David Yakobovitch
DUE TO SOME HEADACHES IN THE PAST, PLEASE NOTE LEGAL CONDITIONS:
David Yakobovitch owns the copyright in and to all content in and transcripts of The HumAIn Podcast, with all rights reserved, as well as his right of publicity.
WHAT YOU’RE WELCOME TO DO: You are welcome to share the below transcript (up to 500 words but not more) in media articles (e.g., The New York Times, LA Times, The Guardian), on your personal website, in a non-commercial article or blog post (e.g., Medium), and/or on a personal social media account for non-commercial purposes, provided that you include attribution to “The HumAIn Podcast” and link back to the humainpodcast.com URL. For the sake of clarity, media outlets with advertising models are permitted to use excerpts from the transcript per the above.
WHAT IS NOT ALLOWED: No one is authorized to copy any portion of the podcast content or use David Yakobovitch’s name, image or likeness for any commercial purpose or use, including without limitation inclusion in any books, e-books, book summaries or synopses, or on a commercial website or social media site (e.g., Facebook, Twitter, Instagram, etc.) that offers or promotes your or another’s products or services. For the sake of clarity, media outlets are permitted to use photos of David Yakobovitch from the media room on humainpodcast.com or (obviously) license photos of David Yakobovitch from Getty Images, etc.
Welcome to our newest season of HumAIn podcast in 2021. HumAIn as your first look at the startups and industry titans that are leading and disrupting ML and AI, data science, developer tools, and technical education. I am your host David Yakobovitch, and this is HumAIn. If you liked this episode, remember to subscribe and leave a review, now on to our show.
Welcome to our newest season of HumAIn podcast in 2021. HumAIn is your first look at the startups and industry titans that are leading and disrupting ML and AI data science, developer tools, and technical education. I am your host David Yakobovitch, and this is HumAIn. If you like this episode, remember to subscribe and leave a review. Now onto our show.
David Yakobovitch
Welcome back listeners to the HumAIn podcast. Today’s episode features Steven Shwartz, an AI author, an investor, and a serial entrepreneur. Steve is the author of the book, Evil Robots, Killer Computers, and Other Myths: The Truth about AI and the Future of Humanity. Steve’s also a multiple-time hero entrepreneur with ventures throughout the technology ecosystem and he invests in early-stage startups, will have a jam-packed episode on all these topics today and more. Steve, welcome to the show.
Steven Shwartz
Thank you, David. Great to be here.
David Yakobovitch
Absolutely. It’s great to have a fellow New Yorker entrepreneur, and it’s great to see the ecosystem, of course, come back in full growth now that we believe the pandemic is potentially behind us.
Steven Shwartz
Full disclosure, David, I’m a Red Sox fan.
David Yakobovitch
I was going to say the Yankees, but it’s okay. I still got some love in my heart for you Steve. It’s good when you think about sports. You think about everything happening in the Northeast and you have a very in-depth background in technology and AI. And I think sports today, my favorite sport actually is tennis.
I’ll be frank and I love one of the benefits, the only benefit from the pandemic for tennis was now, the AI cameras are being used for every single shot. Is the shot in, is the shot out when you serve, when you hit the line and I’m sure similar things are happening in baseball and other industries around the world? So can you start by telling us a little bit about your career and why you’re so passionate about AI and technology?
Steven Shwartz
Absolutely. And I didn’t realize you were a fellow tennis player. I played four days a week. I love tennis.
David Yakobovitch
And we’re going to have to play sometime out in the city or out in Stanford?
Steven Shwartz
Yeah, absolutely. Let me tell you a little bit about my background. So I started my career in AI way back in 1979 when I moved to Connecticut to do post-doctoral research at Yale University with Roger Shank, who was one of the pioneers of AI, then came several AI startups, one of which made a public offering. And another became one of the leading business intelligence products of the 1990s.
That was a product called Esperanza. Then came a number of other startups, some non-AI, as well as, and I did those as both an entrepreneur and an investor. Tango ended up as the fifth-best IPO of 2011 and my most recent company device 42 doubled or tripled revenues every year. From the time we started it in 2012 to the time we sold it in 2019 and it’s continuing to do well. My co-founder is the CEO and continues to run it nicely. And currently, since I wrote my book, I’m advising companies investing, writing, and speaking on AI
David Yakobovitch
And this is not your first book. You’ve published before as a researcher and scholar. So can you tell me a little bit about your journey having previously published and now with this new book, why were you passionate about launching a new book?
Steven Shwartz
I can certainly say it’s my first book in 30 years to give you an idea of how old I am.
David Yakobovitch
Well, AI has been around
Steven Shwartz
It was really hot in the early eighties, late seventies, early eighties. And it was a really exciting time to be In AI. And then it kind of died out towards the end of the eighties when it didn’t fulfill its promise. And then since about 2000 with the success of, we’re starting to see self-driving cars and machine translation and more recently, image recognition. AI has really started to catch on in the last five or 10 years, it’s been exploding but what kind of led me to write my book was, do you remember when IBM’s computer beat the Jeopardy champions?
David Yakobovitch
I still love watching Ken Jennings on TV, but I remember that
Steven Shwartz
Of course, I was rooting for the computer. I thought it was, I just thought it was great when the computer won, but I kind of had a pretty good idea of what was going on under the hood and I knew there was no real intelligence there. It was just massive dictionaries and awards that the system was sorting through and matching up to the questions.
And IBM’s David Ferrucci who led that project eventually published a 10 or 12 series verticals in IBM and an IBM journal that explained exactly how the system worked. And it was really just a lot of really clever tricks and statistics. But then IBM started marketing Watson. That wasn’t the original name, but it became the Watson division so they started calling their technology Watson. They started marketing it as their system can think and reason just like a human. And I was really turned off by that. And then Microsoft had a system that they claimed could read better than people. And there was some fact in that statement, but it’s really untrue.
It could read, perform one reading comprehension test at a better than human level, which is not at all indicative of intelligence. And we’re seeing more and more of the same thing. I had dinner with Roger Shank in New York about four years ago. And I was complaining to him about all these, all this hype about, about that part of AI, that wasn’t true. And he said, well, why don’t you write a book? You’ve been in the field forever. So I did.
David Yakobovitch
And it’s, it’s so interesting, Steve, that you mentioned that just a couple of minutes ago, you were rooting for the machine and I could see a few reasons why we’d root for the machines. First, it was, you’ve been seeing the technology over the years and decades, and you’re like, you want it to get to that cutting edge breakthrough moment, which we might be now in this watershed moment. So the speakers, you mentioned, with machine translation and image processing, but should we always be rooting for the machines? Should we sometimes root for the humans? So what inspired this new book?
Steven Shwartz
I’m just fascinated with how much progress we’ve made in AI over the last 20 years. We have, we can take a picture and our smartphone can say, ‘oh yeah, that’s David and that’s Joshua’. And that’s my two-year-old grandson and go to a foreign country and having everything translated for me. It’s just amazing what AI is doing today. So it’s, it’s real technology. It’s affecting the world and it’s exciting, but it’s not intelligent and there’s no way that we’re going to get to that next step that we see in science fiction movies, where the machines are intelligent and we have the Terminator scenario where the fearsome Terminator tries to eliminate all the humans.
Or there are lots of RRD2. That’s not what’s being built today. What I tried to do was to write a book, a mainstream book that would explain to people how today’s AI really works and why we can’t get from here to there. That is we can’t build computer systems that are really intelligent based on what we have now. And there’s no reason to think that what we have now will ever evolve into real human-level intelligence.
David Yakobovitch
There’s so much to do, to get from where we are to exactly what you just described that human-level intelligence. I think to my former insurance provider Lemonade Insurance, Israeli grown start-up, love what they were doing with technology automation in the insurance space and what their app has done has convinced consumers that an AI-first app, you go in, you have this conversational bot quote, Maya, right? That you think you’re chatting with a human, but you’re not, it’s a decision tree matching different conditions. And everything is just automation. That’s right. But not really AI
Steven Shwartz
I like to distinguish between conventional programming and AI. Where conventional programming is where a programmer writes down a set of exact instructions and tells the computer exactly what to do, where an AI, the computer learns something and doesn’t have to be told exactly what to do. And what people are afraid of is mostly the conventional card. So most of the things that are taking jobs, for example, is conventional software, not AI software.
David Yakobovitch
And when you think about this with the jobs that are being taken. We’ve been hearing much of the last few months about this great resignation and about a lot of the antiquated industries going through automation and cloud services as a result of the pandemic or being accelerated by these contributing factors. And what is it about these jobs is that there’s a task that is repetitive, but you can complete the task over and over again. For example, let’s look at the restaurants.
You can go to the restaurant and the way there sits you and your family and friends down the table and hands you a menu. Do you actually need that physical menu or can you have a QR code to scan that food yourself and save on physical costs and save on the time of service? So, that’s one benefit that has impacted the service industry. And we’ve seen that all throughout different parts of the economy, both before the pandemic, but now being accelerated as the world starts thinking about technology first.
Steven Shwartz
Exactly. And that’s automation. But it’s conventional software. It’s not AI. And most of the examples of where computers are replacing people, it’s conventional software. It’s not AI software.
David Yakobovitch
There’s very few scenarios that are starting to emerge recently. It was in the news that the big company that many of us may use for our online shopping that we may know of is Instacart. In fact, they purchased this company called Caper AI and Caper AI was an AI company out of New York out of Brooklyn that basically built these shopping carts that had these cameras attached to them to recognize the barcodes of your items. So then you get to the checkout when you’re buying groceries. And you can very quickly process that. There’s a little bit of automation, a little bit of a computer vision or image recognition, so to speak there and Instacart purchase them to say, well, our shoppers can shop quicker If they have Caper AI, shopping carts, really interesting use case,
Steven Shwartz
It really is an interesting use case. And I don’t mean to say that there’s no contribution of AI to automation that is replacing jobs or changing the way we do things. And the big place where AI is really being used in the real world is computer vision. And there are a lot of exciting things you can do with it.
You can, in a grocery store, monitor the behavior of people to see what they’re taking off the shelf, putting it back on the shelf. The application you just mentioned makes a self-checkout faster because the cameras can see what’s in the shopping cart. So that’s clearly AI and it contributed a little bit to job loss, but I still think it’s mostly automation that’s doing it.
David Yakobovitch
It’s so fascinating that as a fellow investor, in a startup ecosystem, I gave a lot of thought to when I was founding my syndicate and now emerging fund, you know, what’s the name going to be?
Is it going to be like AI ventures 2.0, machine translation, I would give a name and I came up with a name Data Power Ventures because I was thinking back to all the research where it all started, even when I used to do math competition on paper with no calculators, imagine figure there was a day that we, we did math on paper and in our heads, mental math, let alone, it all comes back to the data.
And so when I also came up with a thesis, you know, my passion on it, none of them were AI. Actually, and I talked to investors they’re shocked. They say, where’s the AI? I said, but I think you have to invest earlier. We have to look at infrastructure if you look at technology. And so I think this might resonate with some of the things that you talk about in your book Evil Robots, Killer Computers, and Other Myths.
So I talk about the fund and the syndicate, I talk about data-intensive applications. I talk about data, software automation, and talk about data developer tools. I talk about real-time insights and human, augmented workflows. I wanted to get your take. What’s your thoughts on some of these, if they resonate with you or you have some contrarian opinions?
Steven Shwartz
My first, big win, as an entrepreneur was an AI system that was in business intelligence, which is a data space. So I’m right, whether you on the on the data side and processing data in AI is a really important area as we both know because we share an investment in a particular company, named Y data where the emphasis is on how you get data quality into your AI models, what they do that’s really interesting, and I hadn’t actually focused on it until I talked to this company was there’s a big industry to clean data for tools like business intelligence that had been around for a long time.
And there are companies that are, multi-billion-dollar companies that provide data cleaning tools, data extraction, and so forth. And you would think that you could just take those tools, clean your data with them, and then feed them into an AI algorithm, to build an application. But that’s not the way it works. Because with AI, it has to be an iterative process.
So, you need to actually clean the data, try out some models, see what’s not working well. And then, you might have to go back and generate some synthetic data or perform other data quality operations and try it again. And it’s an iterative process. So those big platforms that are out there for business intelligence, don’t necessarily work for AI. And that’s an interesting insight that this company we invested in, has come up with.
David Yakobovitch
It’s so fascinating because when you look at these data developer tools, at the heart of the matter, they’re focused on experimentation. And, rightfully so, Stephen, you mentioned that it’s not just a one and done process, human-in-the-loop, data-in-the-loop. These systems are iterating and evolving over time, because, if they weren’t, then they wouldn’t be data systems, they would just be, as you mentioned, that this plain-vanilla software and data just complicates everything. Which is why I think in a data industry, we’re perhaps even a decade or more behind the software industry today, even though we’re emerging into this decade of data, there’s so much work to be done.
Steven Shwartz
Absolutely. And, slightly tangential to that thought. The human-in-the-loop aspect of AI is really coming to the forefront. If you look at AI in the medical area, for a long time, probably the first application of computer vision in the medical area was radiology. Everybody thought that, sure, with AI, you could diagnose illnesses from medical images, better than the radiologists. And it’s never actually worked out that way. I have friends who are radiologists who use those AI tools, and they say yes, sometimes they find things that I might have missed.
But at the same time, they miss things that we would have found. And that’s going to be true with most applications of AI. They don’t do things like people. So they do some of it better, some of it worse. But if you put the human-in-the-loop, and that radiologists can now take a closer look with the AI and find something that the human radiologists didn’t find, they’re going to provide better care, far better care because of the AI. But if you leave the AI on its own, and you try to get rid of the radiologist, it’s not going to work, because it’s going to miss some really important things. And human-in-the-loop is going to become a theme that we see over and over again, in the coming years.
David Yakobovitch
I really like that theme of human in the loop because I’m imagining right now that I’m a radiologist and I get, whatever scan in front of me previously, and I’m looking through it with my eyes and with magnifying glasses and, zooming in the computer and changing the resolutions and the negative and everything right to try to see if there’s an issue or not with the scan. Well, why should the machine just do it? The radiologist should still look at the scan.
But it’ll be augmented by these AI tools, while these rectangular boxes and circles and things that say with this 95% probability that AI thinks that, something’s going on in this scan in this image. And then the radiologist says, Okay, let me fact find, let me run some experiments. Let me see if this is right or not, and they can get to a diagnosis quicker without making those mistakes. I agree with you. That’s where the human in the loop would be really powerful.
Steven Shwartz
Absolutely. And if you look at a really charged subject, the use of AI in warfare, you can take that computer vision system and put it on top of a drone and put a gun on the drone and tell that computer vision system that when you find, terrorist acts, here’s his picture, go find them and shoot him. Well, sometimes it’ll get it right and sometimes they’ll get it wrong. If you’re going to do something like that. I think you should have a human that makes the final decision. That’s not how it’s going to end up working.
David Yakobovitch
Back in May 2021, we had on the HumAIn Podcast, Steven Umbrello. He’s a researcher actually on autonomous systems in the European Union. And he’s at the Institute of Ethics and emerging technologies. And we actually spent a whole hour talking about warfare, AI and warfare. And it was such a fascinating dialogue, everything about the design of the system, the accuracy of the system, the determinism of the system, the debate between humans on this innovation, and where are the nuances for autonomy.
So fascinating, because, one of the biggest insights I had during the pandemic, I was sharing with a colleague the other day, I said, during the pandemic, it seemed as if the whole world for one moment united together, we must battle and fight and destroy COVID. We all came together, we built vaccines, we did incredible research. And now the World, step at a time, getting back to a pre-pandemic world, in a digital-first society. But then, just in the last few months, countries are fighting again. It’s like human nature will be human nature.
Steven Shwartz
Yeah, it really is.
David Yakobovitch
So it’s like, can technology help there? And you bring up a fair point with this. Where is technology appropriate? Or not appropriate to help humans coexist without causing harm?
Steven Shwartz
And unfortunately, it’s just really unlikely that militaries around the world will do the right thing there.
David Yakobovitch
It’s the early days to see what with all this technology, and, thinking more, we talk about investments, we’re talking about businesses that you’ve scaled, and I’ve seen grow. And we’re in this new, we almost say technological Renaissance today, of startups, we’re seeing a lot of startups grow very fast, not only in valuations with venture-backed funding, but they’re building teams of 10s, 100s, 1000s of people. And, Steve, you’ve seen that scale, you’ve led that scale before, what are some of your take on the current technology ecosystem today that you’re seeing?
Steven Shwartz
We’re seeing a lot of the rollout of a specific type of AI, supervised learning, which is a type of machine learning. We’re seeing it applied in many different areas, I actually have a database I keep, for every time I see a new application of supervised learning. And it’s fascinating, it’s being used in almost every area of business or government of the nonprofit world. It’s fascinating how much application there is. But at the same time, it’s a very narrow technology.
And most of the AI companies that are growing up are building these applications where what they’re doing is they’re offering a product or a service. And the AI makes that product or service better, more competitive, perhaps, in the case of the shopping cart example that you gave before, maybe it was something that wouldn’t have been possible without that AI system without that computer vision system, which is the result of a supervised learning algorithm.
So, what you’re really looking at with these AI companies from an investor perspective, is an application. And investors need to make sure that they don’t let off track by the credentials of the people in the company, all the PhDs, they have all the complex AI terminology, and so forth. Because ultimately, what most of these companies are building is a product or application. And you can evaluate that company 95%, the same way you would if it was just a piece of conventional software.
In other words, forget about the AI for a minute, look at whether there’s a market for the product, look at the product-market fit, look at the team, look at the business plan, the distribution strategy, the finances, and the competitive analysis. And then you ask, okay, where does the AI fit? While it’s something that nobody else can do?
Okay, now you go in and you evaluate that claim as a competitive claim, but it’s one of the last things you do when you evaluate one of these AI applications, as opposed to getting caught up with all of the terminology and credentials and so forth. So, I’ve spoken a number of times to angel investor groups and I try to make that point.
David Yakobovitch
That makes a lot of sense. You can have all the PhDs, you can have all the cutting edge technology, the latest programming language, the best technology. But is it commercially viable? Can it actually be monetized, can it actually be built into a business? I know, one, the technologies we’ve seen of late open AI, they launched GPT-3, just in the last couple years, this massive dictionary, so to speak, of phrases that have relationships, and based on inputting one, you get some dialogue that’s generated, it’s actually quite good. It’s not perfect, but it’s quite good. And for a while everyone thought it was just this tool, okay, fun, type the word tennis, you get a paragraph and Serena Williams, how cool.
But now we’re seeing different startups using this GPT-3 technology to create business cases. One of those, Jarvis AI, helps with copywriting and copy editing for articles. So if you’re a marketer, you’re a blog writer, you can generate that article, and then maybe do a little quicker, get some innovation, get some inspiration and I’m not an investor in Jarvis AI, but I find that the business model, the commercial viability, really interesting there. And that may make the difference between great technology or great technology and the business.
Steven Shwartz
Absolutely as the business model that matters. I’m really interested to see how far GPT-3 takes some of these businesses. Because GPT-3, can generate, as you said, a new story or you can give it a prompt, and it’ll do a continuation of that prompt with something that’s going to be grammatically correct. And a lot of times, it sounds like it makes sense. But when you dig into it, a lot of times, the facts are all wrong, they’ve just kind of strung words together in a grammatically correct way, in a sequence that you might hear from a person.
So, when you look at using one of these tools to actually generate, say, blog posts or news stories, a lot of the facts are going to be wrong, and they’re not really going to make sense if you drill down into them. So, what’s going to be the implication of that it’s, is it only going to be useful if, there’s all kinds of search engine optimization, where you don’t really care if what you’re right makes sense, we’re going to generate a lot of crap using GPT-3 and put it out there for search engine optimization purposes? I’ll be interested to see how the tool is actually used out there.
David Yakobovitch
I do agree, actually, with that sentiment that there’s going to be more content created, and a lot of that content is going to create noise. And then that noise is going to get further fed into the next big language models, which will then maybe not be the golden source of truth that we’re hoping for, as these models keep evolving. But, again, it’s technology.
Like, okay, great, you got the next model with now 10 trillion parameters, one yo the by the parameters, okay, but is that going to create the AI robot? In the last year, I’ve seen now there’s been new startups, like always talking about launching something maid in your house, or this robot that can now clean a table, almost put dishes in the dishwasher and get a can of Coke. It’s interesting, but is it there? Is it really there yet? Are we living in the Jetsons?
Steven Shwartz
I don’t know, we’re far from that. Because the computers don’t really understand. So they can’t really have a conversation with you. Anytime they’re having a conversation with you, they’re effectively reading from a script, or using keywords to kind of determine what the input is from their list of possible inputs. In terms of being able to intelligently move around, while robots are still, at a fairly infinite level stage they can’t do that much.
They can’t, it’s very difficult for a robot to go and pick something up and put it somewhere that turns out to require a huge amount of programming and training. And so right now, that aspect of robots is fairly early on, although that part will be solved, what I don’t think will be solved will be the ability to have that conversation with Rosetta robot in a way where they really understand you.
David Yakobovitch
So, let’s make a forward-looking statement. We’re wrapping up 2021. We’re coming into the growth of this decade of data with modern data tools with excitement around all things data, ML AI, Steve, what are you seeing on the horizon? or what’s getting you excited again, about the industry.
Steven Shwartz
So, what I’m excited about as an investor over the next few years are tools that are going to help companies solve some of the social issues that are arising around these machine learning systems. So, for example, systems tend to be biased. And that comes from having biased data. The classic case being computer vision systems that are trained on mostly white males, they do a great job recognizing the faces of white males, but not such a good job on non-white males.
So, that’s an example of a biased system, the people who put the system together aren’t biased, but it ends up biased because the training set doesn’t have a diverse set of data. And what we’re seeing is, companies are having a reputational impact by putting out systems that are biased, either computer vision systems that are biased or hiring systems or loan systems, they’re also starting to run into regulatory issues.
So, in Europe, a system has to be able to explain its reasoning, if it makes decisions that impact people like hiring or loan decisions. So, all of these things put additional burdens on companies that are rolling out AI applications. And there’s a lot of opportunity for companies that are helping develop software and services to help companies build non-biased explainable systems.
And then you have a whole issue around when you build a machine learning system, it deteriorates over time. So, it might only work for a couple of days and then start to go downhill, it might work for weeks, but you have to monitor those systems and go back and retrain them when the performance goes down. And all of that is a lot of effort. So, there’s a whole area in AI called machine learning operations, ML-ops, that’s attracting a lot of attention. And that’s where we’re gonna see a lot of AI companies really grow over the next few years. And there’s going to be some really big breakthroughs in those areas.
David Yakobovitch
I completely agree. We were previously in HumAIn had Weights & Biases when they had raised their series A round and in common ML on their seed round. And these are just two examples of companies that have gone on to massive, not just rounds, but thousands of developers using those tools. So it is the early days. And it’s exciting to see where the industry goes. Steve, what’s next for you as well?
Steven Shwartz
I’m trying to decide if and when I’m going to do something next. Right now I’m working with some angel groups, I’m investing in startups, I’m mentoring some small companies and I’m doing a little bit of consulting. I’m an expert witness in a big case involving Amazon Alexa. And I’m trying to decide, do I really want to get involved in another startup? It’s so much work, but it’s so much fun too.
David Yakobovitch
It’s all about the startup itch which we both have. And it takes us from year to year to venture. It’s been my pleasure having Steve Schwartz today on HumAIn, author of Evil Robots, Killer Computers, and Other Myths: The Truth About AI and the Future of Humanity. You can check it out on Amazon, get your Kindle version there as well. Steve’s a fellow investor and a serial entrepreneur in the technology ecosystem. Steve, thanks so much for joining us on HumAIn.
Steven Shwartz
This was great, David. Thanks for having me.
David Yakobovitch
Thank you for listening to this episode of the HumAIn podcast. Did the episode measure up to your thoughts and ML and AI data science, developer tools, and technical education? Share your thoughts with me at humainpodcast.com/contact. Remember to share this episode with a friend, subscribe, and leave a review, and listen for more episodes of HumAIn.