data.bythebay.io: Panel: Artificial Intelligence: Who is Winning?Panel d4
Recording: data.bythebay.io: Panel: Artificial Intelligence: Who is Winning?Panel d4
uh before the panel we conferred and and some of the panelists expressly disallowed certain topics which I will mention otherwise we're not going to talk about building data teams we're not going to talk about data Lakes except I just want to ask one uh thing uh so a year ago I asked pit P pit is kind kind of WR the notion of data scientists so I asked him is it okay for a legitimate data scientist to say data Lake and pit said literally I quote probably not that was one year ago I just want to ask before we put this question to rest is it still uh not okay to say it or is it kind of becoming inevitable like Donald Trump the data Lake like we're going to be stuck with I I think oh I I'm trying to keep it safe for work here uh yeah it's kind of the equivalent of like uh go jump in a lake okay if you say that yeah go go get a data Lake and come back to me only ours reservoirs and rivers lakes with reservoirs data reservoirs okay data ocean is it Okay no Okay well there was the lame thing about narwhals being unicorns from the data ocean that I was trying to popularize unsuccessfully so that's may change things this is beautiful uh so uh probably no no data Lakes um and uh so we have four great panelists let them introduce themselves starting with Richard hello everybody I'm Richard I uh graduated two years ago from Stanford the PHD in uh deep learning for natural language processing and computer vision then uh started a little startup called metamind and we just recently got acquired by Salesforce and now I'm the chief scientist at Salesforce hi I'm Shon zillis I'm a partner and founding member of Bloomberg beta which is an early stage Venture Capital firm and so our firm exist to invest in companies that are transforming the future of work and So within that the lens that I focus on is anything data and machine learning oriented and so just to give you a little bit of a flavor of the types of things that we're really excited about and invest in um within this data machine learning World there are three categories so the first and this was the one we started investing in was tools for data scientists and machine learning practitioners so we've got about four investments in that space covering different areas of that uh the second and this is the one we're most active in is U machine intelligence for the knowledge worker so basically cutting across every Enterprise function um the majority of these tools end up using Text data as an input but we've also made investments in in companies that are primarily using audio and visual data as well and then the last and this is the one we think is is emerging and we're hoping to see more activity in uh is machine intelligence for various different vertical Industries and so so far we've made two investments in agriculture one in healthcare and one in education and you know we're we're really hoping to see you know healthc care is heating up very significantly ly um we're hoping to see more things in cities uh I'm personally obsessed with Elderly Care and what machine intelligence can do to help the plight of you know our aging populations um and then very finally you know just more on a on a Founder level you know we're a seed-stage investor we love helping founders with early go to market fit and so we tend to focus on companies that are problem first and so you know just an example of that is we're an investor in a company called Tuli Technologies which is in the agricultural space it's basically data to help farmers make better irrigation decisions so they started out with just the problem of you know how do I help farmers maximize crop yields while minimizing water because as we know you know in California especially water is an increasingly large issue so problem first is is sort of ouro hi uh I'm Pete scok and uh thanks for the introduction uh pre-introduction but uh yes I was uh data scientist at LinkedIn I was an early member of the data team and uh helped my particular Focus was around building data products so I'm kind of at that intersection between uh machine intelligence and actually applying it uh to products that people use in the real world um while I was at LinkedIn I think it grew from uh 30 million to over 300 million uh users um and uh all of those members on LinkedIn were doing things like updating their profile creating their public Persona and building their networks so that you could see who they're connected to and find job opportunities um business deals things like that uh particularly things like uh LinkedIn skills and endorsements um if you love those uh you can blame me if you hate them you can blame me after um but uh more in particular uh you know I I think a lot about um using data and algorithms to help make um work more efficient and to help people be more effective in what they do um and I think it's not just doing more of the same things faster but actually uh by making the connections that you can make with AI you can actually enable new things that you couldn't do before um and maybe that's something we can talk about later today so now I'm working on putting that into practice uh with my own startup um so I have a new uh stealth startup unfortunately still stealth but um what's your domain I actually so uh little little joke on my part I put down stealth. a uh on my uh data by the bay profile and I got the domain and put it up just just for you uh so we have uh something to point to um but no seriously I I'll be talking more about this in uh in in in the near future uh but I I will say that I'm thinking about how can we make knowledge workers more more effective congratulations um what is still our panel as um as always um Alexi tries to bring me onto his panels um and Surround Me with Super Genius so it's always fun I'm Shri I'm uh founder of U the H2O AI community and um we've been building a Grassroots community of AI and math movement in many ways bigger algorithmics as a core thesis uh data products is what we are building uh some of the pieces you're doing now is more um how to make data scientists more efficient was first Core theme behind the company in fact how do we make data scientists happy um I think from there we've kind of now um trying to the next goal which is is kind of what Pete was mentioning in the data scientist at LinkedIn try to build data products right sort of so we now enabling data products in most of our customers about 5,000 companies use H2O today on a weekly basis 12200 of them on a daily basis and we are now seeing verticals um adopting AI in a rapid fashion so vertical is the new horizontal that's kind of Core theme um and data is the only vertical that you need to focus on so I think that's going of being a very common um uh thread amongst most of our customers custers so we're building a lot of tools for AI um our ultimate Vision really is what is um to make us more human using Ai and and towards that we're building a lot of more visual intelligence how do you coexist with a very Superior intelligence like AI how do you not treat them like a bot bot actually is actually a derogatory word in my word my view of the world you're probably looking at AI as being a substantial partner in the growth ahead and in the discovery that's really right ahead of us so it's one of those wonderful times to be in so thank you for having us uh thank you guys uh so the title of this panel was you know who is winning and it was a little bit factious but actually I think uh I need to provide some background uh uh in between the call for papers for this conference and the actual conference two of our presenting startups were acquired uh both by Salesforce right prediction I Simon chant talks about spark and machine learning as a service and uh uh Richard's metamind was acquired and Richard became Chief scientist of Salesforce so I think we can frame this question in a way we see an arms race right companies acquiring in you know AI startups with expertise and apparently they they want to win uh in a larger battle so can you talk about what is it they kind of fighting about where where the winning will be and why is such an urgency for established companies to acquire deep expertise in certain AI Fields I don't want to start all contrarian but I think uh I'm not a big fan of the title I think if anything the answer is humanity is winning because we don't have to do a bunch of boring work in the future and just like you know 200 years ago 90% of all people alive were working in the fields um that seems kind of silly now because we have tractors and other bigger machines and uh that will be there may be a short period where it's a tricky transition just uh like the Industrial Revolution um but in the end it will free up our time to do interesting work so I think overall if there was an answer it would be the answer would be Humanity in my in my eyes um now if there is some competition it's probably mostly over the AI Talent um it's not so much about the models because fortunately and I really have to credit a lot of the senior members of the AI Community for this uh like Jeff Hinton and Yan Lun and and uh Andre in and uh yosua Benjo uh who've been working in this field for a long time and of course all those were just this the well-known deep learning researchers there are a lot of other uh famous uh machine learning researchers and and there's a huge uh community that really cares about pushing the state-ofthe-art in the open in the public domain and sharing their knowledge and setting up conferences in ways that encourages you know knowledge sharing and fortunately once all these uh famous people got hired into these top companies they continued uh with that and continued to publish and hence are pushing the state-of-the-art together um even though the companies might might compete uh on the business side and so uh in that sense I think the only uh sort of war that is going on is the war for talent and that is also in some ways the answer of why why there are Acquisitions um of course for small startups uh you can have a lot more impact if you can uh apply your tools uh and and yourself to you know companies that already have a huge amount of resources have a huge amount of data which is you know data compute uh and algorithms are so the holy trifa of uh getting getting things done in Ai and so it's it's a perfect mix um yeah I'd chime in on that uh last point so uh I'd agree with the first the first point about Humanity winning ultimately um technology moves on and we adapt and improve um but when it comes to companies so uh to that point larger companies obviously have an advantage um and I think it's largely because they have the information so I'd say information and not data um I think people get hung up on data so there are a lot of large Fortune 500 companies that may have a lot of data but it may not be the right kind of data it may not be act actionable information in the same way that Google um has a lot of data right or that Amazon has a lot of data um and so I think the big players we probably know who all the big players are they the people who are competing for that talent and who are acquiring startups or getting people out of Academia and out of those large companies I'd call most of them information Focus companies some of them have a little bit more of a hardware or design Focus Tesla maybe is an example that Bridges both where there's signs that they are doing Ai and thinking about it in a in in a in a way where they're pushing live updates their their their cars are online as opposed to being offline um but if you look at something like apple I think I I was joking before the panel that this would be a funny topic but I I do think there's a real battle happening across all these companies we just saw Google launched their equivalent of Amazon Echo and I think it would be interesting to ask who do you think is actually going to win you like Google would be the traditional answer but um and and but then think about Siri Siri is the one that's been around for a while uh but I would argue apple is doing a lot of things that are limiting the data and the information they have to operate with and they're doing it for what you know what they think are good reasons but I think in the end it may harm their users in terms of the the quality of the product so one quick anecdote here uh that I keep hearing over and over and over again is usually when we talk to Founders they have in their heads some sort of strike price for oh well you know if they paid me x million I would be willing to sell the company but nothing below that but the interesting thing in the machine learning world is is we've seen a lot of the founders say well you know I would actually be only be open to being acquired by X or Y because of the specific data set they have so it's interesting to think about the fact that they're not leading with a a numerical amount for you know what they would sell for but they're like oh actually you know if I were able to power my platform with with this data type and have access to these other customers this is actually sort of a strategic thing in my mind as well um which is interesting and um two other quick points so one is on you know when we make investments at seed stage a lot of these companies are machine learning first and by that I mean they're choosing a problem type that's only solvable by virtue of you know some of the Innovations technologically that that have happened in the last couple years and then we're seeing companies and the best example of this is something like uber which is machine intelligence second right so they put together this Marketplace and I mean Uber has a lot of things they have going for them but arguably one of the reasons they've become so dominant is because they were able to put the right data science and machine learning team in place and that's how they really Amplified the value of the product uh and then the one point and I'm curious to hear you know everyone in this panel has worked with with Enterprises um so I'd really love to get your take is is we've seen a lot of these big companies be entirely binary about whether or not they have machine intelligence expertise so you see these platform companies like a Salesforce or a Facebook um or Google pretty much they've got it you know from from the very get-go they've had machine learning as a part of what they've been doing they understand the product iteration cycle they understand what talent types they need and how to motivate that Talent with the right data and incentives um and then the rest of the world is near zero and so the struggle there has been you know how do you educate people enough to use self-service platforms to get themselves to catch up and so that's kind of just one thing I I hope we can tease out over the course of the panel so I think um I guess it's a tough it's an interesting question I'm a Founder who's kind of a fighter we've been right it's a long game none of this was a short game AI was not when we were fundraising we got kicked around by across all all these investors nobody was interested in ml they wanted faster SQL right when you go ask the customers to say Highway slow faster horses faster SQL not a car right so we so we defied the odds and we still built an AI platform and data science platform and R we we used R Community to really get the Embraced as fast than python Community all open source folks so um most Revolution start by being totally derogated right so nobody every ignores you to to death right so that's when you actually build something interesting and Google themselves I mean one of my investors Angel Investors he had an opportunity to acquire Google for a couple of million dollars because the founders were tired of building a machine Learning Company themselves right so uh you kind of tend to think about this Acquisitions as winners but they're not really I mean if Google had sold in that particular investor angel of my seed investor ended up selling to Amazon you would not have Google today so I think really if you think a long-term scheme of things you need to build a platform for to build more platforms and and build a networks the ecosystem around raise a forest not a tree and that's what Google really did is it build a full Forest full of ideas and you need to fund that ideas with lot of energy continuously and continue to dream again and again and again within the same company a startup is not one idea right it's it's a million ideas that you go back to back and continue to invest in them invest in your people it's the people who are building the company not products the products keep changing and AI today and tomorrow will be like basically whether stealth. a or you call it health. a or insurance. a you can have go change one vertic at a time and it's not true some of the big companies actually are some of our best customers are not in the Bay Area they're in Midwest somewhere in Cleveland or or or Dallas so it's not actually true that big companies are sleeping they're actually changing dramatically faster uh if you see how Microsoft changed overnight into it a Microsoft Ma aai I think change is possible except you don't need to change all of United States you need to change the Silicon Valley of your company and you need to find that and and tap that nerve so you can bring CH across a big vertical company and that's happening we're seeing a lot of that happen and being Partners in that change what makes us what makes this whole thing exciting the the other big difference you'll see between Ai and a non aai companies is the winners in AI so historically you would say there's a new trend coming you need to be the second best second to the last so the bear gets the last guy and you're still running fast right in the AI rise the first person who crosses the line turns around and becomes the bear so if you don't become the person who adapts your disruptor you're going to the entire vertical will become lost so I think that's kind of what is really driving change in lot of these big companies these days thank you uh I think it we really got a a broad spectrum of views uh on where this field is going I want to kind of dig in a little bit and kind of take apart the pieces right I think we're deal with pretty complicated system so uh today multiple uh components of of U machine learning AI powered companies were presented and I think there are some of the most complicated startups around right because for instance deep learning is the current most uh interesting hot uh uh set of algorithms but you need to go to school to to properly do it right you know even things like support Vector machines are much similar to take apart uh right but but for deep learning you really need to uh to have proper expertise which will take at least you know a couple years probably to to build up properly and we at Nitro we we looked at uh various stages and we kind of uh funded an internship with Stanford alic group to derive you know the basics from from math first principles and uh did the paper review which basically entailed you know several months just to understand uh and the whole like the knowledge is embodied in the community you will not just understand by reading all the papers you need to speak to some of the people so so we have this technosocial system where certain C knowledge is embedded you can glean some of this you need to go to conference you need to inter with people to to understand the algorithms themselves but then you need a scalable system we need a distributed system which is another uh theme of this conference which we call data pipelines and we basically have about half of the talks uh which also requires properly you know career right and data engineering is essentially uh an art in itself so you need at least two pieces now we have VC because uh the the effort needs to be funded you know top best of these people need to best assemble together so uh I would like to ask you uh which which parts of these do you think are most important and how how in your experience they interact how if you you run a company how we putting them together and and basically if you want to achieve you know world domination how how should you put the whole thing together I I think you left out a key another key piece which is uh the problem to solve right so I think that's I think you and Siobhan could probably uh chime in after this with more details on that uh what it looks like from an investor point of view but um from my my my view of the landscape I think uh it's really hard to marry what you just described so a that's that's a deep bench kind of startup right so you need people with the AI Talent who otherwise would be going to a Google or Facebook or they would be uh doing well in Academia plus you need to build the the infrastructure in a sane way not you know nothing wrong with academic code but you know it needs for Enterprise or something it needs to be robust and scalable and all those kinds of things um but then you also need to be solving a real problem that they care about that's often one of the hardest ones for technology startups is is you build some techn techology that could be great um but people don't want it you're you're you're building the wrong car you're building a Jeep instead of you know the car they want or or they don't even want a car they want a they want a horse a flying machine um so uh I I think that like getting those things together that's that's the art of building a new product um and I think if you're a big company you can absorb a lot more hits right so really it feels like a startup uh you only have so many bullets um and you have so much runway uh for this particular type of company I think like so one one thing I've seen is um either you need to have a bigger War chest so that you can build the back end and the front end and research with the customer the front end um or uh you know maybe you shouldn't uh you should be at a bigger company uh where you can take more swings uh so I think kind of there are two Cate categories that that we look at and invest in and one category is the pioneers and one is the integrators right so I'll just give you two contrasting examples so you know what I'm talking about um on the Pioneer side these are folks that you know World exports like you know Richard um who have focused on you know pioneering these new techniques and then they're going to apply them to a problem so we made an investment in a company called Deep genomics which is you know out of the same lab is Jeff Hinton a fellow named Brendan Frey who'd been working on these algorithms for 15 years and then 10 years ago you know he had a personal medical issue him and his wife you know had uh a child yet to be born that couldn't be diagnosed and they're like we you know I'm working on these algorithms that in theory should be able to solve this problem so he spent a decade just focusing on Healthcare saying you know if we were going to solve a problem here or if we were going to adopt algorithms to these types of data sets how would we do that and so you know he recently got to the point where he's like okay you know we think this is going to be a big deal we want to throw more significant resources behind it because we can we think we can change the world of molecular Diagnostics by kind of closing that genotype phenotype Gap with deep learning because it's impossible to do it otherwise or without a neur neural network based approach um and so you know that's a category that's certainly investable I think the one thing that everyone has to keep in mind when you look at companies like that is it just it just takes longer to commercialize right because you have a technology but you have to run a whole series of experiments for what it's going to be best at solving within a domain and then on the other side and I I would say we've got about 20 investments in this category it's it's called the the integrators who are like all right you know I keep on top of everything that's happening in terms of scientific papers coming out you know I look at all the different Frameworks out there you know whether I'm dealing with tensor flow or nltk or anything out there and I think that this problem is basically solvable if I can intelligently integrate and adapt these Technologies so an example of something like that is a company we invested in called texo Has anyone used texo ever all right people on the panel have um but essentially what they're doing is if you think about spell check you know we wouldn't ever send out an email without using a technology like that think if this is spell Che 2.0 for any domain specific type of text you can basically optimize it based on knowing everything else that's out there so the particular example they're dealing with out of the gate is a job description so we all have to write job descriptions um on average we suck at writing job descriptions and it's no fun so what they've done is you know they've read millions of job descriptions and they also understand how long it took to hire a person how good that person was what were the demographics of that person after the fact they're using fairly rudimentary machine learning Technologies to back out okay you know how can we 8020 this how can we use data to say you know this job description is way too long you probably shouldn't use these words if you use this set of words it's going to be gender biased towards women and you would probably want to rebalance that um you know they haven't invented any technique what they did is they understood a do domain very well and they were like what can I pick and choose from uh and so that for us I mean that's the free lunch that's out there you know to Richard's point about about the fact that most folks working on AI come from Academia and therefore are very giving to the community there's so much stuff out there that if you find the right problem you can just take existing algorithms and and use it to solve those problems I think you you said most of the right things I think the underlying pattern here is if you're in that second category you need some interesting data advantage and laser focus right and so I'm still curious how they got that data so you can get the you can get all the job posts but then you actually need to know the companies and then later on like how well that person did so that was I'll just quickly answer yeah one one of the reasons we actually invested um so we look at data companies as kind of have you ever been to a really cheap hamburger place where you know you get like this much bun on either side and like this much as the Patty so we're really interested in the problem and the user interface for the problem and the underlying data and so the reason we invested in large part here is these guys had strong relationships with Microsoft Amazon they actually went to Bloomberg to get a whole bunch of this data and so B basically this was all large amounts of Silo data within organizations and as soon as you got six or seven of them together that was enough of a core data set to be like okay we can actually do something it's awesome yeah I think data is a good um good place to start I think a lot of public data sets are available even while we were popularizing some of the tools in the early days with meetups and stuff customers would be finding public data sets there's private data sets they data alliances to be formed I think Google has the 359 degrees view of the world but that 1% uh one degree left is amongst the corporations so corporations will come together and build data alliances and that's probably going to happen in the next probably 12 months um and a lot of our own customers are building alliances amongst themselves so they can get better models from the same data um there's a lot of new math techniques that come about called consensus modeling which you can share the models without create the model as if it was just one continuous data stream as opposed to broken up data um data monolithics um I think the bigger piece here is was of course data is one part of it apis API economy is really on you can get Jon from a lot of these things scraping or um open source is huge right sort of uh we built an entire community and customer Community from just pure bread open source and marketing much like um Alexi was one of our um big hand Alexi who runs an open source Consortium and data by the way um uh every year year on year I think a lot of good um Talent is interesting I think think um Talent can be grown as well um I mean obviously there actually the other way to look at the company description that um Shimon gave us the two types of AI companies the one that are started by the professors or or professors in the universities most likely going to get acquired or fail U the second one is real entrepreneurs who going to build companies for the long I think Jeff Bezos he was building really a an AI company in 1997 his shareholder conference s say the same thing so those those are the kind of people who built or the horizontals long before we are thinking of them as horizontals so truly entrepreneurial stuff is going to come the the next 10 years it's still just the beginning of AI phase I would say thank you uh so I'll ask one more question uh of which panelist and I will ask uh folks in the audience who have questions for the panelist to please uh line uh up uh uh in front of this audience mic which we have here if you have questions so then we can uh open the floor uh so my question will be uh each of you uh obviously has a lot of experience in either the field itself or invest in it and uh and you have key sides what is your key Insight you know what do you see is your key strength of your company or in case of Shan investing and where do you want like what do you want to do with it like where do you want to advance the field what is your vision of for from your angle what do you want to do in the next year to maximize this insight oh um that's a good question so I think the I I was talking about this the other day with our team so uh our first hire was actually a designer um and so as a data company I've working in a large organization we're thinking a lot about problems of of actually doing things within an organization how do how do knowledge workers become more effective how do you get things done um and one of one of the challenges in any organization I use the example I think I tweeted something about this the other day was uh drop cam and and nest and so they're within Google which is the mothership of all AI companies yet when you know they drop cam was acquired a couple year a year ago or a little over a year ago um but I get 10 notifications a day on my phone from drop cam which are clouds passing by in front of my house and it thinks there's a person you know coming coming to my house right so it alerts me um and on the other hand they have Alpha go and they're beating like the world champion go player so what's going on here right um why does my drop cam so stupid right I it's not that people at Google are stupid right they have you can get these smart people together and they still can't cure cancer right what what is going on how can we make uh people more effective within any kind of organization um I think the biggest challenge for any company and especially for data companies is bridging those two worlds right so you have the designer who thinks okay I'm going to build this awesome thermostat or this awesome camera and they're very design focused uh developing a new product and then you have the AI people who are kind of maybe decoupled or brought in later so how can you bring those two people to those two groups together in tandem because one I ideas are not born in a second they develop and they evolve and so I think that's unique skill set that we strive for in my organization is is to bring those two things together and it's not easy it's not like pleasant but you you work through it and I think you come out with a stronger uh product in the end you want to go sure well I don't have a company so this is going to be slightly different from the Venture perspective but I mean realistically you know we're in an interesting time as it relates to the things I focus on because they're just so hyped up to the nth degree right now um you know what do we have going for us the fact that we've been you know investing in data science and machine intelligence well well well before any of them were sexy which is good and bad so it's um good from the perspective of we've seen thousands of these things and we understand sort of subtle and unique patterns for you know what tends to successfully get a company from C to a and and what doesn't um the downside to that is it makes us on average slightly more jaded right just because we see some of the red flags out of the gate and so you know in dealing with a Founder we're going to be very brutally honest about the fact that you know we've seen this experiment that sounds like a good idea 20 times and here's why it didn't work and here are some alternative strategies and you know so some people don't necessarily want to hear that but um you know I'll just share one anecdote we invested in a company called uh Domino data labs and I remember I met the founder a year before they were fundraised just because he was looking for advice because you know I'm constantly you know writing reports and researching the space and I said you know I've looked at so many of these data science tool companies and I don't believe the B tods model exists yet and I was like I know it well and I can't wait for hopefully you to prove me wrong but here are the four things that I've been waiting for and so I kind of laid them out for him um and you know I don't think he was too happy at the time but then he came back a year later and he was like and I was watching the market and I was like okay you know we're starting to move into that and one of that was just the real pain of Team expansion within large organizations and that being sort of you know the thrust of you know I don't believe an individual data scientist often will buy a tool but when you're trying to convert data analyst at an Orit will so again like kind of annoying out of the gate but we're just really really thoughtful about what it takes to scale the data science or machine intelligence org and hopefully that helps the founder um I think the reason I'm super excited about Ai and the way I'd like to push forward the the field is there there two things the first one is artificial intelligence as a as a field is super exciting because it makes us uh pushes us to ask questions about our own self like what is intelligence why are we so special compared to other animals um you know is there is it motor intelligence linguistic language intelligence visual intelligence it it really asks us and and pushes us maybe to you know extend definitions of intelligence to you know artificial ones and so on uh so it has this kind of pretty deep aspect that is very scientific and and kind of uh almost evolutionary um at the same time it is just incredibly useful uh and uh so the reason um you know I'm building both like a research lab as well as like an applied uh group uh at Salesforce is that they have a huge amount of data and um I don't know if you saw but in the last couple of days Mark um sort of talked about um artificial intelligence too and and if you have that kind of leadership um and excitement and you have all these resources um at hand you can have so much impact and they're uh really working with tens of thousands hundreds of thousands other customers um and uh their data set so it really becomes the true met mind if you have a an AI group uh inside an organization like that scale um so math is actually um I remember in Santa Fe we had a researcher who would say math is actually the only provable religion out there right sort of um and the word data I mean the source word for data is datum datum is Latin and became popular in 1700s so actually Source word Sanskrit datum means to give to impart energy to impart a mineral with high rich in energy uh so I think that data as a whole data science is a search for truth right so amplifying that whether through automation through AI through more deeper algorithms that are more accurate I think is it's going to amplify our speed at which we disc discover ourselves right which is kind of the deeper part here um as a company we we are building people right people are the product of a company and a startup um and we continue to dream more beautiful dreams right how do you come up with more beautiful data products so design um uh designers are our product managers really we don't have product managers so we actually get designers as the interface between customers and the products we're building um this the one one step further past beauty Beauty sells itself and bu Defending Your Community or nurturing your community love with beautiful data products that's kind of one of the core thought process there I think about six or nine weeks ago I would say I woke up with a much more richer dream and that dream is actually much more simpler but actually much more deeper and that I would say my vision is to build 10 women CEOs in Silicon Valley and unleash them kind of much like PayPal Mafia we about 47% women in our company are size 60 still small um but our vision yes our vision is to actually lead build real leaders like our vpf Marketing vpf sales or a lot of really hardcore leaders who think holistically because the way to bring change is simple things my working mom raised me and so I was always envisioning it be 50% women right sort of and Silicon Valley which is kind of the high point of civilization today as we think think of it is really extremely gender biased and extremely um I mean most people don't take the bus right so we we think of um futuristic ways and hopefully hybrid cars will get us there but really there's a ton of um Innovation that can be done in social life right so I think and that can be done through work because work occupies work refines us and so how you build culture that is very flat inside your organization where data science is the art of asking lots of questions so you'll be asked lots of questions in your company building because you are hiring a lot of data scientists and they'll be cynical most of the time and sometimes and most of the times they'll be right 99% accuracy rate so you're going to be uh knowledge if you think thought of knowledge is power and knowledge economy is what you're building you're very likely to be the least powerful person in the company you started it um and you're going to be surrounding yourself with much much more smarter uh people have thought seven seven steps Beyond The Go game right sort of so I think that's kind of the people you're surrounding yourself with and it's probably been true for all companies but in specifically if you're thinking the data companies everybody is going to be asking for data and signals and there are lots of ambiguous signals that you you see as a startup so you still have to have the faith right kind of the belief and I think the combination of belief which is what Richard was talking about as well as the pragmatism of building a business I think it's really truly one of the best times to build because in many ways infrastructure is easy to get like open source Cloud um getting communities is also reasonably um possible if you're honest um and and build trust with your customers so a lot of the other pieces have been truly commodified in some sense the only thing that's really you're trying to build is build a very trust um an economy of of truly um more giving than taking data as in to give right so thank you thank you all for this beautiful insights uh let's open the floor uh to the audience please introduce yourself and then ask a question I'll be first so I want to touch base on two points actually you've been expressing just a second ago one is about the economy second one about the evolution and no matter what part of the Spectrum in social and political life you you stand on you cannot deny that market is always trying to improve right so we had blacksmith which were replaced by the uh hydraulic processes we had you know professional horse riders that we replace by The Cars and the drivers and now we see that the drivers are being replaced by the driverless cars so nowbody rides the horse for money Nobody Does the blacksmithing for for money right so we're always trying to improve right so clearly artificial intelligence is an improvement upon the natural one right so we get in short of the natural intelligence so we're trying to create something better so here's my question to the steem panelist don't you find yourself waking up in the middle of the night in a cold sweat thinking of what you doing in the daily life will cont contribute essentially to the demise of the human natural intelligence so the people would use their intelligence as a vocation not as a profession thank you please introduce yourself uh my name is Constantine bunik I'm a CEO of the company mcor so I am guilty as you probably as well thank you guilty no I think um so one of the big transformation why is AI sexy right why is it's really cool and actually one of the goal for the companies is to make AI boring and if I if AI is still boring it's not still Sal sexy after H2O then we failed right so our goal is to make AI so boring that you don't call anyone to say we saw a driver we saw something that is cool based on AI much like search today is boring um I think the key piece of that is that historically software Engineers were the last historically this last 20 years 40 years is were busy looking at others as they were automating their lives out of existence right sort of so AI has the capability to consume all software so software was eating the world AI is going to eat software in some sense all of software rule-based software most all of rule based software is going to become AI based software patent recognition based software so and so that's opportunity one kind of like the Y2K opportunity one you can of can trans at all rule based code to AI based code and that's why Microsoft is excited it Google is excited about it or all soft Oracle is excited about it all software houses are going to be excited about change in big software right but the col v m is that we are going to find that offer Engineers can themselves be automated out of their existence which is kind of an interesting part devops. aai for example is one of the new piece coming up here and there but I would say a lot of software automation for software and automat atic Cod coding of software is likely to happen it big it's a before internet after internet movement of is the before Ai and after AI mov um and I think uh a lot of um lives will be changed and that's actually true and I think a lot of um and and the good way to keep human intelligence in the loop is to build visual interpretation which is human interpretation of intelligence if you can understand what's going on from the blackbox model to somewhat more of a gray box model then you keep the kind of the the human in the loop for the bigger automation that's happening for AI but 100 years from now is AI going to be around or human going to be around that's a question to be asked AI is going to be around yeah um so this is what I I the panel topic was uh listed as so I I I thought about this a bit before um so the fir first thing I'd say on the general topic is um I don't wake up in a cold sweat because this is a long way off I think I go with Andrew in on this one about uh someone asked him a similar question and he said I don't worry about this in the same way I don't worry about um you know income redistribution on mars or something because we're not there yet you know we don't have a colony up there um so I think it is a long way off and if you're a practitioner in the field and most of the people I think in here are you you know what you're doing dayto day and how difficult it is um where're you spend a lot of energy and a lot of research and a lot of time getting something to do uh getting an algorithm to work in a very narrow problem and a very narrow uh implementation uh kind of on guard rails with training wheels right um and so to take that and say it's going to become self-aware and so the singularity all that kind of stuff I said the sword sorry uh but it's uh a shot after I think that's that's way way way off in terms of the near-term impact on jobs I think that's more real um uh you see it with regular software companies and the Uber and things like that already it's causing ripples um what I would say is uh it's it's automating as as as uh as he mentioned uh things that already were automated and making them more efficient um I think certain jobs and certain roles in the short term if you can be automated you should be thinking about the next thing already right um and I the good news is I mean as humans I mean new humans are born there's new generations um we we saw this at LinkedIn actually what's happened over the last like 20 to 30 years is the the frequency with which people learn new skills and new careers new jobs is increasing rapidly you used to have a job for 20 or 30 years more and more people are are Reinventing themselves constantly and you have corsera Udacity things like that um so I think that the reality is there's always new work there's always new jobs we have to be more creative about it um I don't I'm not an economist I don't know the short-term reality in terms of like Universal income or you know how how do we buffer and how do we Bridge these gaps I think is the key question um but in the long run in the same way like you said hydraulic press is we're going to come up with new things and they're going to be difficult and the machines we build until we get to that Singularity are not going to figure it out for us right so we're going to need people to figure that out so I I'll just be very blunt about the way I think about this and and why I do what I do because I've I've thought about this a lot um my fundamental premise in general is you can't play defense against technology like I just take that as a given and so given that you know what do we do what are our obligations and for me it boils down to three things and and the first is you know every technology is going to have a net positive and negative effect um I think the best we can do is is try to over index on the good side of things and and how do we do that so I I will be blunt about the fact that I will only invest in things that I think will have a net positive effect on society and so you know investment is a piece of that there are a whole bunch of nonprofits doing absolutely incredible things with data and machine intelligence um from you know dealing with climate issues to um digital Defenders of children which is an organization called Thorne and so we're seeing a lot of good happening there uh the second piece of this is making sure a new technology type is effectively democratized right and and there are a lot of folks that are are working on this problem um I volunteer at open AI which is was which is hoping to do that we're also lucky to Richard's point about the fact that a lot of folks are academics and so are pushing this out into the world just to make sure more of the community the is distributed and more of the community is involved in in shaping what it becomes and that sort of bleeds into the third piece is is I think we're obligated to get in front of you know regulations and systems that will help deal with this technology and I think one of the risks we face is you know to your point we're learning skills much and much more quickly but I think the pace of change is accelerating and so you know 50 years ago when you when you had a 20 year 10e ad to Corporation if you needed to be retrained often the corporate would step up and take that that role um to make make sure you were going to be okay for that next skill set um that's kind of dried up and so you know my colleague at Bloomberg beta um Roy bahot has is heading up something called the future of Work Commission and the idea there is we don't know some of the these answers one of them fundamentally is is this shift any different than anything that's happened before um so working with economists to your point to you know answer those fundamental questions getting everyone to the table from public and private industry to just say okay who's going to own this there's going to be worker displacement whether it's white collar with a lot of these analyst tools or whether it's blue collar with self-driving cars um how do we make sure we get in front of this and you know keep the American population and the world population working because you know bad things happen when people are out of work did you want to we're good okay thank you I think we have time for one more uh audience question if folks have a question hi uh my name is Santos and my question to the panel is uh today AI is uh maturing from its infancy we are able to talk to it AI is able to talk back to us we are playing with it like a small child but um and it is growing very fast right uh the Aging is not like human aging that it will grow one day at a time it has uh exponential growth are we playing with a monster or a sadu right so we actually think that AI will be like a sadu or a guru in our company it's not a bot I think bot's the in the American the the new V word of it's the same as it's racist right so to think of AI as not being sentient is like think of data in Star Trek right sort of um they are probably more likely to be um reliable and less human error than than human error right so I would say um it's actually the opposite we building that's one of so we good Kudos on the open source we we chose to be open source because it could be substantially monster four years ago we had the vision that this could be so dramatic that needs to be open right and so open source kind of makes it Democrat democratizes it but in the deployment of it you will see some real substantial monsters being deployed on the on the world and luckily there hopefully there enough Engineers enough ability to debug these tools right sort of it's a it's it's truly you need to be able to debug a complex system uh on why this model failed to send this patient to ICU right I mean literally did I mean that that's the kind of AI That's being deployed today and it could save lives by being applied correctly by early diagnosis of disease we've seen people use it to fight cancer ultrasound image recognition for example early pre like 10 years ahead of human eye so before you we get to this dangerous State I would say to some degree a lot of good will come together as well and hopefully have just like humans you would have good intelligence and evil intelligence and hopefully it'll right so for that point I would like to add that yes when you say about a saint and a monster yes that decision is still with humans but we are crossing that limit so fast that at some point it will just go beyond the human hands altoe it is already Beyond human I think actually um uh to some degree it is good that it is beyond human because the humans are actually more likely to mess with it right uh to some degree I think um the bots so I had a question um for my own conference last month have all what how would Bots meet together and build do a data by the bay themselves and and that was a that was a line on the on on my on right data by the Bots right or Bots by the bay and the answer was maybe they already have met 10 years ago and we don't know right so so why do we think we They will announce it to us Singularity was not televised right I I'll just echo my my previous answer so I think that the Bots by the bay is is a way is a way off um but yeah at some point it's conceivable and probably in the near future that Purpose Driven built machines we're going to have to address some real moral dilemmas and I think we're already pretty much there on that point when you think about self-driving cars so there's been a bunch of writing and thinking about this and as a software engineer there there was always a classic dilemma and when you went to school uh you learned about good practices ethics and principles in engineering like you're building a bridge you have a ethical responsibility that bridge doesn't fall down when people drive over it now when you're building a car it's the classic Isaac azimov Laws of Robotics dilemma is it protecting the person in the car or is it protecting the other cars is it going to do some math to try to minimize total casualties these are things that we haven't figured out and they're they're happening and the technology as often happens is leading us figuring these things out um and similarly you know people I I'll throw this out there we're it's a panel we're supposed to be a little uh uh controversial but people are very worried about terrorism right and uh they think about uh okay is a drone is an AI going to come kill us is a drone going to kill us just think about cars you have self-driving cars self-driving trucks those when I learned how to drive the the instructions were this is a massive weapon and I had a you know some former drill sergeant doing driver's ed and saying you're getting behind the the wheel of a massive weapon and so now we have a software system that's coordinating centralized across all of these self-driving cars that's a massive weapon so there are things like this that as Society we're going to have to think about and I think that's more realistic than saying we're creating a monster the monster is us and we're the programmers programming these things you you you just jumped on the point that I was going to make because you're 10x smarter than me um but I was yeah I was just going to try to be quasi controversial and and point out that often it's the fact that we are the monsters um you know you can Envision a not so distant future where self-driving cars have dropped the amount of deaths on the road by a THX and sure it's going to be horrible the first time a self-driving car hits a kid um and we're going to be 10,000 times more mad at it than when a drunk driver hits a kid and so that might set things back but you can actually it's not that inconceivable to imagine a future where it's actually irresponsible for you to get behind the wheel because you're a THX more of a liability on the road than a self-driving car is and if you hit a kid it will make sense where it's just like wow that really selfish person they felt like driving and therefore my child you know was hit by a car uh and the same thing is true you know one of the things that kind of baffles me is if you take a look at hospital situations the RS of drug misadministration are extremely high um and a lot of deaths in hospitals come as a result of that and we could have computer systems that could help solve these issues um we choose not to and why do we choose not to because we would have we would prefer to have a human to blame at the end of the day when a catastrophic event happens even if it doesn't lead to an optimal outcome and so you know in that case we choose for ourselves to be a monster when there could be a better outcome and it's going to take a lot of time for society to change its views on that um not that there aren't massive issues with some of the systems we're building you know in in areas like reinforcement learning if you're rewarding an outcome you're not talking about the means to getting there and you know that gets back to bom's famous paperclip problem of you know you ask the machine to build Paperclip that use all all available resources which is every you know metal component in the world then synthesizes other things and and and destroys the world right so I'm not making and you pointed to the why problem of you know can your system actually explain why it did what it did so there are huge problems we're going to have to deal with um but I think we do have to understand that there's a lot of harm that comes today that can be avoided so we should just be a little thoughtful about that as well all good and some doctors actually are so thoughtful they want to use the AI as their like first um diagnostic agent right sort of become and some of the pragmatic doctors we've spoken to like it would be it become criminal not to ask a computer what are the top three uh choices to choose um because maybe they're clouded by their emotional state of the day um and I think and that's actually one of the key piece right so humans are going to be good at what they're good at which is build emotional energy um build a lot of um community and trust and the more we are able to build trust with both the AI and the human intelligence that surrounds us I think um the more open we are to ideas that are not ours and I think machine Moneyball is is a difficult act right it it takes courage to follow the sus that the the data says uh when when the Oakland Ace were doing that Ace had the Oakland has had given up on them right sort of and to that degree they let them do a money ball on their favorite team um so day to-day decision making is really hard and it takes a lot of courage and so when data or the AI says you need to do this it still needs uh courage for the human to act on it and I think that's kind of an interesting uh human AI um discussion conversations is waiting to happen very soon and I think um I think Global markets are probably not yet um capable of dealing with this Mass transformation of of knowledge and information has always been the what we've been trading as in an economy and I think to some degree um AI could be playing that economy key key role and the key player in the market could be AI itself right sure thank you uh let me summarize actually I think it's uh really fascinating that the main theme which emerged from this panel as I hear it is actually that you know human intelligence is is uh what is actually emerging supported by I Ai and uh and she talked about building humans and I kind of want to just give you all guys a little challenge so we found that uh the data by the conference the main value which we built as human connections and and so actually even Foster connections among the panelists so a piece of History last year as a result of our previous panel one of the panelists joined sh's company so uh I'm not you know encouraging you to do that but uh I think each of you actually as I hear it actually interested in in human development right so she thinks of you know building new leaders uh Pete thinks of uh plans to improve productivity at workplace uh sh obviously invests in Things She Believes In and human they're basically human first and Richard I think empowers a whole bunch of people through the tools because Salesforce eventually basically governs uh large workforces so and Alexi brings us all together so big hand to Alexi thank you thank you so so I I just wonder uh if you guys can continue this conversation IM more yourselves and see kind of what what betterment of human condition in Silicon conve at least we can we can enact and uh also another thing I want to you know ask all of you guys as as you noticed we have a very strong overfitting signal if you submit a paper to this conference submit a talk you you you're high likely to be acquired by a Salesforce so I I really encourage all of you that's my to get out of here uh yes uh uh to do that and and the third point I want to make is that we have you know I think believe the best happy hour uh in any data conference we have you know 6 to eight happy hour with full bar excellent food and conversation so encourage everybody to continue this upstairs and thank our panelists for for the excellent [Applause] panel