Listen to the latest news, expert opinions and analyses on the ever-expanding world of artificial intelligence, data science and machine learning, broadcast on leading talk radio shows and premium podcasts.
Deno versus TypeScript
"Well, let's get to the point. How does type script fit into dino? Won't types crew field into dino as more safe way to writer programs way to make things more certain on expected. So you know that when you write with tight script, your force to type everything. So you always know what you're getting so that kind of fail-proof typing is extremely useful for large shallow scripts software. We recently had to write something that was critical in with money. We went with dice grip approach to make things super stable and predictable. I like that that were predictable in script. You always know what you're getting into the Phoenix Dino can ward with type script natively, you can type refile Andino will parse it just like any other showers creep filing dino doesn't care about their whether it's types creep or chevette script. So they is that there are some issues that the team. Found with a script because they were using task for DNA core and also for the user facing Api the cold but they were finding that they compiling the type script cold was taking a long time and it was increasingly getting longer and longer but they. They've found most important was that they didn't have found that tie screed was helping the organizing the code because I one of the issues I mentioned that they ended up with two body classes in separate locations. Also. They were maintaining like two tie script compiler host one for Dina Co the core. And another for the external user code, but both had a similar purpose and go. In the end they decided to go back, just plain chevette script for the core there were still retain the user code will still be typed. Chequered will still run through the type script compiler but they ended up in working with Chavez. creeped on. Leave for. Dinner. Core. And how does the type script support within dino compared to that of node? Well as mentioned, you can write native dino native the Thai script and lauded on Dino. You can get all the definition of the types in India no riding for example, Dino space types you'll need to sit up anything for Deana to understand types. Quit you just ride your program, it doesn't need any compilation. Dino will just run your. File. Wow that's interesting. How is that possible? How can you run type script without first compiling it down? Well the competitor is already integrated on the inner. Oh interesting. So so it's happening. It's just your. You don't ever have to have the the output of the Java script file from type. Script. Expecting interesting.
Huang's Law, the New Moore's Law?
"It got all lost in the scramble of other news but remember invidia bought arm and that's a pretty big deal. So to pieces about that, I let me introduce you to. Some, people see as the new Moore's law and it might explain why Invidia is doing what it's doing quoting Chris Mims in the Wall Street Journal who I believe has actually coined the term quote as chipmakers have reached the limits of atomic scale circuitry and the physics of electrons Moore's law has slowed and some say it's over but a different law potentially, no less consequential computing's next half century has arisen. I call it wings law after INVIDIA Chief Executive and Co founder Jensen Wang it describes how the silicon chips that power artificial intelligence more than double in performance. Every two years while the increase can be attributed to both hardware and software it. Steady progress makes it a unique enabler of everything from autonomous cars, trucks, and ships to the face voice and recognition in our personal gadgets between November twenty twelve and may performance of nvidia chips increased three hundred and seventeen times for an important class of a calculation says Bill Daley chief scientists and Senior Vice President of research at Nvidia on average in other words, the performance of these chips more than doubled every year a rate of progress that makes Moore's law pale in comparison and quote. And then also from the Journal, a portrait of Jensen Wang and his founding and stewardship of Invidia as the company has. Just exploded in all sorts of ways. In videos market value has soared to three hundred and nineteen point, eight billion dollars surpassing Intel's valuation of two hundred and fourteen point five, billion dollars even though INVIDIA had ten point nine, two, billion in annual sales in its latest fiscal year compared with seventy one, point nine, seven, billion for Intel. Videos bet on some of the hottest fields, INTECH video gaming and artificial intelligence have fueled investor enthusiasm. While Intel has stumbled with some of its most advanced chips, it's all a long way from nineteen ninety-three when Mr Huang. Dreamed up in video on his thirtieth birthday at a Denny's in San Jose California with two like minded engineers they bet on a future where consumers demanded better computer graphics, which would require specialized high performance hardware that wasn't available at the time and
What it Means to be "AI Ready" - with Matthew Mattina
"This is Daniel Magellan. You're listening to the and podcast. We speak this week on the topic of. Readiness. What is it look like to truly be ready as an enterprise if you're a consultant, you're selling into enterprises and you want to build a success wave. This client stands in what might need to be worked on or if you are an AI champion within an enterprise, you want to get an understanding of where do we stand how ready are we should be an awfully helpful episode our guest. This Week is Matthew Martina who's the head of The machine learning research lab, A. R. M. A. R. Ramsey multibillion dollar semiconductor and software development company wholly owned by Softbank Softbank One of the biggest venture funds in the world based out in. Japan, and Matthew speaks to us about his criteria and his way of thinking through with Ai Readiness looks like in an enterprise again, if you WANNA a checklist to list of features away to assess take view on your own company or that of your clients. I think at this episode should be awfully helpful. If you're just getting started with deploying a, we have a free guide called beginning with Ai. It's special guy for non technical professional. So if you do not have a technical background, but you still want to understand what is it realistically look like to deploy artificial intelligence were the key factors to understand for a adoption. If you're not the person writing code, you're more focused on the business strategy side. Of things then you'll WANNA download that free pdf brief it's an e. m. e. R. J. dot com slash B e g, and then the number one. So bg like beginning and then the number one that's RJ DOT COM slash bg one that pdf should give you some extra details to layer on top of some of the insights that Matthew provides for us here today. So further ado this is Matthew mcconaughey with arm on the and business podcast So I'll kind of dive in first here on this topic of Ai Readiness and ask you about what you consider to be sort of the core components, the core aspects of Iranian s within the enterprise obviously a lot of moving parts here what comes to mind for you? Yeah. That's a good question I think. One of the core questions is one that I think people sometimes miss with respect to a I is. Now there's the problem that you're trying to solve. Of course, understanding that from the get go is key in pretty much any scientific or engineering discipline, but then with Ai. Knowing how your machine learning or a model actually going to be deployed. So what is that model gonNA run on in the field as it can run on a some kind of a big server in a cloud data center somewhere with no terabytes of memory and an of GPS and processors, or is that model ultimately going to be deployed on some kind of you know very constrained embedded device say you know in a in a o not censor or mobile phone or a car and everything in between? So think what we sometimes see is that a model will be developed by a data scientist or or. application will be developed without a good understanding deployment and where that gets prickly as you've developed this model, it uses you know. Fifty gigabytes of memory and then Lo and behold actually want to deploy it on a constrained device that has you know two hundred and fifty six kilobytes of memory, and now you need to do some surgery. Got It. So readiness here you're talking about you know not only involving the model, but involving sort of what are we going to run it on DC? This is potentially part of the four thought process for companies obviously, not everybody's GonNa have devices. Out in the field, people have security cameras, La- run things on mobile phones you know in in cars or maybe heavy industry the have it on a boat somewhere maybe other folks are just GonNa have stuff up in the cloud but for you, it sounds like maybe that thought process should happen as we're coming up with ideas not sort of after we've developed a great model idea that those have to be married to hardware sort of at the brainstorm phases kind of what you're getting at. That's exactly right as part of the upfront? Planning. Stage of enterprise preparing for a readiness. Yes. Some consideration for. What devices is this actually gonna run on and what are the key characteristics of those devices and and the interesting about it is that like I said, you can build models you know and build ai applications that you know recognize faces and use lots and lots of memory or they can have models at recognize faces and use very little memory. And making that trade off and understanding that that trade off will need to be made between accuracy and memory upfront will save people pain down
Using Existing Edge Hardware for New AI Capabilities - with Roeland Nusselder
"So Roland. Glad to have you on the program. I. Know we're going to be talking about a at the edge. I. Think in order to have that conversation based on where your firm is focused we should talk about micro controllers and a tiny am l. This is for for you folks really really big opportunity for ai at the edge he maybe not up what we're talking about today. Yes sure thing very me of course. So tiny mel is machine learning or a I own really cheap low power hardware and then usual micro controllers. Some Mike controllers are very cheap low-power chips and they're literally everywhere are hundreds of billions of my controllers in a roads sets also why they can be cheap. But it's very challenging to deploy machine earning or to run machine learning or microcontrollers and its resulting but maybe it's good if I did a bit of a wide so important to run machine learning microcontrollers. So one way to do this, I mean if he thought, you could think that you can just sense the data back to the cloud depressing Darren very heavy and. Expensive jeep use bid. This is often not a very good idea. First of all, they're bench with limitations. So if you have a camera that's connected to Wifi network and they sent up the whole camera to the clouds, if you have multiple gummer's connected through same wife network that just doesn't work your your wife is down immediately than there are things like latency take. Time to send it to the cloud to person Darah Senate back there reliability issues, diner radiant, and it's done. You still want to make sure that your product works their privacy issues. You don't want to send feed data or audio data through the clouds and dinners energy consumption. That's actually a big issue because sending data to the cloud or even if use WIFI consumes a lot of energy and that's not good. especially not if you have a battery powered device Saddam L. Solstice by running machinery and work by running two workloads on the device itself the very cheap low power chip. But. The thing is it's very difficult to run machine learning on a chip and that's what our companies folks them. Got It. When you say chip in this case, you're you're talking about microcontrollers. Yeah. Exactly. Okay. Got It. Got Maybe. So you've walked through a couple of instances an. Familiar. With you know the edge as as sort of an idea and the intersection of of Iot nfl you focus on this space pretty ardently though maybe we can talk about some of the cases where hyping data to the cloud makes sense and some of the cases where a dozen year security he brought up bandwidth. There's a lot of these practical concerns. Can we tie this to you know potential business cases Hey dan here's an example where it completely makes sense we we gotta send this stuff up it might be at the edge, but we got to send it to the club and here's an example where we really should not be doing that Do you have any? We can talk about The main thing here is that you want to have devised a battery powered. That's an important issue because makes much cheaper much easier to install vices. So for example, if you have a small camera in a grocery shop to detect if a shelf is empty or not. You want to make the device battery-powered at getter with smoke, and you can do that if the machine learning workloads is running on a mic controller and it just sense a small signal to the cloud if the shelves empty or if the if the shelf is enough to empty and you always want to tasks on the edge, if that's bull doing reason, why would not want to do it on the edge is if you need so if the Model is so complex that requires a lot of energy to run and requires very expensive. Large ships skull jeep use said, that's why you want to do hoax over example for very complex and opie models of complex NLP tusks you you might want to do it in cloud, but if it is Bulbul, if you game run it locally, you generally want to do that because of bent with. Issues, because of latency issues, because of reliability issues because of privacy issues. So a good example of adults that you want to run locally is, for example, an H. system where on a heating air conditioning system where you have a small camera which idex if there are human in room, and then if there is a human into room, the heating air conditioning system automatically turn on or turn off. You don't want to send it whole feed your feet to the cloud it. It's not great for your free event with of your Wifi network. Another great example where we ecstasy for between is in retail. So you're starting to see devices that are battery powered that have a small camera and that have a small microcontroller controller, and it runs in a little deep learning model to detect if the shelves empty or not car, and if the shelf has empty than the signal to store manager that someone needs to fill up the shelf again. Yeah. Yup or for example, as small camera detects how many people are waiting in acute and at the stores manage store managers do more effectively locate their their stuff thus like debt or you can do gay seduction off shopper and detect. What kind of products show purser January interested in said at the shop nick and do better product placements those kind of thoughts. been if you're running ds on larger chips that are more energy consuming, you have to connect them today Tristan Nets, somehow that makes it much more expensive to install much more painful store owner. So if he can make this battery-powered, you can just clip it on a shelf or you can just Louis this evening for example, would you really like if you can make these device better half these kind of
"I'm Derek Limb. Currently, going into my last year of undergraduate study at Cornell University. Very neat to what in particular you studying. Study math and computer science majors and more specifically I do work in the applied mathematics side of things in the machine learning and data science side of things very cool for me I come more from the computer science side. I have a vague sense of what applied maths means, but I will able his vague I'm curious do you have any particular methods or ways in which that blends nicely with the data science that you might WanNa share oh? Yeah. Applied Mathematics is already a vague. Term. So don't worry about that. But yeah, it's very broad term but tools that I like to use, and that are generally very helpful. Are things like numerical computing numerical when your algebra optimization and label a lot of those as applied math but it's definitely a very blurry term. Yeah. Neat areas though a of interesting research going on in that spot, the paper I invited you on to speak about is I, guess related to that but sort of unique and interesting in its own way title. Is. Expertise dynamics within crowd sourced musical knowledge curation, a case study of the Genius Platform. So maybe a good opening question here is, what is the genius platform for anybody who doesn't know? Yeah. So there's a website genius dot com to really great website hosts all sorts of different texts content mostly by it primarily hosts song lyrics, especially, rap lyrics because that's where the origin of the site came from. So what happens is you use crowdsourcing use the power of the crowd to transcribe. Song. Lyrics. Molly's different songs at an after that users also come and annotate the song lyrics. So if there's some lyrics that use some type of terminology that not everybody might know or some lyrics for which is interesting backstory that's what annotations seek to answer an annotation provide all types of very interesting knowledge and information and all types of different media to help explain what lyrics are about could you contrast the content to some of those user contributions to maybe what wikipedia users put into the Kapadia. Yes. Yes. So a lot of the annotations written with I'd like to call it like a view of culture of rap culture and they will use the slang terms and all this that rappers use I mean obviously because you have to explain the lyrics that rappers use but also they include parts of hip hop culture end generally music culture that are prevalent today such as there are a lot of twitter links and links to interviews that artists do in these annotations, and it just provides a lot of really interesting stuff when I think of Wikipedia or if we I guess rewind ten or fifteen years to win, it was getting going it seemed like every stand up comedian. Joke about you know it's going to be accurate because literally anyone can change it. But that did seem to pan out wikipedia startlingly accurate in a lot of situations. But maybe part of the advantage of Wikipedia has is the goal is to publish. You know the sort of unbiased truth that's what an encyclopedia should be. Leaks are I don't know that there is a ground truth in lyrics. Are there issues of pollution in the state of set like people who are into the Polish Dead Beatles conspiracy tagging the lyrics that They think are clues or something like that. How do they keep it clean? Yes. So there are a lot of studies on wikipedia that show in certain fields actually wikipedia quite accurate in covers a lot of information, but there's also a lot of interesting studies unlike edit wars but I have not really found instances of edit wars on genius, which is nice and maybe people don't make Super Controversial annotations but I definitely noticed when analyzing the annotations that a lot of these users are definitely putting a lot of their own. Into, these notations, and actually quite a good amount of speculation and you know, maybe the stakes aren't as high for genius as they are for an encyclopedia that a ton of people as a very important source every day. But it's also something that I think genius Kinda wants to avoid because if you read some of their like frequently asked questions, pages or things like. That they will recommend you to add a lot of factual sources but I'm sure that they definitely like the opinionated takes sometimes also not everything can be factual when you don't have the original artist interpreting your lyrics. But yeah, I don't think it's that problematic and I actually think in fact, it has an interesting twist to all these annotations it allows users to add an. Interesting twist to their own annotations.
Trends That are Changing AI Hardware and Software - with Marshall Choy of SambaNova
"So Marshall Glad to have you back on with us today, you guys have made some great headway in the last year or so since I had you back on the program and I know today, we want to talk a little bit about trends. You've seen in AI hardware. You guys obviously have finger on the pulse here you raise a lot of money got some of the sharpest focusing industry when you look at what are the trends moving dollars. Moving the smartest technical in this world of AI hardware how do you like identify those are the big ones for you. Thanks for having me on Dan requests. He again, there's a number translator going on in the space day and We've kind of narrowed down on really the three most prevalent trends were seeing to our interactions with customers now in the market and really using our view, these trends kind of inform our internal thinking in our our development processes and so first and foremost. I think you know nobody's probably better qualified to declare this. First friend is our CO founder and chief technologist Kounellis Colton. who was the father of multiple? End. His declarations been the multi processing is kind of reducing in its utility in coming to an end of life as a result of the slowdown things like Moore's Law Dennard scaling. So if you look at a lot of people in the space, they're continuing to build A. Systems processors based on a on a core base approach. As cores themselves become less and less effective, inefficient, just putting more more course together multi courtship minimal ecosystem only fields you in an even more inefficient system, and so what you need is something it's going to be more flexible and more performance than the core approach and so on. This, you know we'll. Just for a second year, Marshall I'm aware that you guys spend a lot of time in the nitty gritty hardware, a lot of our audience they're going to be you know our head of compliance at a bank or head of innovation at our ECOMMERCE. They probably understand you know coors at a very conceptual level like, oh, yeah. Intel has those you know something like that. Yeah. But even for me understanding core and multi-core at a conceptual level defining that, and then I guess contrasting it with with what could be would you mind to finding those churches? The folks at home can kind of imagine this? Yeah, I mean think of it this way right I mean in the past with with. Transactional processing computing core banking taxation like that people assembled systems with many cores were. Had fixed functions and usages for those specific software operators. With machine learning and ai the whole software development and delivery model has changed in acquired a whole different type of computational capability, and what that requires is the ability to have a much more flexible silicon infrastructure to run the applications to effectively provide the software, what it wants needs to do data flow execution operation. So that's really kind of the core difference here. Is really being driven by the software, which is mandating the needs for different types of infrastructure. Why do cores ORC with? Let's say software as it was you know, let's just think of whatever kind of software we want. We want to imagine wise a core suited for kind of traditional it traditional software I. Know You made the contrast with machine learning, but I'd be interested in yeah. Yeah I mean traditional computing is the Very predictable and in in deterministic in nature, it's all about calculating to the degree of accuracy. For example, if you're trying to calculate someone's bank account statement, want that to be you know too many many. Ratchet. As opposed to nation and so with a core approach in a traditional compute approach, you affectively hard code in these functions. Antics of these, you know well known operators that are GonNa be used the software.
Exploring The Future of Health through Dreams and AI with Antonio Estrella
"Welcome back to the podcast that I have the privilege of hosting Tony Australia. He's a managing director at Talladega Investment and advisory for health tech and insure tech startups. He's also a fiction novelist Tony's a global thought leader and fiction writer and digital health with experiences working in Asia, the US and Europe as a startup founder investor or Britain ovation leader and strategic advisor Tony currently sits on the board as an independent director, for C, x group, and Savannah CTS as both. An investor and adviser Tony Partners with Asia focus companies who are working to develop solutions to change the face of cancer human longevity and population health with core IP stemming from AI genomics blockchain smart devices, his previous work within both life insurance at metlife and farm out with Pfizer, it was focused to drive measurable business impact allowing him to help entrepreneurs enhanced their product market fit and commercial growth plans across Asian markets, his debut fiction novel comatose, which will touch on here. In today's discussion is a fiction novel about Lucid Dreaming and it's all about health tech fiction something that will cover with Tony as well. It's available in bookstores today in the UK and Amazon globally. Tony is has done tremendous mono- work and he spent some time at University of Pennsylvania's wharton getting his MBA there the London business school and the University of Pennsylvania School of Engineering and Applied Science in electrical engineering. So a tremendous individual and it's a privilege to host them. Here today. Tony thanks for joining the next. So the pleasure to be here, thanks for inviting me to share some of my thoughts and insights with with your audience. Absolutely my friend. So tell me a little bit about your journey. How did you decide on healthcare? So I academically studied electrical engineering and that's actually where I caught the bug Ford being more entrepreneurial minded and how I focused by professional life I used to build and race solar electric race cars really. Little coffee that I helped build up and and I started my career in consulting and during that period was great you know lots of. Ways to learn and be mentally intellectually challenged. But in two thousand, I had just finished doing work in Silicon Valley and that was the first Internet wave and lots of excitement about transformation and as I started business school I really thought about where did I want to dedicate my time and energy in terms of industry focus for several different reasons including personal wants healthcare just jumped out. I love the fact that you can build technology and it helps people live longer have better quality of life I had a couple of. Personal Peoria friends who dealt with health issues. I had an aunt who passed away from kidney failure and so all that just came together for me to say I can wake up every morning. Feeling excited that what I do is helping at least one individual of a better life love that man yeah. It's a compelling reason to choose the field and with your knowledge and background you've been able to make a big impact and so I'd love to hear from you. Tony will you think is should be the big thing. On health leaders agenda and how are you approaching it back when I started my first business in two thousand one, there was a lot of emphasis in terms of whereas the healthcare industry in the US the US at the time and fast forward through time they're still an enormous amount of of focus in the US in the healthcare sector is digital health or health tech has grown the US. Market clearly is an important one, but I'd say that equally as important that on every health leaders mind should be what can they Learn from what's happening in. Asia and Asia whether Asia's an opportunity or not is there are there things that Asia offers in accelerating growth and scale and product that can be leveraged for for their business and couple of facts about Asia that I think are important for plus billion people forty four countries over two thousand languages spoken and normally large region and from an investment perspective this two, twenty, eighteen we saw the Asia approaching the same amount of investment to help tech startups is in the US style so within the next. Eighteen months you'll see that Asia, actual have more capital being deployed from the venture community and startups. So when I say that every health leader medically look at Asia, it's because the region is just is as awards today with with a much greater growth potential in the number of people countries. So there was a book I read recently by Kaifu who was a venture investor, in China, who formerly headed up Google China and used to work. For Apple and driving their early AI, and he doesn't amazing job painting the picture for China's one country when when important region round where they're going with a and how it's different than the US and I think that's the key thing that a takeaway for health for health leaders it's just a different technical environment data standards, and in the way that the tencent and Alibaba by do have changed China much the same way that Google facebook. Changed West is lots of learning that can happen man that's fascinating stuff Tony and folks I forgot to mention to you that Tony Lives and works in Singapore. So he's he's been there for the last five years this time around but definitely, a global health leader focused on Asia that knows the INS and outs. So critical critical piece of of information there everybody. To know. Tony, without a doubt there's there's opportunity over there. The money's flowing over there. Give us an example of of what you've seen is working and creating results. Yeah. The landscape for Asia is complex As I said, there's lots of countries and so before a answered that question, let me give a little bit of context as to how to think about the region. So. One is mentioned China and you can group Hong Kong and China together from thinking about one of six hubs in the region. The other hubs are the Indian subcontinent, which obviously is driven largely by India, but there's other countries their third. It'd be Japan for the be the Korean. Peninsula, which includes South Korea Fifty Southeast Asia Singapore and then six to be Australia New Zealand and I didn't do these in any order of size of just kind of went north to south and regret yeah, an each hub has. Similarities that that make a logical grouping whether it's economic development or cultural lifestyle history or climate.
Google Will Block Its Autocomplete Suggestions For Some Election-Related Searches
"Is removing some of its auto complete predictions and evolving the upcoming election. The tech giant said it will remove predictions that could be interpreted as claims for or against any candidate or political party. It will also remove suggestions about election security and voting methods. Uncle gave the example saying it won't give suggestions of parties or candidates. If a user types the words donate to the auto complete suggestions operates by what the user is typing into the search box, which then suggests searches.
Portland enacts most stringent facial recognition technology ban in US, barring public, some private use
"Portland, Oregon has just passed what experts say are the strongest limits on facial recognition in the country. City Council unanimously voted to bar city agencies from using the technology and to ban businesses from using it and retail stores or other public places. An official with the Portland Business Alliance tells DowJones exceptions should be made for hotels or banks that let customers opt into facial recognition experiences.
How The Consensus Voting Mechanism Works
"So my name is Mush. Pot. Suzuki like the car last year a student at Yale University in Computer Science and mathematics and I'm supervised by Adrian Vata and from goalie. I'm very happy to be here and thank you for inviting me. I got exposed to your work. When I read the paper, I invited you to come on and talk about mostly today is how many freemasons are there consensus voting mechanisms in metric space so no less than three interesting ideas for me in the title maybe we should give some background to open up with consensus. What does that mean is that? Just the majority or how will we be using the word today? Consensus mechanism is Mecca Zeman which to select a candidate they're only if everyone agrees on it for instance, they said you're running for some group Peter, you get accepted only if everyone in the group agrees on it and you want you accepted otherwise. so that's like unanimity the Neha in here. Yeah. It's basically the same as unanimity but in different. Settings. Unanimity also implies that it's a property where you have if everyone agrees on that, you accept it but something can happen. If you know, let's say you can use different rules otherwise there's a different sort distinctions there, but it's basically the same as unanimity that particular choice leads to some interesting things and you guys are presented this really concise analogy to the freemasons. Well, I suppose getting down to the actual counter freemasons trying to conduct a census here that just formalizes. A funny title to give because as you know that freemasons are interesting sort of group there. Member only if everyone agrees on sort of it, fits our mechanisms today run their group through consensus voting Gotcha so it wouldn't surprise me if just given how much sort of folklore the surrounds the freemasons that there might be at least a few listeners confused about whether or not there are real organization. Could you throw a few facts Edis? have. Amazing temples in Montreal, phrases, I live in Montreal. So they have a huge temple here and they're very active and I don't know too much about their inner workings or I've never been inside the doors but they're very oh yeah you can go to their building for instance so real organization and in terms of getting accepted into it if I wanted to go in that building up there in. Montreal is this the actual mechanism they literally use or is this a sort of an analogy? So I don't know this for sure but I heard I read online that in order to get accepted that everyone in taboo are everyone in that group has agree on it. So is the medicine that they actually use. Cool. Well, I guess whenever that started it wouldn't be surprising that it could grow quickly. Right, if I was thinking of starting my own such organization and I I, invited my wife. Now we have two members and perhaps she and I can agree her sister joins and my sister joins. But at some point, someone's going to say no to the next member what can we learn about that and how do we study it? What are the interesting Totta questions? So you have two different settings. Here. So wondering is not remodeled people's opinions and you say I prefer my wife says I prefer my daughter you know or or so and so forth we need to the preferences of people. So one of the classic ideas in voting theory and actually just modeling through machine learning or any sort of the setting is this prioritizing people's opinion or privatizing people. So for instance, let's say you could be. Leaning like the political left or right. There's a spectrum for which you are lighted. Let's sure you can be at the centre or you could be very right thing or it could be a very level. So we can model that as any number between minus one and one, but two one being the right-wing minus one being the left-wing. So then that's one example of how you prioritize people right? Because you wanted to this rigorously and mathematically. So we need a method to represent people and our model is at. ISSUE, have this in which you represent urine. You couldn't space in some point. Is You your opinion and you vote for someone who similar to? Let's say you're very right wing than some candidate comes in then you're more likely to vote for someone who is right being than left-wing to someone who is closer to you in opinion or characteristic that's our model of but this question was asked actually before in one emissions but in one dimension. So you're basically just have an inch between minus one and one, and this was an extra Unin by very famous. Nogal. Yes. Paper in two, thousand, sixteen, two, thousand, fourteen I forget. But we did this in higher dimensions in actually specifically two dimensions because it introduces so much more complexity when you go up in a dimension, right so Francis, why do we need higher dimensions is because before you you're presenting people left and right but people are more complex than that. You can't just represent a person by just one number for instance you. have to use multiple features as people say. So you have this multiple characteristics of people and they become dimension. So we do this in two missions surprisingly the mathematical difficulty of asking this question John Huge becomes actually much more difficult in two dimension than one dish very interesting and I guess maybe the answers grounded in a lot of details. But is there an essence to what that challenges is it that there are more simulations or? Is there a complexity theory aspect of this? Why is a two dimension so much more difficult than one so for instance, if you look at random shape, you can characterize this voting as looking at something called random convex or some random shapes. The question bows down on understanding certain random shapes and here assuming that the candidates are appearing. Let's say uniformly at on interval you have existing group members and she accepts somebody then he becomes. Group member and you evolve. So the candidates are uniform at each time step and so to answer the question we knew looking at this sort of random judge shapes in one dimension there's only one shape is just an interval intimate conviction in one dimension is just an interval. Some number between attack could be minus zero point five to zero point one is, but let's say you go into the mission all of a sudden there's so Many different shapes so many different things second happen if the shape is convinced or even just not convex in two dimensions does different more complex shapes can get in one mission. You only have interested in two dimensions you have gone as you have China goals, you have something that approximates your face even that could be shade in two missions, but that's not going to happen in Wand mission
Creating Decentralized Artificial General Intelligence
"So Ben will start off by talking about the idea of centralized governance globally there's pros and cons. People argued that past a certain threshold of AI, we may require it for certain topics. We will certain topics we won't. When you think about centralizing, we're not centralizing a governance. What do you think those important distinctions are where do you stand? The funny thing is when you set it off with the phrase governance I, immediately was thinking about is doing the governance. Out. A hours that. I've been that is in the end. Where we're going to get to with the advanced technologies that humans are rolling out. Human beings are not gonna be able to coordinate human society. In effective way, we're GONNA need a gradual transition. APP powered. Governance of humans but. Together, they're certainly in the early stages we will need wise and judicious in agile human governance of of a is end because initially the is aren't generally intelligent enough. To govern themselves in the rather uses tools by humans. human institutions enduring during the government is are mainly being tools. I mean governance of AI, is mostly about governing how various humans and institutions are using a I, which is a very important in difficult problem right and then the really interesting thing will come in the transition between these two phases of justice got great like. So in the long run, which could just the couple of decades coming in the long run we're looking at a is doing governance yet. End In immediate term were looking at. Okay. What regulations do we make? Stop people from doing? Nasty things with the I in military Jordan positive uses of. But then in the intermediate stage between assume phases, you got a eiser gradually getting more and more autonomy gradually getting more and more general out right and how how the regulate things in that context against quite interesting and and Saga Roy and some countries, some legal systems are trying to take. Modest steps in that direction already, but there's a tremendous amount of on those and so if we think about the state of affairs today there are folks who say, well, you know when it comes to privacy of data, you know maybe that should be done within individual countries. Well, when it comes to you know win facial recognition can and cannot be used well, that's gotta stay within countries. There's other folks that argue that other things may be a lethal autonomous weapons or. Some facets of kind of human rights or maybe to some degree privacy fits in there. I know some people feel very strongly about facial recognition globally that there's some threshold where even with today's tool like a I ben that there should be some broader set of standards that humanity can kind of play by for the sake of our aggregate wellbeing in both peace and prosperity other people really think everything as far as today goes is a country decision in that stat would you agree with that of see certain thresholds where it does make sense for global standards to fit in or are we too early for that or is it never a good idea? Where to list. We have now in the commercial sphere is concern. Regulation of AI is not especially different from the regulation of many other types of of software or hardware tools. The boundary is is quite difficult to draw right like weather Cambridge Jan Letica with crunching people's facebook profiles know using basic statistical analysis in an excel spreadsheet or using a machine learning algorithm. It doesn't really matter on the dozen hundred of manipulation. Let people using their day that it's. It's the same thing right I mean. With face recognition I, mean whether you consider that a higher it is what it is. If you have something more complicated than sunlight recognizing who someone is from. From. There Gate up from the other people walking next to them or something I mean you may need more general intelligence directing those people from more indirect cues but in the end the. They had to go on regulatory issues are. About the same right it's more about the the pet optical were building. About the the degree of. Intelligence. That were embedding in different parts of it. I think once they is get more autonomy as agents and are you know making their own? Choices in the world without humans tightly in the loop. Then you get into a fundamentally different class of of ethical and end regulatory issues than we. You know we made some small engine number of years away from that. The commercially rollout software. We're not there yet right. We're talking about. Is. In terms of national versus global regulation or In US become state versus federal Allen, stay on a pragmatic basis. This is GONNA be national. Regulation. For the immediate future because the international community can't even regulate nuclear weapons effectively right in that. That's very clear. What is a nuclear weapon in what isn't a nuclear weapon? There's not a fuzzy dividing line there. And also, there's not a lot of humanitarian and lucrative commercial uses for nuclear weapons that are very very similar to the Nazis is one avoid right? So even in the silver clear case like that. The international community is doing a pretty bad job, but I mean in a case where the nasty things. are a few lines of code away from highly lucrative commercial things are highly beneficial humanitarian things I mean how? How. Is International Community? Impact is going to cope with that very very uselessly would be would be
Report: Algorithm question complicates TikTok sale
"Late last week China updated its export rules that might sound like a small thing but it could have a big impact on the tech industry. One of its changes focused on artificial intelligence technology including things like content recommendation algorithms. It just so happens that an advanced content recommendation algorithm is central to you GUSTA Tiktok. The Algorithm is a big reason why companies like Microsoft Oracle and others are so interested in buying the social media up the new rules from China could apply to any sale of TIKTOK US operations. So deal talks are in limbo and the impasse is raising the question what's TIKTOK WITHOUT ITS ALGORITHM Here to explain as a reporter. Aaron thanks for joining us. Things revved me. So what is it about tech talks algorithm that all these companies are interested in what is the do? It. Surfaces content you know I. It's. It's like any other sort of use suggest auger them. You you see on your facebook news feed or browsing instagram in the search function you know it's just it's just catering in tailoring content for you, but it's pretty remarkable the level of engagement that Tiktok as seen because of the algorithm and they they have they have different algorithms, free country tailored to different cultures and countries and. Regions and they have so many users engaged. They get so much data that they're able to really tune these algorithms for maximizing the kind of engagement they get, and so you know without it it's you know what is tiktok except a pretty basic video app it is really the core of the product. It is pretty essential and TIKTOK has a really big audience. Are they enough to keep some buyers interested on their own? Yes. So if the American companies were to buy, Tiktok with out the algorithm than what they get is is a pretty good video APP with a lot of users, and if the product doesn't work the same way, if it doesn't, it doesn't lead to that kind of crazy engagement they get now the algorithm powering it I think there's a real risk that this audience would leave the platform. Right. Now they've they've around one hundred, million monthly US users you know in these companies would have to develop their own out of the room but that would take quite a long time months possibly years and I think by then it'd be there'd be a big risk of these users leaving the platform, and of course, the stakes are pretty high. The trump administration has said that tiktok either has to sell the operations or face potential ban and. Talk is a lot less appealing to these companies without the algorithm and without its audience. So. Looking foward what's next for this deal? Yeah right now, it's kind of a above the company's heads. You know right now it's this sort of geopolitical tug of war between us and China both governments trying to weigh in on the deal and and decide what how this will be best for each country and the company's inside that country. So right now it's it's really kind of above them and it's waiting to see what the Chinese government will or will not allow to be sold and what? The trump administration deems appropriate insecure for Americans. So all these companies involved in this deal are any kind of wait and see mode in some ways. I mean. They're trying to negotiate these things but they really have to wait for these governments to decide what they'll allow to happen. Yeah and zooming out to that geopolitical level I as a huge engine behind development in right now, and it's the thing that's being restricted here by both the US and China. Why is that an and what could that mean for tech innovation the future? Well I. Think There's been certainly love commentary about the sort of Cold War, going on. In artificial intelligence is very central to that. Certainly, tiktok uses the artificial intelligence technique called deep learning to develop their algorithms are so the results in so much engagement and these same serve techniques a deep learning are being deployed and weapon system. So so whoever has access or the the ability to develop some the best says Systems here I think has a huge competitive advantage. So both sides are trying to keep a very close eye on who has access to that and how it's how it's deployed and and you know China doesn't want to give away the keys to the algorithm especially around Tiktok, which has been there really big global success. One of their first big global successes in the Texan
Voting Mechanisms And AI
"Steven Hi I'm professor RTP OF MATHEMATICS AT USC University of Southern California Excellent, and tell me a little bit about your general interest within mathematics. Before we get into the particular paper, I wanted to talk to you about a few different topics generally speaking probability probability generally construed its relation to computer science in particular theoretical computer science. Would we wind up somewhere near what is it Polynot mealtime? Generators. I don't know about a number. Generous. Followed meal time things more specifically clavo problem that can't be solved in polynomial time, and then you WANNA approximated solution in USA. How well can approximate? How can I prove that? This is the best. You can do things like that under the general category of hardness of approximation suppose why knowing lot of those cases you have one benefit may be many but benefiting a lot of problems. Like that is you can tell if a solution is valid or you have some function you're trying to optimize for I. Don't know if the same is true in voting. Is there a global way that we'd all agree that the outcomes are good or the processes? Good. Maybe that's a good way to get into your topic designing stable elections. Exactly. I mean there's a lot of A. Link to Wikipedia Pedia page somewhere it's a table and it has a list of desirable properties voting methods and there's at least maybe ten or twenty cents properties and it's impossible to have all the desirable properties no matter which voting method you have there's always gonna be some that has some that a dozen but the one property that myself and many other people who focus is how can the voting method be protected from corruption and that could be mostly what people in this community of worked on is looking at random vote corruption. So everybody cast their vote and then Tyson with some small probability they will randomly. Change some votes, and then the question is which method best preserves the election's outcome. So that's the quantity that you want to say maximize. You want to maximize the probability that the voting method preserves the outcome. When you compare the original outcome to the outcome after the votes have been corrupted one quantity, you can try to maximize very interesting. I definitely want to come back and talk more about corruption but you've got me intrigued with those properties and I know there's many of them may be I don't WanNA put your memory test, but could you talk about one or two and maybe discuss you know a Controversy around them or why they're important that sort of thing. Yes. There's a bunch one desirable property of voting method is that it doesn't succumb to the spoiler effect as we know, the Electoral College does. So how can we think about this spoiler effect the main let's change the names to some ancient name. So we don't have to deal with political of discussion in the moment, but let's say we ran election whatever two hundred whatever years ago, and there's George Washington running, and there's also a clone of George Washington running as candidates and I some third candidate on the. Fact factor means the fact that two of these clone George Washington running while people who originally if there just one George Washington, they'll just vote for that first one but two of them you'd imagine you know the original George Washington Supporters A. Vote First Josh Attendance on my vote for Evil George Washington or whatever you call the second one. So the fact that the original supporters of the person gets split between the two separate candidates we know an electoral college that means that it decreases the chance of either one of them winning, and for example, I think a last election cycle Bloomberg said, I'm not gonNA run as an independent because of this effect, you can steal votes away from someone in a sense and it can. Ruin the chance of say some candidate that may be you kind of support or something. So that's a desirable property of voting method that some of them have and some of them don't doesn't have the spoiler effect. That's that's what you are. There mechanisms than the can eliminate things like that. How do we build something like that into voting framework one voting method that avoids this it's become popular to certain people you know on the perfect voting with, but it's called instant runoff voting so. Different than what we're used to thinking about your vote is no longer just your favorite candidate. It's like a ranked list of candidates like for example on. Once going back two hundred years or something maybe your first choice most preferred candidates George Washington may be your second most preferred candidate is out in. Alexandria. Hamilton third most preferred candidate Harriet Tubman or something I don't know every single person makes list of preferred candidates and they all get submitted into whatever the election methods, which is song way of taking all those votes and just saying, okay, here's the winner and so one I think mentioned already one. Popular ranked choice voting method is called instant runoff voting on I believe it's used in Australia might even be used this coming election cycle in I'm not entirely sure but anyway so the important thing is this voting method does not have these spoiler effect
How to Use Calculus to Design Learning Machines
"An. Important key idea for many machine learning algorithms is the concept of gradient descent. The basic idea is that suppose we have a machine learning algorithm whose performance can be measured in terms of performance function. An example of a performance function might the prediction error of the learning machine? Learning machines prediction error depends upon the training stimuli that we used to train the learning machine, but it also depends upon the parameters of the learning machine. As previously noted, the parameters of learning machine are justed during the learning process with the goal of improving the learning machines performance. Note that the prediction era of the learning machine is essentially a function of its parameter values. If the learning machine has a good set of parameter values than it will have a low prediction error if learning machine has a bad set of prominent values that will have a high prediction error. So, this is where the concept of great sent comes in. Our goal is to come up with a learning algorithm which adjust the parameters of learning machine at each time. The learning machine has an experienced in its environment in such a way that the. Prediction error of the learning machine decreases a little bit. And this is done by an approach called gradient descent. It, turns out you can prove the following important key theorem. Suppose that we make a very small perturbation to all of the learning machine parameters and specifically suppose this perturbation. That we add to a particular parameter, value is defined as some negative number multiplied by the derivative. Of. Mode applied by the derivative of the prediction error with respect to that parameter value. It can be shown that this method of updating the parameters has the property that the prediction error for the. Parameter values will always be a little smaller than the prediction errors for the current parameter values. This is called the method of gradient descent. and. This method of updating the parameters of a learning machine so that it's prediction error performance improves is a key idea for many machine learning algorithms. The concept of great upset was first introduced in episode sixty, five of learning machines WANNA. One. Now at episode twenty, three of learning machines went to when we introduce the concept of deep learning neural networks in a feet forward deep learning neural network. A unit is essentially a function which computes estate value from its pramod values and the states of the units which is connected. Each day value is represented as a number. Intuitively, we can kind of think of the state of a unit as analogous to the degree of activity of neurons, perhaps inspiring frequency and the parameters specify the degree to which the activity of one on influences another. Of course, the next challenge is the method for how do we compute these derivatives to implement a gradient set our them in many machine learning applications. The prediction error is a very complicated function of the parameter values. So it's not immediately obvious how to compute the necessary derivatives. In fact, even computing, just one of the derivatives can be extremely challenging. Yet, we often need to compute many such derivatives.
How to Use Calculus to Design Learning Machines
"Before beginning our discussion, it would be helpful if you're not familiar with vectors matrices to review episode eighty eighty-two Learning Machines WanNa win. An. Important key idea for many machine learning algorithms is the concept of gradient descent. The basic idea is that suppose we have a machine learning algorithm whose performance can be measured in terms of performance function. An example of a performance function might the prediction error of the learning machine? Learning machines prediction error depends upon the training stimuli that we used to train the learning machine, but it also depends upon the parameters of the learning machine. As previously noted, the parameters of learning machine are justed during the learning process with the goal of improving the learning machines performance. Note that the prediction era of the learning machine is essentially a function of its parameter values. If the learning machine has a good set of parameter values than it will have a low prediction error if learning machine has a bad set of prominent values that will have a high prediction error. So, this is where the concept of great sent comes in. Our goal is to come up with a learning algorithm which adjust the parameters of learning machine at each time. The learning machine has an experienced in its environment in such a way that the. Prediction error of the learning machine decreases a little bit. And this is done by an approach called gradient descent. It, turns out you can prove the following important key theorem. Suppose that we make a very small perturbation to all of the learning machine parameters and specifically suppose this perturbation. That we add to a particular parameter, value is defined as some negative number multiplied by the derivative. Of. Mode applied by the derivative of the prediction error with respect to that parameter value. It can be shown that this method of updating the parameters has the property that the prediction error for the. Parameter values will always be a little smaller than the prediction errors for the current parameter values. This is called the method of gradient descent. and. This method of updating the parameters of a learning machine so that it's prediction error performance improves is a key idea for many machine learning algorithms. The concept of great upset was first introduced in episode sixty, five of learning machines WANNA. One. Now at episode twenty, three of learning machines went to when we introduce the concept of deep learning neural networks in a feet forward deep learning neural network. A unit is essentially a function which computes estate value from its pramod values and the states of the units which is connected. Each day value is represented as a number. Intuitively, we can kind of think of the state of a unit as analogous to the degree of activity of neurons, perhaps inspiring frequency and the parameters specify the degree to which the activity of one on influences another. Of course, the next challenge is the method for how do we compute these derivatives to implement a gradient set our them
Apple's iOS 14 privacy settings will tank ad targeting business, Facebook warns
"Earlier this year apple released a Beta version of its new operating system. IOS Fourteen contains new privacy features that make it harder for companies to track users from APP TO APP yesterday, facebook said, the changes would cripple its audience network business. That's the part facebook that helps it serve targeted ads to facebook users when they're using outside APPs reporter Jeff Horowitz says, this is just the latest clash between the two tech giants. It fundamentally stems very different business models, facebook sales. ADS and they collect as much data as possible and from as many sources as possible apple historically at least sells devices. That's kind of a more focused on the immediate user obviously, and those different interests and different business models. Kind of conflict in which facebook generally wants the free flow of data and apple is I think known for being fairly stingy on that front, which many of its users will tell you exactly what they want that for based on privacy concerns.
What AI Readiness Really Means - with Tim Estes of Digital Reasoning
"So, Tim will kick things off and get your perspective on what Ai Reading this means. When an enterprise says walked, we want to become a I ready. We want to start using a I what kind of components have to go into that? Yeah, well, I think the first thing is you had to have infrastructure that sounds so basic especially with the cloud bud, the larger enterprises a requires a good functioning process to allocate infrastructure with their on premise or cloud. And then data governance of data can be used for training and validation around any process it's going to be tested. So it's all too often that you know one group in enterprise wants to try something. The aren't really the owners of the data that is required. To validate what they want to try. And they are not the suppliers of the infrastructure. So you might run into a substantial gap. The could take you know a sixty day or thirty day pilot. Or PSE and make it a nine month process because you're waiting on them to sort out data governance and infrastructure availability. So those are two pieces you know something about the education side of it. In terms of you know this this dictation you want to build and educate yourself to understand the difference between certain techniques, but it's always overalled because in the end of the day I I'm a little bit more pragmatic I think there's certain techniques which are better for. Some things and others, but obviously, the most sexy technique that talked about the time or different variations of deep learning. Yeah and we could go into braces but the phillies he'll is in most cases, the customer doesn't have the data sets available to train a really good deep learning class fire and so or an engine of some kind. So I I think that what you find actually is it's not just that they have data general had dated is prepared a certain way. Often to teach a machine that the machine can perform the task and that's really the that's the area. So these maybe the this question you elite in some other things but you know basics, infrastructure data governance I can pull they need to run the test fast and as a as a vendor or someone the outside I mean I would becoming in asking these questions now because I lived through being wishful thinking and. This is really exciting CTO and they want to do this and they have a business stakeholder that wants to do this kind of application. They think we're the answer I've been through that whole dance, and then you find out that of the whole dance that dance might take months four and then you wait nine months for data and infrastructure to be available in the large bank. Yeah. Well, not surprising at all within a large bank you're lucky it's not a eighteen months or something. So you bring up infrastructure you're bringing up data does this mean in the process of speaking to whoever your initial? Champion as your your initial kind of point of contact who you think is GonNa either signed the Checker help sign the check the you really have to be clear that sort of what infrastructure you need to access of what kinds of data you'll need to access of the state of that data with that person and or with whoever they need to rope in serve as part of the process of working to a pilot. So like doing that diagnostic I, guess as you go as you progress forward. Yeah that's right. So I mean I'm naturally gonNA give it more from the vendor sides, of course. But if I flipped the hats and I'm I'm actually in the buyers persona what I wouldn't WanNa do is the last thing I want to do is to put a lot of energy into something that could create real value get excited marketed internally, and then find out that getting infrastructure having data governance process in place where we can get the data necessary to test the system is not really well figured out or is figuring out but the restrictions that make this not work. So I, think that there's a good upfront investment in that but there's a difference between that and sort of what I might call the the Data Lake Panacea. We're everyone wants to have this. Highly Organized Library of data with the Dewey. Decimal system in their enterprise. And that's not gonNA. Prize is unfortunately function. So many it efforts in an enterprise are responsive the business as a higher priority to eating across business lines. That you'll almost never find as you will a pristine data infrastructure. So you really WanNa make sure the process to pull data, put it in the compute environment do that safely and would security sign all's that should be enough to get moving, and so I think if you try to go four steps beyond that. You have much bigger challenge and essentially trying to boil the ocean and I think a lot of people went down that road with all the WHO do vendors to be blood. You know the idea that just got to spend all this money on that and then from that. You end up having all this application. These applications become so easy and here we are five years later it must. We're seeing what applications besides restoring my you know might my loan scores or some other batch structure process? You could probably done some other way.
The Evolution of AI Chips and Their Business Impact - with Dr. Gregory Diamos
"So. Brag. I want to kick things off by talking about this theme of a I chips kind of term being thrown around now around software serve built to purpose for ai he gives a bit of a background on what they are and maybe even where they're starting to move forward today. Sure. So if we look historically on ships, there's a few trends that are really important One really powerful tragedy in computing has just been personal computing and especially driven by the C language. So in in the past, actually, you'll probably aware everybody these days has personal computers interact with computers. They're probably aware that computers have gotten substantially faster. Since they were introduced. If we look back fifty years, the speed of computers for the types of applications that we know in love the. Power personal computers and power the Internet have gotten tremendously faster. So can actually measure this. there's a trend around. Around nineteen eighty driven by DISA-, Group of people who are early founders and early manufacturers computers. To try to quantify the speed of computers, one of the things that they developed at that time was called the stack benchmark, and this is a way of determining really objective. How fast is your computer? Suspect started around. One, thousand, nine, hundred, five, a little bit early though not a really danced around nineteen, ninety five. and. This was a metric that. Anybody was producing a computer You could runs back on it and it would give you an idea of how fast it was. So, if we look at stock performance going around nineteen, hundred, five, about two, thousand, ten, it increased basically exponentially over that thirty year run I'm so we saw this enormous over a thousand times improvement in the speed of those computers. From around nine, hundred, eighty, five, two, thousand, ten. So for for various reasons, actually after that it's it's still been increasing, but it's slowed down a little bit. So we have been seeing the same. Performance increases out of our personal computers out of our phones out of the computers that are powering are our data centers. As we have in the past. So the the modern view, those those machines like the current name that we usually call them in industry is is called a CPU. CPU sometimes dance for Central Processing Unit really I think about it as a as a computer that's designed to run a specific type of application like an in particular it's designed to run kinds of applications that we build an in languages like C or or Java, or. Running on operating systems like windows or Android or us. So those computers in those chips are powering quite a lot of the workloads, quite a lot of the applications or the the computing technologies that have been transformative in personal computing in in smartphones in in in data centers. And they're connected the world through the Internet. So, ai chips are actually something different. And There's actually a convergence of maybe two forces here. That are driving this and one is the the forest I just mentioned that. You. Know as we look at the existing designs. For various reasons sometimes people one of the reasons people often referred to as the death of Moore's law whether or not. It's actually dying is is actually a highly contentious topic, but the death of Moore's law is actually one force. That's. Slowing down progress in in Cebu speeded efficiency Almond CPU development. There's some other forces to one fourth that sits a little bit technical. It's related to it's usually called Denard. Scaling. This is just the idea that it's it's getting harder and harder to make computers consume less energy. Some of the technologies that have been driving reductions energy are kind running out of steam. So for various reasons, actually progress in CBS has been slowing down not stopping but but slowing down since about two thousand ten, actually a little before that. Is, really showing up an industry you know after about two thousand ten. In around that time, a lot of people including myself were working on this. It seemed like a really hard problem. You Know How do we actually? Continue the enormous improvements we see in computing. Face with all of these challenges, there's actually a number of people in the industry who call. They're all of these walls that you'd run into and you know. So there is a joke in the industry that we basically run into a brick wall at this point. There's no way around this wall or it's it's really hard for your the leading experts in the world to figure out what to do next. You mean like n Hey, Moore's law. Hey as denard factor that all of these are just kind of like physical barriers that you couldn't surmount Yes Yeah it's hard to say like actually You know who? Maybe there will be a breakthrough in the future but in a lot of people spend a lot of really smart people spend a Lotta time young. Yeah. Yeah. And we haven't been able to find a way around young tough stuff. Yeah. So Luckily, at the same time, we figured out how to get a it to work. And that was so fortunate because the type of a sometimes called machine learning I'm just GONNA call it a I. The type of a technology that started to work really well, sometimes people called his deep learning one of the driving technologies is. The back propagation Algorithm in neural networks. And this technology is wonderful for computers. You know if you want to build a computer like a CPU that's good at running windows or running. The android operating system on your phone. Is Hard, you know the CPU. Is the best way that we know how to do that. But if you want to run something else. So if you want to run an AI algorithm like neural network if you want to train with back propagation. You can do something different you can do something completely different. And that opens up so many more opportunities to continue to improve performance and so you know since two thousand ten. As some companies have started to invest in these technologies I. Think. Actually. GP Use especially GP's from. Our an early era of investment. we've seen enormous gains like we've seen 10x gains are actually in terms of real applications, not not all real application running real applications actually seen a hundred times gains. In performance
The Injustices of AI
"You are about to meet to Gutsy. multi-award-winning film directors with stories that connecting contrast to incredible confronting crucial films where artificial intelligence is being used in the service of good bad and possibly plein rotten. Lucia terrorism. Sleeve those. WHO ITS MUSCLE IS INSERTED GAVE David France director of the documentary. Welcome to Chechnya went underground to document the current persecution imprisonment, torture murder of lgbt people in Chechnya and the. Going rescue mission to get them out to safety. Kid Dot as facial recognition mis identification, and then you start. To search, this is an innocent child. System is becoming mechanized shall nation Thanh Director of coded bias documents the rise of this called Algorithm, ick Justice League. There are fledgling movement with this mission to rescue us from the insidious crepe of biased computer algorithms into pretty much every aspect of our lives. Now, they films both feature at this year's Melbourne International Film Festival and they join me from a Balmy summer in New York. City thank you for having us. Absolutely thanks for having us the injustices in these films. Real and raw and happening in the world right now, and there's this. Shopping, and shocking sense of urgency in both of these films for both of us. Why are these films that you were compelled to make? Now what drove you to these stories David? This is a story about an ongoing genocidal program in the southern Russian republic of Chechnya a conflict which we were all informed about in a series of articles published in a newspaper. In Russia back in two thousand, seventeen, it produce headlines around the globe. It was a horrifying revelation of a campaign to round up and eliminate all lgbtq people living in the. Chechen republic it generated our government leaders around the world were outraged by it and demanded justice. But the story immediately fell from headlines and what I discovered some months later was that it the the crimes themselves had not stopped. And in fact that ordinary Russians were responding to this in really heroic ways I spent eighteen months embedded in his underground network this underground railroad of people who were actually physically rescuing individuals from. Hiding them. Tending to their physical wounds in their psychological ones and trying to get them out of Chechnya if and to relative safety and other parts of the world the access you got. WAS INCREDIBLE THE TRUST In Janet easing credible Shalini what about you boss is this absolutely riveting vital interrogation of the wine which machine learning algorithms are effectively shaping lives in the most potent and yet most secret ways. Why will you compelled to make this film? Well, I think all of my work deals with how disruptive technology makes the world lesser more fair. And when I stumbled upon the work of joy Leney and Cathy O'Neil the author weapons of mass destruction I sort of stumbled down the rabbit hole of the dark side of big tack and came to realize quite shockingly. that. You know these computer systems that we give our implicit trust to and entrust with such decisions like who gets hired, who gets health care how long a prison sentence on someone may serve. have. Not Been often vetted for accuracy or for racial or gender bias, and that comes across in the making of this still where. Joy Leney is just trying to make something like a snap chat filter were right? And put a mask on her face and stumbles upon the fact that commercially available facial recognition doesn't see dark faces are women accurately she's
Forging International Consensus About the Future of Intelligence - with Jerome Glenn
"So Jerome got a lot to talk about here in terms of artificial intelligence governance, artificial general intelligence. The reason I think this conversation will be fun is because you've thought through some future scenarios with with very large organizations for many years. Very High Level and you've learned a lot in the process of what is the process for pulling together different stakeholders imagining, what will the future be? What should we do I? Mean very complicated. You go about it. Of course, one of the first things you do is you gotta find out the state of the art of whatever it is you know is there is, let's say five elements to it or ten elements, and you know was the state of the art on this element on this element, this element, this element. Now myself I won't know enough to do that. So we have a global network of networks sixty five nodes. Return Network himself within countries, and so I can say, here's where we are so far and they tell me what else ought to be considered. So there's so as global sort of a state of the art assessment finger. Yeah and then within that with take a look and say what questions were not asked the authorities have been asked. and. What questions were as but answered, superficial. That gives us questions to as in a Delphi study, which is a questionnaire goes around the world. And the results of that then becomes guts content to create draft scenarios. We send address narrows back out and everybody hasn't at Pat and presides over, and then we can say, okay, what do you do about this scenario? What did you do about it? You'll see a good action as well as scenarios this sort of a general approach So you talked about the Delphi study I actually recall you bringing this up the first time you and I chatted I don't remember who has five years ago or something wild like that. Speak briefly about wooded Delphi study is so I like finger on the pulse what are we missing? Pulling, those ideas together and then there's this kind of dispersion to generate even more. What is the Delphi study. Delphi questionnaire. Whose second round. Is. Determined by the results of the first round. And third round is determined by the results the second route. the reason for it was that there were generals and admirals and experts that don't always the same room with each other at the Rand Corporation.
Reining in Complexity: Data Science & Future of AI/ML Businesses
"And I come from the physics computational physics background, and we both kind of been pushed into this. Data. All data science and I don't know if that is coincidence or if we have an affinity for them before we get into that though this kind of competing view of the world, which basically says sequel can do everything and this one we spent a lot of time actually looking at the data science or the data landscape and it feels like there's two worlds there's like the data warehouse maximalists. We'll stick all date on the data warehouse and then we're GonNa do sequel and then we're GonNa have some extensions to sequel like you see popping up like big query whatever and that can do everything needs to be done, and Oh by the way if someone's using python and are all, they're doing basic regressions and so we can just make that a simple extension and we're done and then there's the other view of the world I like to call the hoop refugees which is. Actually, we do hardcore computation and we need our in python because of stuff we do is very sophisticated. I I know you're squarely on one side of those but I wonder like, do you think there's a convergence that happens these stay two worlds does one become a relevant like what happens there just because you oppose extremism doesn't make you an extremist, right David? How's maximum? I see this world and it's the old yarn about I guess, I. Don't know there's there's so many variants of this but Allen Parlous a great computer scientists has some really great quotes about somebody Reverend sees about these kinds of things but I would say that to the idea that everything be expressing sequel. which sequel with how many extensions because the end of the day and with how many like extensions on extensions and multicolored on your post grass? Pipe, colonel, I guess you're doing the sequel, but you're running a python script knows let's not really don't count and frankly a lot of stuff runs access and BB in this world isn't sequel I think if you choose to look at the world through a particular Lens, you can choose to count everything else as. And rounding errors. But if you take off those lands as you see a much more diverse landscape and I think that's where for me, I see the space for sequel I understand the reasons why it has evolved into particular kind of animal like the shark is still the best predatory fish in the ocean. But it's not the biggest Predator in the world and I think there's something about that that if you're in the ocean, you're going basically shark like if you're to eat a lot of fish. So if you're in that business data analytics world especially because a lot of business data looks like fish, it's evolved look like food for the sharks. So that's kind of the way it is but what Hoodoo opened. Up Back in twenty twelve I called it the hoodoo battering Ram as we're not GonNa win the Hoop Gate. We'll let the vendors go and fight against the Tara's the oracles, all the classical data warehouse guys let them do that thing. Once he battered down the door, we're GONNA come flooding in with all sorts of heterogenous approaches to data science analytics, things that are hard to ask in sequel. And Moreover, there's a term use which I don't hear used very often now obviously or the shadow it, which is used it, but there's a shadow data management that's a are far more serious and dangerous problem when I was at an messing bank, they had a million dollar oracle database sitting somewhere and it was too slow to actually run the analytics they needed, and so what they did is they had an instance of this oracle database cost a million bucks and what they did is the only quarry they ran was a bit full table dump into a CS meet and then they took CSV. And they did everything else with it and it was python scripts Ram Java. Crap was bunch of other stuff and it was sort of like. So if you're a data manager, if you're like in the data management practice, you say, well, we have another big old million dollar instance stood up our data management techniques are great Potemkin village against right. But then when you actually go and you ask the developers, Hey, where's the source data for the stuff whereas prod data coming from? Like, oh. Yeah. This file share back slash black slash something or the other or you know that I'll I'm like that file what about database don't touch the database to Brennan right so there's this kind of stuff going on in everybody listens knows what I'm talking about that shadow dated management is absolutely a pernicious problem and data science is just eating it alive because did ask the question you want to ask you have to integrate us together master data management is about silence And all this stuff you've hit to the site which I just think it's so Germane to what we're here to talk about, which is this clearly problem domains which sequels totally fine for right Yep and you can argue the problem domains, which is just not any sort of hardcore statistics is just not very good for and the point of us being on this podcast is actually talk about like listen we're kind of new types of companies and you types of workloads. And they're around kind of processing data and I hear you that this shadow data management is a real issue and you can make an argument why that exists because people are stupid or they don't WanNa do good workflows is like literally we don't have the tooling to deal with data in the right way. One question that I have that I would love to hash out with you is are we a fundamental shift in workload that requires a fundamentally new set of tools? And a fundamentally new type of company or is this just more of a transition where we can kind of put into service the old tools and I just want to be a little bit more specific, which has in the past you had your toolkit of systems approaches and you have software system, and you'd kind of pull them out and applied to the problem the sequels, one of them, and we kind of understood how systems behavior and we kind of understood how the company's both around the behaves. As an investor looking at a lot of data companies, they just don't look the same types of tools they use the type of operational practice they use. The one that you pointed out was a great one, which is outdated becomes a primitive what actually apply like software techniques to in a way, but we don't have the tools to do that. and. Then we've written posts about margin structures look a lot different where you go at your company different and so I just do you think this mess is because data scientists don't have formal S-. Yes trainings or do you think this is an entirely different problem domain and we should actually look at what the future looks like for that and development. Tools, etc. This is at the heart of Oprah talking about this is absolutely the heart and I will try to start from the top, which is this concept that every baby or every child is born and the reason that they think their childhood is normal right? They think of like your childhood like normal thing. So have developers coming online in late two. Thousands let's say and they think this is the world even me as a professional starting in ninety nine while this was just what there is the more you start researching history and looking back your life. You know what? We're just building in this industry, which is layer it's frozen accident on top of frozen accident frozen accident very very few times do. People. Make principled intentional revolutionary shifts. Right? It's. He basically band-aid a substrate. Okay. So starting from the top, what I would say is that there is no law there was nothing hard in stone that Moses brought down from the mountain said all information systems must be deconstructed into hardware and software and data there's no such thing it was information systems will stop. The fact that we had different cost structures for innovation in hardware versus software versus networking, and so forth. That has led to different rates of innovation different places, things like that, and so when a business steps in and says, okay, what's on the shelf that I can use to accelerate my business processes? Then it makes sense because this thing that thing like when you buy a car, you buy the car and then you put CDs in the car you don't go by car with a CD prefect
Using Pilot Projects to Set Enterprise Expectations
"So back will kick things off talking about the components of making a business case for a I in the enterprise. A lot that goes into you know, should we deploy this? Should we adopt this? What do you think of as those component parts? What makes that a? Business case is clear if the I serving the outcomes that you're interested in. and. Therefore, I think the business case should be aligned with your own business case if you can review where it is that you are optimizing various different outcomes that's what is all about it's optimizing outcome. So that alignment should be there. Now, a lot of what we see in ai in various different businesses is sort of okay. We've heard good things about ai being able to classify images or do translation or whatever. Let's just incrementally look at places within our business where this might help. That's one way of looking at it but the way I would recommend looking at it is is in fact to work back from how you imagine. Ai Impacting. Your Business. Until. There are both sides of that coin. There is the hey, take your existing. You know, what are your key thrusts? Where's your money in your attention going to executives care about, and let's see where I alliance with those and let's pick projects. Can we move the needle on the same outcome? Okay that's good. That's that's one side. The other side is to your point maybe bringing into the strategy room. Hey, look here's the representative use cases of this new capability. It's important. Here's what it could do your. Maybe we should find some errors we can build his capability that way you recommend. Really let's hit the key business dead on. I think so I think. So I mean it's the difference between thinking of it as being incremental versus thinking of it as being disruptive and I should be destructive. That's the our our tool. So why not use it and expect from it? The outcomes that were were seeking from the business. To not have the toy application not all kind of thinking about where to say I fit questions are toys but some are like we're going to do it in a corner and then we'll say we're doing we're saying is you know where is your muscle headed at? Can we just drive ai behind that to go for that outcome? So for you making the business cases about just having your defined outcomes I think a lot of folks would say that what else are thinking about if I'm an executive at a big company. And I've got to make the call. Okay. We can go ahead with this. What else do I need to see any way to know other people have done it? It's worked. Do I need to understand conceptually what it does do I need to know these the underpinning capabilities need to build in order to enable it what goes into so I get behind my Keith? Ross what else is there Absolutely I think it's very important for you as a business to consider this as a differentiator. If we're talking disruption, that's what you want to own, and if you want to own it, you have to bring something special to it. You can't just be using it as an out of the box tool. You have to bring in the domain expertise that you have, which includes the definition of the outcome, but it also includes how you would actually implemented how you would deploy it and how you would benefit from it, and the onus is on you the business person you know that the business the enterprise to make that definition looking at others saying, well, you know other folks are using it in this manner and therefore I should probably be able to use the same way I think doesn't quite work. Any what ends up happening is Oh you know a lot of people are using a I to find more insights in their data. Let me see if I can find more insights in mind data. Let's just unleash it on the date and see what gives us. Well, that's like shirking the responsibility of defining what you're looking for, and why, why, what, what would you use that insight four and how would you know that what you have actually derived is going to be useful for you or not so to me, there's a piece of this. I was actually thinking this just recently where a lot of the business models have moved to being look, there's a very special kind of unique data going. Through my enterprise that data is my goals, you know if I if I own it, if I owned the data that goes through it, then you know that I can monetize it. So that seems to be an established fact I. think we need to start thinking a one step beyond that and saying that data in of itself is useless. It's what you make of the data in other words, the models that you build on top of that date on the way, you actually make use of it that is going to end up being your differentiator and in order for you to be able to monetize that you need to actually capture it in the form that is useful for you and your. Niche your special place in the world of business. Actually, if you get to that point, the data itself becomes less interesting It's important that it's piping through your system, but it's the abstractions that you're a is actually making off. That's that's really the goal
Groundbreaking New Material 'Could Allow Artificial Intelligence to Merge With the Human Brain'
"Groundbreaking new material could allow artificial intelligence to merge with the human brain. Well, that's what Elon Musk has been working on with the stuff. This is not directly connected to that although I suppose he could use this material in his product. Scientists have discovered a groundbreaking bio synthetic material that they claim can be used to merge artificial intelligence with the human brain. Previously has been hard to use the typical stuff you would use for some sort of a cable or wire things like gold, silicon and steal. They can cause scarring when they're implanted into your mind to turn you into a Cyborg because you're not made of metal. So, unfortunate, your made a guts and skin and stuff. Good old fashioned biology data. How dare you gooey stuff the gooey stuff The meat sack. What they say that's you full of meat. Yeah. So obviously, we're outdated and we have to get upgraded. We gotta get upgraded and so in order to do that, they're going to use. A type of material we got the idea for this project because we were trying to interface rigid organic micro electrodes with the brain, but brains are made out of organic salty live materials. Those pesky brains. pesky brains it wasn't working. Well, we thought there must be a better way we started looking at organic electronic materials like conjugated polymers that were being used in non biological devices. We found chemically stable example that was sold commercially as an anti-static coating for electronic displays. This polymers known as P. Dot. Org pedo maybe it's a French philosopher by the name of bidault. Sounds appropriate or dot. Like t dot no says t dot. Back in the day ever that the that didn't last that long or did it. Did. Never. Really. was never really the choice in my circles no. T dot in order to reference the city of Toronto was never a popular choice in my circles. Was it a popular choice in your circles? No no, it's just all you do is say Toronto Toronto without the T, at the end. Yeah it's all you have to do. Anyway show Toronto Raptors because when we checked up on the game, they were up like thirty they're up like thirty. But who knows probably since we started filming this because we were like that gave is done and then probably now Brooklyn Brooklyn came all the way back I. Doubt it. Not, against those raptors. This versatile I it'll polymer was recently discovered to be capable of transforming house breaks into energy storage units due to its ability to penetrate porous materials and conduct electricity your brain, your porous gooey brain needs to be penetrated, and this material will allow that to happen. So you can be upgraded. Understand Down for that. Okay. You'RE GONNA be number one. LISTEN UP LISTEN UP Ilan. Wills going to be the first guy on on your brain implant assembly line just stick them in a chair put him through the conveyor. And upgrade that man. Meet Suck upgrade the man.