The Exchanges

Q So I have a couple of questions really that stem from that. And why is it now in the last probably 18 months, you probably agree with me in saying, why is it that everyone from kind of CEOs in Detroit to head of Salesforce is interested in machine learning? What caused that rise in public interest?

A That's a great question. You know, one of the things that's been The slower trend is the fact that data is so much cheaper to store, and compute is so much easier to do now. And what that means is that people are able to worry about higher-level problems. So that's one, that's one thing. And then the other thing is that, in part, thanks to a set of guys who hit out in Canada for a little while, diligently working on a set of techniques that continue the work of, you know, folks from 40 years ago, we've got Cases where we can look at deep learning and say that neural nets actually work and produce really, really impressive results. Those two things, one slow and one sort of kind of dramatic because the results were so impressive, meant that people started paying attention again, and suddenly you would look and you'd say, wow, the work that Google has been doing for, you know, kind of quietly for the last 15 years is now yielding an incredible amount of fruit and creates all these opportunities. And the other part of it is there's this great investment in In large data systems. And then the natural question becomes, what do we do with it? You know, we have now these great analytics and we have this ability to figure out exactly how long it takes to, you know, sort of take one widget from one state to another. But what does that mean? And what are the actual applications that are …

AI assessment note: “Those two things... meant that people started paying attention again”

Q And talking about kind of group thinking there, I'm always very intrigued by herd mentality, and particularly now when we look at machine learning as a space, there's There's clear characteristics of herd mentality towards it. So with that respect, does the kind of mass popularization of machine learning, does it concern and irritate you as a machine learning enthusiast well before the herd arrived?

A You know, I'm always glad to have more company, right? You're right that certainly this is an area that lots of people care about and talk about. You know, to be honest, like wanting to be early, one of the reasons why I wanted to be part of a seed fund was so that we can invest You know, two to four years before everyone thought something was interesting. So I, I, for one, am very, very glad to have everyone else join the party. And I think that part of that means that we'll all get smarter about it. The truth, though, is that there still remains a relatively small set of folks who've thought about this type of business model, building these types of applications, and we're lucky to know a lot of them.

AI assessment note: “I, for one, am very, very glad to have everyone else join the party.”

Q So I have a couple of questions really that stem from that. And why is it now in the last probably 18 months, you probably agree with me in saying, why is it that everyone from kind of CEOs in Detroit to head of Salesforce is interested in machine learning? What caused that rise in public interest?

A That's a great question. You know, one of the things that's been The slower trend is the fact that data is so much cheaper to store, and compute is so much easier to do now. And what that means is that people are able to worry about higher-level problems. So that's one, that's one thing. And then the other thing is that, in part, thanks to a set of guys who hit out in Canada for a little while, diligently working on a set of techniques that continue the work of, you know, folks from 40 years ago, we've got Cases where we can look at deep learning and say that neural nets actually work and produce really, really impressive results. Those two things, one slow and one sort of kind of dramatic because the results were so impressive, meant that people started paying attention again, and suddenly you would look and you'd say, wow, the work that Google has been doing for, you know, kind of quietly for the last 15 years is now yielding an incredible amount of fruit and creates all these opportunities. And the other part of it is there's this great investment in In large data systems. And then the natural question becomes, what do we do with it? You know, we have now these great analytics and we have this ability to figure out exactly how long it takes to, you know, sort of take one widget from one state to another. But what does that mean? And what are the actual applications that are …

AI assessment note: “neural nets actually work and produce really, really impressive results”

Q I so agree with you with regards to that challenge, and the question then that's posited to me from that is, will we have machine learning in a box that could be integrated into the majority of most corporate systems provided by the likes of IBM? Will it be a kind of box solution, or will it be a personalized one-off per company solution, do you think?

A There's always been the dream of making programming so easy that everyone will be a great programmer, and I think similarly there's a dream that it'll be really easy to build models And I think what actually happens is when everyone's able to program, you end up with really bad programs. When you have these little black boxes that are not thoughtful about what data they're taking in and how they're using it, you end up with basically bad analysis. My own bet is that you'll see a lot of that happening over the next few years as the number of high profile projects inside corporations end up not yielding fruit in part because the problem was not defined well enough, or there was an assumption that If we just trust this black box without being thoughtful about processes around it, we'll end up with sort of terrible outcomes.

AI assessment note: “When you have these little black boxes... you end up with basically bad analysis.”

Q And talking about kind of group thinking there, I'm always very intrigued by herd mentality, and particularly now when we look at machine learning as a space, there's There's clear characteristics of herd mentality towards it. So with that respect, does the kind of mass popularization of machine learning, does it concern and irritate you as a machine learning enthusiast well before the herd arrived?

A You know, I'm always glad to have more company, right? You're right that certainly this is an area that lots of people care about and talk about. You know, to be honest, like wanting to be early, one of the reasons why I wanted to be part of a seed fund was so that we can invest You know, two to four years before everyone thought something was interesting. So I, I, for one, am very, very glad to have everyone else join the party. And I think that part of that means that we'll all get smarter about it. The truth, though, is that there still remains a relatively small set of folks who've thought about this type of business model, building these types of applications, and we're lucky to know a lot of them.

AI assessment note: “I, for one, am very, very glad to have everyone else join the party.”

Q We mentioned that, uh, you, you said about the theological challenges related to Kind of the future of work and what will we be doing? And then also the kind of Elon Muskian, will it destroy the world as concerns? You said there were other more immediate challenges. What do you perceive those immediate challenges to be?

A The biggest challenge out there right now is that there's not really a good metaphor for thinking about how to apply machine learning. I think that we watch movies that will tell us about falling in love with the robot, or we'll read economists telling us that things are falling apart. But the real opportunity, I think, is figuring out specific processes and specific places where building good bottles to help us make better decisions, or actually, oftentimes, to make decisions for us. That's the actual interesting opportunity out there right now, in part because I think machine intelligence is not going to be, at least in the near term, you know, one big intelligence, but as it is for our own brains, many small intelligences and many, many small systems that help us make Good decisions or make decisions for us and do things on our behalf. And I think that's the identifying that and also figuring out both where the low hanging fruit, but also critically for a lot of corporations, what's actually possible. Figuring that out is a hard challenge in front of us. And I think nobody knows what they're doing yet, right? There are still huge questions around how to think about machine learning, about, you know, what it even means to build these models. That's the challenge in front of many of us.

AI assessment note: “The biggest challenge out there right now is that there's not really a good metaphor”

Q I so agree with you with regards to that challenge, and the question then that's posited to me from that is, will we have machine learning in a box that could be integrated into the majority of most corporate systems provided by the likes of IBM? Will it be a kind of box solution, or will it be a personalized one-off per company solution, do you think?

A There's always been the dream of making programming so easy that everyone will be a great programmer, and I think similarly there's a dream that it'll be really easy to build models And I think what actually happens is when everyone's able to program, you end up with really bad programs. When you have these little black boxes that are not thoughtful about what data they're taking in and how they're using it, you end up with basically bad analysis. My own bet is that you'll see a lot of that happening over the next few years as the number of high profile projects inside corporations end up not yielding fruit in part because the problem was not defined well enough, or there was an assumption that If we just trust this black box without being thoughtful about processes around it, we'll end up with sort of terrible outcomes.

AI assessment note: “you'll see a lot of that happening over the next few years”

Q This is a meta question for you now. What do you think the optimal relationship is between man and machine? You know, Musk is terrified and then others are terrified about kind of too much integration and too much love.

A So I don't, I think that that's the wrong way to think about it because it's not, you know, machine is just the metaphor we use right now for technology. The truth is that we as people are Constantly being transformed by technology and, you know, whether that's agriculture and eating, you know, eating wheat and having that affect our tummy to deciding to wear clothes because it keeps us warm, right? And so, I think that that, you know, sort of, there's gonna be inevitable progression of how different technologies end up influencing us. And, you know, in some ways, you and I are incredibly augmented by a whole set of things, you know, sort of everything from living inside buildings to, like, using light. To be honest, I don't know that AI is gonna be necessarily that different. Which is say all those other technologies were incredibly transformative, and this will be another example. And people who come from a thousand years ago will look at us and sort of marvel and be confused, but we will still feel like we're human.

AI assessment note: “I think that that's the wrong way to think about it because”

Q We mentioned that, uh, you, you said about the theological challenges related to Kind of the future of work and what will we be doing? And then also the kind of Elon Muskian, will it destroy the world as concerns? You said there were other more immediate challenges. What do you perceive those immediate challenges to be?

A The biggest challenge out there right now is that there's not really a good metaphor for thinking about how to apply machine learning. I think that we watch movies that will tell us about falling in love with the robot, or we'll read economists telling us that things are falling apart. But the real opportunity, I think, is figuring out specific processes and specific places where building good bottles to help us make better decisions, or actually, oftentimes, to make decisions for us. That's the actual interesting opportunity out there right now, in part because I think machine intelligence is not going to be, at least in the near term, you know, one big intelligence, but as it is for our own brains, many small intelligences and many, many small systems that help us make Good decisions or make decisions for us and do things on our behalf. And I think that's the identifying that and also figuring out both where the low hanging fruit, but also critically for a lot of corporations, what's actually possible. Figuring that out is a hard challenge in front of us. And I think nobody knows what they're doing yet, right? There are still huge questions around how to think about machine learning, about, you know, what it even means to build these models. That's the challenge in front of many of us.

AI assessment note: “The biggest challenge out there right now is that there's not really a good metaphor”

Q Where are the greenfield opportunities in machine learning for you?

A Oh, there's so many. There's so many. I think that if you were to break down every single process inside an organization, and you'd say, which are the ones that are easiest to digitize? That exercise is sort of the exercise that we go through right now. And I think that everything from Redoing logistics over to HR, you know, sort of rethinking the way that contracts are being written. All those are huge, huge opportunities. And then there are a whole set of opportunities around rather than selling products into specific companies and for one specific function, there are a whole set of companies that can be remade because that one core function will be so valuable. And so we look a lot of that, that part of the world I'm certainly quite excited about.

AI assessment note: “everything from Redoing logistics over to HR, you know, sort of rethinking the way”