know, I guess there is also, I should acknowledge, you know, in terms of the role philanthropy should play, there is a legitimate role for allowing uh some of the money in philanthropy to be spent on some of those kind of bigger bets or potential existential risks. Um, and you know, that's where some of the money that's going towards sort of long-termism and effective altruism may be, you know, but the thing is, even if these things are of sort of vanishingly small possibility or probability, you know, they're the the the potential upshots of them are dramatic and terrible. And so I'm I'm sort of sanguine about the idea that people should be allowed to spend some money on that. I guess the challenge is that it feels disproportionate at the moment to a lot of people, because actually EA and long-termism gets gets a lot of money by virtue of the fact that it appeals to a certain sort of um, you know, kind of rationalist tech billionaire donor, and that is totally out of step with the money that is going towards supporting things like allowing um you know kind of more traditional grassroots uh civil society organizations to get a voice in debates uh about EI. So actually if those things were a bit more balanced, people might not feel so uh kind of uh angry about it. And and then just I guess not to be sort of overly simplistic about it, why you know what are the challenges that are preventing some of that stuff happening at the moment that we need to address? Well, one is you know, I think at an organizational level, we need more knowledge and awareness about AI amongst non-profits and civil society organizations, and you know, that means everything from developing their own knowledge base and skills around AI, although realistically, it's not about kind of bringing in-house machine learning expertise, I don't think. I'm not sure that's going to happen. So I think there's gonna be a lot to be said for sort of partnering with the technology industry uh in order to kind of have that knowledge sharing. Um, I think there is more to be done to make sure that sort of trustee boards um have uh the skills and knowledge that they need to make strategic decisions about uh about artificial intelligence. And sort of linked to that is the point that I think organizations need to have a technology strategy in place, even if they're not technology organizations or don't see themselves in those terms, because increasingly all problems to some extent are technology problems. Um so I think and and I think not seeing AI as an IT issue is absolutely vital. It needs to be something that is seen at a strategic level by the people kind of leading organisations. I think another thing that's really important in terms of allowing civil society to play some of the roles that I've I've suggested it should play is that we need to find space and time for for civil society organisations and grassroots organizations to be able to sort of stop for a moment and to look up and look ahead. And that's often very difficult, particularly for small resource poor organizations who are struggling to raise funds and to do the work that they're doing and find it very difficult to think about their own futures, let alone think about the kind of collective future of civil society or you know humanity as a whole. But I think it is important to find uh ways of uh kind of providing the infrastructure to allow that to happen and also funding it properly. Um so I think we need you know more money from uh from the funding side of France to be going into creating those spaces. And then I guess the the final challenge, um, you know, which without wanted to leave things on too glum a note, is that you know I've said about the necessity for civil society organisations to play more of a role in debates about um about the development of AI and the impact of AI and the ethics of AI. The challenge is being taken seriously. I think there there is a there are a lot of sort of problematic uh imbalances in power and knowledge um when talking about these these uh topics potentially between civil society organisations and the tech industry and government, which goes to a sort of more fundamental problem about um an imbalance of the value that is ascribed to different forms of knowledge, where if you're talking about AI and you're a sort of small civil society organization and you're trying to talk to Facebook or you know another large tech company, the difficulty is you know, if if they know about how the technology works and they know coding and they've got developers in there, they you know, you wouldn't necessarily have that knowledge and that is seen as kind of highly you know prized, high status knowledge. But I you know I would argue that that civil society organization, if it has knowledge of you know the communities that it works with and the social issues that it's that it kind of um that are part of its mission, that is equally valuable knowledge, but it is not, I think, in in that context, seen as equally valuable knowledge. So I think there is a real sort of asymmetry of power. Um and I think that that's a real challenge that we need to overcome. Okay, well I think that brings us to the end of a uh skirling, whirling run through um some thoughts on the the impact of AI on philanthropy and you know the way that philanthropy may have an impact on the development of AI over over coming years. Um hopefully that's kind of given you some food for thought. Um if you uh want to read some more stuff on philanthropy and AI, I'll put some links in the show notes to where you can do that. If you want to read more stuff on philanthropy and tech and philanthropy just more generally, uh do check out the website why philanthropymatters.com. Uh you can follow me on Twitter at Rodri underscore h underscore Davis or at philiteracy, uh which is more about sort of history and theory of philanthropy. If you've got ideas for people we could talk to on the podcast or topics that you'd like to see covered, do drop me an email. You can find all my contact details on the website. Other than that, just like, subscribe, tell all your friends about it, leave us a nice review on iTunes or wherever else you get your podcasts, and I'll see you next time.

SPEAKER_03 1:09:40

Bye.