thank you very much it's a pleasure to be here this is my third time speaking at rise of AI it's pretty cool from the very humble beginnings to to what it becomes so today I'm going to be talking about towards an AI Commons and I will preface this with a question if there was a KPI for Humanity what would that be would it be maximizing the GDP of everyone would it be making everybody super super happy what might it be and what does that actually mean how would we measure that so if we were to become a little bit more pragmatic what that what might that mean or maybe does it already exist it turns out that there is a pretty good KPI or a set of KPIs it came out in 2015 by of all things the nations of the world agreeing on seventeen combined KPIs the UN sustainable development goals and these are things like no poverty zero hunger and so on all anything for this by 2030 now in there there's a
lot of talk about the climate crisis and so on so it's a good thing that about four or five of these are actually dedicated towards the climate crisis climate action like full of water or life on land and so on so this has really become a galvanizing set of goals for the United Nations and many NGOs and institutions all around it since 2015 there's been growing and growing and growing momentum so if we want a KPI for Humanity maybe this is it by the way they are they also have about 180 measures to see how well they're doing against these goals so the one question is then okay we have these goals sort of to help humanity for good then the question is can I help Canha AI help towards these KPIs for civilization the UN SDGs well if you can look around and it turns out that there are quite a few projects doing AI in ways that can help the SDGs for example satellite imagery can be used to predict the afforestation if
you see that things are being more and more deforested well you once you measure it that's the first step towards optimizing it and reducing deforestation another one is interestingly it turns out that once you can measure livestock moving around it's actually a very good predictor of whether conflicts are going to happen basically cattle don't like bullets so they run another good example is micro-insurance for crops so these are a few examples of AI and satellite projects that are towards these SDKs there's more AI for smart communities smart cities things like a tailored smart city where the traffic lights are much more efficient so if it's a midnight midnight and you're the only one sitting at sitting in a stoplight wasting your time running your gasoline and so on so there's a lot of things that can be done there or epidemic modeling when mad cow happened in the UK every single cow in the USA was killed but what if you could zoom in and pinpoint two which heard it actually emanated from and sort
of as this movement has been starting to happen these specific cases there has been this emergent emergence of a community around it called the AI for good summit the first one was two years ago in Geneva and at that time there was a dozen UN agencies that came as well including the itu itu which is a telecom arm of the UN XPrize the ACM which is a major computer science organization and more so it was really about these people getting together for the first time mostly policymakers less so developers we're getting together for the first time to say what can we do to help catalyze this movement to help make AI improve good rather than just you know selling more ads so this has emerged last year it grew there is about 500 people who came and about 20 or 30 UN agencies and this year it's actually half anymore there's going to be 60 plus agencies and more than 1,000 people at this in just a couple weeks so there's a community for me
around this saying hey you know we can use AI for good so to the question of well can a I help for good for the UN SDGs perhaps the kpi's of civilization the answer is a resounding yes which is great news but there's a challenge the challenge is that of scale so right now there's a whole bunch of people out there that have ninja skills for AI you know AI researchers at universities maybe many of you in this room you're really good at solving problems maybe once the problem is solved deploying it to industry maybe even scaling it up but on the other hand there are the people with the actual problems the problem owners maybe people trying to stop deforestation the people who are trying to stop on the next brick outbreak of mad cow disease this sort of thing and these people often don't have good ways to formulate the problems to vote it into something that then an AI researcher can take and work with so there's this gap between the problem
owners and the problem solvers so the question is what can we do to resolve this gap and in resolving this gap can that make a difference towards scaling up AI for good so right now there's sort of a smattering of these AI for good solutions that I talked about a few of them here a few of them there and none of them have really hit massive global scale yet but imagine the vision of there in the next couple years there are 50 problems to that 50 problems solved in one jurisdiction each initially but then deployed to ten jurisdictions than a hundred jurisdictions so 5,000 AI four good deployments that would start to move the needle that's pretty exciting so how do we get there well going back to this gap between the problem owners and the problem solvers let's try to address that what can we do so we need to flush out the ecosystem that's the first thing so the problem owner is they often have data or access to the people who have
data and then the data owners themselves are storing the data and maybe their own compute centers or maybe on the cloud and as well the people solving the problems running the AI algorithms they are of course talking to the storage and compute services as well so this is a bit of a fleshing out of the ecosystem obviously it's still simplified but it gives us a feel so and notice that the problem solvers are talking to the data owners and the storage and compute but this still this gap imagine we start with this but we say let's structure it a little bit more imagine that we have some sort of connective tissue such that the data owners the storage and compute cloud services etc and the problem owners all of that is leaving as a big fat pipeline towards the problem solvers so think as a wonderful example many of you in the room probably know about image net it came out out of basically a PhD student at Stanford in though in 2011 I believe
and with that definition of these problems you know many many images and people saying hey classify which is a dog which is a cat and so on from that simple problem it was then well-defined enough for dozens and hundreds and thousands of AI researchers to go and attack those problems right using their own resources and so on so that was that's an example of taking one problem and making it available to the problem solvers and it isn't just image net if you go back 20 years we have things like the UCI repository which is a repository of it used to be about a dozen problems but now there's hundreds and thousands our Open ml so imagine if we had some utility some network that connected all of these that connected the data owners and so on with the problem solvers what might that look like you know we can think of it as a public utility network and what is a public utility network well you know we have the lights around us looking down
as shiny on us and that's supplied by electricity and the electricity is this public utility network coming from this public utility network the electric grid it's just there right and we have the waterworks supplying water to us we have sewage systems we have roads all of these are different networks that our public utilities we also have the natural ones like the air that we're breathing and the soil that we're standing on these are all basically substrate for us as humans in civilization now here's the cool thing there's a whole new breed of technologies that makes it way easier to build global scale public utility networks this is block chains block chains are public utility networks now in the room probably many of you have heard about things like Bitcoin so Bitcoin one way to think about it is simply digital gold and you might ask who owns it the answer is no one someone created it more than ten years ago now they put it out as just a piece of software other people downloaded
it and started running it and those pieces of software that were running on different computers started talking to each other and bit by bit by bit it went from you know one nodes to five nodes to ten to a hundred to a thousand two now there are more than 10,000 nodes running this Bitcoin public utility network it is not owned or controlled by any single individual or government or corporation it's just there people have established a value on it you can go on to one of dozens of exchanges and trade it a Bitcoin for US dollars or Euros or otherwise these days it's worth about 7,000 or 8,000 dollars goes up and down but the point is it's a public utility network and for the very first time we have something that is scarce in the digital fashion like gold gold is scarce except it's digital and light gold no one owns or controls it this is a really powerful idea it turns out that you don't need to stop at this scarce token used
as a replacement for gold you can use it in a lot of other ways you can use it to build infrastructure to start to shape incentives to start to get people to do stuff and that's actually basically the superpower of lock chains so next here we go so imagine now going back to this challenge of connecting problem owners with problem solvers we have a network just like the Internet just like the World Wide Web and on top of that just like Bitcoin but it's dedicated towards connecting these people the problem owners are the problem solvers towards a commons a commons as in a shared resource pool that anyone can access just like we have a shared resource pool right now of breathing this air or our access to the electric grid now it's for data and AI services and if we were to flush this out a bit more this is what it can look like so this is the most nerdy slide of the talk so if if your eyes glaze over that's okay
but for the the engineers in the room and the data scientists here's what it can look like for the problem solvers imagine you're using your Jupiter notebooks and you're accessing data and storage and compute if your problem owner you're entering your problem you're specifying some Excel spreadsheets or CSV files saying this is a classify a classification problem and of course you're working with AI people for that and then the data owners and storage and compute similar things you're providing front ends for this and it's all connected in one common substrate that is global this public utility network what might this do well the whole thing here is it's all about unlocking the data we see time and time again you know there's lots and lots of people who have AI problems that they're happy to solve but they have a shortage of data so if we can unlock data for two then we can unlock and equalize the opportunities and benefits of AI so one thing imagine if there was 10 or 100 marketplaces for
data or a thousand buying and selling different sorts of data medical data and automotive data and some of it is free like a data Commons and some of it is is if priced so think like Bloomberg but instead of just Bloomberg providing data for financial services across the board and instead of having to hook to a hundred or a thousand different api's it's all in this unified way that connects so you have many many networks plugging in so imagine we have this network that just makes all these data marketplaces available on top of that you might ask well what about this compensator this free data that could be used to benefit a whole bunch of people of course it costs money to put out there here's an example for you it turns out that right now if you want to predict the weather you actually have to have a really good model of global data of weather data throughout the world and right now there's a problem because in Africa there's a black hole there's
not enough incentive for people in Africa to put up really good weather prediction state weather sensor stations to create to be able to predict the weather well in Africa what this means is that when we're trying to predict the weather globally and understand global weather trends it's very very hard now it turns out that actually if you're a telecom provider you can actually change your cell phone towers you can start to gather data on those towers just a little bit of cost and then the data that's gathered from those two towers things like the change is in AM the lines and so on this can actually be used to predict the weather and there's actually research out of Israel that professors have demonstrated this so it's a pretty small Delta for a telco provider to do this but of course it's not worth or well because it doesn't drive their business and they don't see an end result but imagine if the network itself was paying for this data so you might ask well how
does that work and the answer is it's basically using inflation so Bitcoin let's go back to Bitcoin Bitcoin is actually this self-sustaining network out there and there's tens of thousands of people running nodes why are they running them well it turns out that if you add to the security the Bitcoin network by running a node by adding to the hash rate then you have a chance of getting 12.5 bit 12.5 bitcoins every 10 minutes you have a chance of doing this the more security you add to the between network the more chance you have of getting these 12.5 bitcoins and that's pretty great because that's about a hundred thousand one hundred thousand euros so bitcoin is using inflation it's printing bitcoins in order to incentivize people to add to its security imagine that you have another public utility network that is printing its tokens in order to incentivize people to supply data for the Commons so now I'm a telecom provider and if I see that I provide some of this data for whether usage
or otherwise I can be getting paid maybe ten thousand dollars every ten minutes for this data and that's the idea here so block rewards coming from this to incentivize the data supply so even though the data is free for consumers the suppliers of the data can get paid pretty cool huh finally there's a third way and this is a big challenge many of you have probably been thinking about GDP are right and inside a hospital if I'm a friend of mine he's a researcher using genetic programming to make models of cancer and he tells me that he's a happy man if he has access to a hundred data points from a hundred people and the thing is he's got medical data he's got sorry DNA data so he's got thousands of input variables and only a hundred samples his models are terrible he can hardly predict anything and you know what this means is that these models of cancer maybe they can predict six months in when you're at stage three cancer but imagine if
instead of having 100 samples you have ten thousand or ten million or a billion samples imagine how much better the models can be right instead of predicting six months in maybe it's six days in instead of stage three cancer at stage one cancer and of course that sounds great but you run smack into privacy issues privacy laws as well as simply ethics right but imagine instead where you can say you know what we're gonna build these models we're gonna go to a hospital number one and we're gonna update the model from that and then hospital number two and update the model from that and do hospital to hospital to hospital or even consumer by consumer by consumer right so this is the realm of federated learning now traditional federated learning decentralizes the data but you have a middleman that's orchestrating all of the learning and you have a middleman that controls the final model and of course if if you have a middleman that controls the final model it can get reverse engineered and then
you lose all this private data it leaks out but imagine if all of it is decentralized every single layer at the level of the hospitals and the data itself at the level of the orchestration and at the level of the model itself if all of that is decentralized part of this public utility network where every single person controls their own data then you can actually untap the potential of billions of people's datasets for predicting cancer for ridding Alzheimer's and so on and so forth and beyond medical as well self-driving cars and so on how do you do this well the summary is bringing compute to the data that is the core along this track let's keep going so overall I've talked about this whole idea of this AI Commons you can frame it in various ways as a compute by plying as decentralized orchestration as Sabre for data so just like when you go to buy airline tickets all these different airlines are providing their data whether it's air canada or Lufthansa and so on
but you get to access that same interface no matter what and the reason is in between there is a unified database saver and Amadeus together imagine if that you know unified database was there for all the data and no one under control it that's what's possible now imagine if you're a data scientist you get to keep using your own tools you can use Jupiter and that you can combine this with web 3 wallets things like a theorem and so on or Apache spark and guess what this is actually already software integrations with apache spark and so on every single dataset is an asset every duper - notebook is an asset every algorithm you write as a PhD student or researcher as an asset it can be sold or it can be given to the Commons for free this gives you new powers way more data like I talked about data Commons or enterprises being able to sell the data without worrying about the data escaping you get provenance in your data in AI training this
is critical for GDP our GDP our requires that you actually can explain your the data that you're providing to consumers well what better way to have that if you can actually have a very clear trail of where all the data came from data honey pots every single big enterprise right now has a massive data honeypot its massive liability you have huge costs for insurance imagine instead of that where every consumer is just holding that data and they're giving permissions to the enterprise's to access it for the benefit so goodbye data honey pots and of course more dollars for anyone who's generating data or cleaning it labeling it doing feature engineering and so on I've talked about some applications things like the satellite data to see the impact on water quality better predicting Parkinson's so one project we're working with is based out of Munich and Singapore and so it's two hospitals in Munich and Singapore they're having trouble sharing the data in the traditional ways because of gdpr and so on but now they can
actually share the benefits of it by modeling at this higher level using this public utility network as the connected substrate grow Asia it's a spinoff from the World Economic Forum what they're doing is using AI to help farmers allocate their fertilizer better to increase yields and so on as well as you know for governments themselves right we have GDP are and so on so the the data authorities in the government of Singapore are actually using this technology to give optionality right now we think it's a either/or you can either see the government can see all the data all the time and at the cost of privacy like we see in China or you go the opposite direction sort of like the status quo with Europe and GDP are where it's hard for anyone to see anything imagine if you can have your cake and eat it too that's what's possible now so during the German laws and European laws of GDP are can get reconciled and this is actually what Singapore is doing so so
with that I will conclude with the following slide what I've described to you is this overall goal of asking how can we use AI for good for good as in these KPIs for Humanity the UN 17 sustainable development goals it's happening and now there's technologies to do this at scale leveraging public utility networks and around that there's a foundation that has formed it's a movement and a foundation called AI Commons many of the top EA researchers in the world are starting to join in as part of this and last year actually on stage we had Fred Werner who runs communications at the International Telecommunications Union as well as many others the XPrize folks and others basically announcing the AI Commons it's been slow and steady growth since then there have been workshops and basically this is a movement and a foundation that is driving AI for good at scale so I encourage all of you to join this movement of EA Commons to change the world and I will conclude with a question do you
want to sell ads or do you want to change the world thank you