Josh Sutton has been helping clients leverage advanced technologies to drive business transformation for the past twenty years. Most recently he has assumed leadership of Publicis.Sapient’s Artificial Intelligence Practice, helping companies leverage established and emerging artificial intelligence platforms to generate business insights, drive customer engagement, and accelerate business processes. With more than 22,000 people around the world, Publicis.Sapient, part of Publicis Groupe, is the world’s most advanced and largest digitally-centered platform focused exclusively on digital transformation and the dynamics of an always-on world. Some representative AI programs that Josh and Publicis.Sapient have been involved with include the following:
- Leveraging a semantic AI platform to generate unique customer insights for a global wealth management firm
- Creating next generation virtual assistants for a wide range of companies including leading theme parks, automotive firms, and retail banks
- Using machine learning tools to optimize media planning and buying activities for large CPG and airline companies
- Using a combination of AI technologies to build “smart search” tools for global energy and financial services firms
Byron Reese: When did you, when did artificial intelligence first appear on your radar? When do you recall being first intrigued or lit up by it?
Josh Sutton: From a personal point of view, that would really go back to my early childhood. I think artificial intelligence has always been one of the things that I’ve been passionate about. As somebody that was always fascinated by technology and what you can do with it, subsequently going to school at MIT, it’s always been in the background of my thinking. But from a more recent point of view, coming out of the most recent AI winter when I really started to look at artificial intelligence, and when we as an organization started to realize that it might be time to start getting much more intelligent about what is and isn’t possible, was in the time frame of 3 or 4 years ago.
At that point in time, I was running our capital markets business, and as you would expect some of the leading hedge funds in the world were dealing a lot with what today we would call ‘machine learning’. At that point in time, they were just calling it ‘very smart technologists doing very smart things to make a lot of money’. We were being asked more and more frequently, “What is actually going to be possible?” and “What are some of the technologies that are coming to the forefront?”
Then as we really started to move forward, you had IBM making a big media splash with Watson winning Jeopardy. The momentum continued to pick up with the acquisition of Deep Mind by Google in 2014. All of these things started to pave the way for us to realize that we needed to get deadly serious about how this was going to impact the businesses we operate with and all of our clients, as well as our own business. So, we started looking at what are the different technologies out there. ‘Artificial Intelligence’ is, unfortunately, a horrible term, but one that we seem to be stuck with. But, as you alluded to, it’s a general catch-all for a wide range of different types of platforms, and often times if something becomes well understood, it ceases to be ‘Artificial Intelligence’ and starts to be a different type of labeled technology.
Now, with that as a backdrop, what we’ve done within Publicis.Sapient is focus on two primary areas. One is looking at the technologies and tools that are available, and figuring out how we can deploy them to transform our own business, and be the absolute market leaders when it comes to the application of AI-related technologies to the field of advertising, marketing and communications. The second area that we’ve chosen to focus on is helping all of our clients on the consulting side understand how to put together an enterprise architecture that’s focused on AI, cognitive, whatever terms you choose to use, but really enables technology to drive insight generation, drive acceleration of business processes, drive business transformation through technology capabilities that weren’t possible 5 years ago – helping people understand what’s actually possible. Because this is a field where, even by the fairly generous standards of the technology industry, I think there’s been some fairly egregious discrepancies between what companies claim they can do and what they can actually do. So helping our partners really penetrate that myth of what’s real versus what’s hype, and how do I put that together to solve the business problems and challenges I’m facing today? So, let me stop there, because that was a fairly long monologue.
Where are you in your product life cycle?
Well, we’re not building products. We’re partnering with all of our clients, and putting together the right solutions for what we do internally. So, as it relates to that first part I talked about, around applying AI to our business, I think we’ve made some remarkably substantial strides in applying machine learning to the field of media planning and buying, and what we can do to really shift the entire focus of the media industry, from the segment level to an individual level. That sounds like a very simple thing, but it’s something that wasn’t possible to be done by humans, and wasn’t possible to react in the time frame, that you need a lot of time to really create the value that you want. So we’re able to generate returns for our clients right now that are extremely meaningful, saving them tens, and sometimes hundreds of millions of dollars, from advertising expense, by using these technologies. A lot of that is proprietary, how we’re using the different tools and applying them, and some of that is leveraging data that we’ve also put in place. From a partnership point of view, we partner with north of a hundred companies to aggregate different individual level data attributes on people that we are seeking to advertise and market to. So that’s been something that has been very fruitful for us, and continues to be. But we’re always on the lookout for companies that are building products that do it as well, because we’d much rather buy than built. That being said, I don’t think anyone’s put in nearly the focus on this particular space that we are.
On the other front, which is what are we doing for our clients, there’s 3 or 4 main areas that clients are coming to us right now from a generalized use case point of view. The most common, as you would expect, is around the entire virtual assistant chatbot space, and we’ve been working with all of the technologies that claim to play in that space, which is growing literally by the day. I feel like there’s always somebody new that claims they’ve got the next best mousetrap. In general, what we’ve found is none of the core platforms on their own do everything that you would hope for. Especially given that people have expectations, in some cases, driven by the media, watching movies like ‘Her’, where you have a virtual assistant that sounds like a real person and acts like a real person. Now, we’re still away from that. But we’re making some real progress in being able to put different platforms together, and combining some of the virtual assistant frameworks and platforms with some of the more robust cognitive platforms, and some of the larger semantic players to produce some things that come close to that, if not there all the way, but a lot closer than what you see in production today. So, that’s been exciting for us and a lot of our clients that we’re working with.
Another big area we focus on is, well, obviously, the targeted media that I’ve talked about, but also the automation of manual processes. This is an area that, strategically, we’re hoping a lot of companies look at, and people always go to cost cutting, as the reason you automate processes. What I’ve found consistently is that the companies that are really benefiting from these types of programs are those that are focusing on taking tasks that typically take an extended period of time, and reducing the time it takes to complete them, for a strategic advantage in the marketplace rather than just the cost saving component.
And the third area is around the extraction of insight from unstructured data. Now, IBM calls this ‘dark data’, but the problem that most companies deal with today is, 80% + of the data that you have access to is unstructured, and the amount of data is roughly doubling every year, so for companies to make intelligent, strategic decisions, leveraging the data they have access to, they need to figure out a way to systematically extract insight from that data, and that’s something that we’re really getting a lot of traction with. Particularly with some of the semantic platforms, which are a bit less well-known than the Google’s, Microsoft’s and Watson’s of the world. These technologies are often coming out of academia.
You say, with regard to your passing reference about the movie ‘Her’, you say we’re a little bit away from that. You mean literally we’re a little bit away, or do you mean we’re far away?
No, we’re a ways away from that. But I do think what we are doing, looking at how we can create virtual assistants that handle open-ended questions and conversations better than the frameworks today, is moving us in that direction. So it’s a baby step in that direction, but I don’t want to overstate it. It’s still a huge road between here and there.
And do you believe we can, do you believe we will build an AGI.
Absolutely, I absolutely believe we will build an AGI. I think it will be from a different framework than what we look at today. This might go down in the weeds a little bit, but if you look at all the major players right now, and by major players, I kind of group 6 of them together that are north of a billion dollar spend type of category, and that’s Google, Microsoft, IBM, Amazon, Facebook and BaiDu, and while there’s lots of other players trying to move into that, I still think those are the big six.
Right. And you have people, just in these 6 companies, you have a range on when we’ll get an AGI, from 5 years in one case to 500 years in another case, within those companies. Why do you think there’s so little, why do you think there’s so much disparity on when people think we’re going to make it?
I think it’s because everybody that’s close to the technology understands that where we’re going with machine learning, and I group machine learning as a very broad category that encompasses everything from your deep learning neural nets, through to your very basic algorithmic approaches, that is very good at enabling platforms to pattern match, and to learn based on experience. But what it doesn’t enable is the application of common sense and inference and reasoning about how the world works and the way you and I process information. So, that’s, to a certain degree, the dirty little secret of AI. There was actually an article in the Wall Street Journal a couple of weeks ago, where Yann LeCun from Facebook was quoted as highlighting the gap. Now the question becomes how do you fill that gap? Because if you proceed on the thesis that what the platforms today can do, and will be able to do in the not-too-distant future, is outperform humans, if they can’t already today, at learning from pattern recognition. To augment that, at a minimum you need the ability to apply common sense and a human construct of how we view the world to do that. So you cannot just learn from pattern recognition, but you can apply that to a framework, and extrapolate insight from that. That’s where you have a bunch of different people pursuing that over time, Ray Kurzweil has been a very vocal advocate that that’s where the entire singularity conversation happens. You’ve had people like Doug Lenat dedicate his life to it with Cyc and Cycorp, and what he’s been trying to build there. What Marvin Minsky was doing at MIT with ConceptNet is a big enabler of that, and a lot of those technologies. The Cyc’s and ConceptNet’s are not talked about as much, because they’ve been coming out of the academia and they aren’t as commercialized, they’re not robust at the same level as say TensorFlow or Deep Mind is, but directionally they’re moving in the right place to augment machine learning, and I say ‘augment’, not replace. But to augment machine learning with that common sense capability, and when you combine the two of those things together is when I think you’re going to see something that much more closely resembles AGI. Now, there’s lots of different schools of thought about how you do that and what the layer beneath that is, but frankly that goes way beyond my technical depth. But I think that’s why you have such a wide range of opinion about it. There are some people that philosophically just do not believe that what Minsky and Lenat and others have been doing will ever be successful. There’s other people that fundamentally believe that they are correct and the roads they are going down will yield results sooner rather than later. And then there’s a big chunk of people who say, ‘You know what? There’s a gap missing, and humans, by human nature, will learn how to fill that gap, but it’s going to take some time to do it, so we’re probably in the decades rather than years type of window.’
What is your take on the Chinese room problem, which argues that computers can’t really ever be truly intelligent? That since there’s nothing in the computer that ‘understands’ anything, you can’t actually have an AGI. [Note, this is a classic argument against the possibility of a general AI put forth by the philosopher John Searle. It is worth looking up in Wikipedia. But the basic idea is that because a computer is completely mechanistic, it simply follows programming. No matter how clever it looks, it doesn’t really understand anything.]
That looks like a very philosophical argument, or debate, which is probably very germane to the book you’re writing, which is ‘Is the human mind a machine or not?’ Do you process information according to a set of hard-wired objectives, guided by the rules that we learn, or is there something more than that? And I will not even attempt to answer that one on the record.
Do you think computers will ever attain consciousness?
Depends on how you define consciousness. If you mean, ‘Will they be able to interrogate themselves and question whether a hypothesis they put forward is correct or not, or is aligned with the objective they’re trying to achieve?’ I believe the answer’s absolutely yes. Will they have emotion the same way that you or I have emotion? And relate to that? I think that gets into a realm of the ‘it’s possible, but unlikely’, because I think that’s something that, the concept of emotion would be to get into the human brain and look at is it a set of rules or not? I believe emotions are, to a certain degree, one of the hard-coded things that help steer us from a driver point of view, from a subconscious point of view. And that’s a way we process information and how our hardwiring works. But when you get into computers, will it mimic that? Very likely. Will it be the same as you and I? Difficult to say. I would lean toward ‘not’, but I would lean toward the action of what you would consider consciousness, of being self-aware as a possibility, yes.
Do you have an opinion on how quantum computing is going to lay into the creation of an AGI?
I think what quantum computers are going to do is just dramatically increase the processing power. I believe that one of the challenges that you’re going to have, if you go down the road that we are today, which is a very machine learning-centric road, is incremental processing power is not going to give much more yield than what you have right now. Because you’re already constrained, to a large degree, by training data. And lots of people talk about having big data, but the reality is very few companies truly have big data at the scale that’s needed to drive machine learning, and get to the point of understanding that you and I operate within. Most of the technologies today require a large amount of data rather than small. When you start getting into the semantics and reasoning engines of some of the Minsky-esque type of philosophies, you start to reduce that, but not a lot of companies are going down that road yet. And although I think all of them in the background are, they just aren’t making a lot of noise about it.
So, you were very clear in your statement earlier, that the most successful practitioners of AI in your estimation are using it to do a process more efficiently, or speed up a process, as opposed to cutting costs. Do you think that on net, these technologies that we’re building right now produce jobs or destroy jobs, on net?
On net, I think it’s neutral. I think in a relatively short period of time you’ll see more job destruction than creation. Over an extended period of time, it will be a net neutral, and I base that on historical precedent. Every time we’ve had a major change in technology that’s changed the way society operates, we haven’t worked less, often times it’s back to work more, but we’ve been much more efficient and able to do more, and have a better quality of life resulting from it. I think this is no different than previous industrial revolutions, in that it will drive a higher quality of life. Over time it will drive the creation of new fields that, frankly, we can’t even fathom today, not having it as a part of the fabric of our society yet, but will seem intuitively obvious in twenty years. But the one thing that’s different is the rate of change that’s happening, it’s much faster than any other industrial revolution. So if you go back to Davos, the fourth industrial revolution construct, one of the big challenges with this one is that it’s happening in a period of years rather than decades or centuries.
Are you optimistic about the future?
Absolutely. I think, as a species we’re hardwired to continually try to improve ourselves, and improve our circumstances, and improve the environments within which we live. I think that, while there will obviously be exceptions to that, and always are, in how people use technology, we will use these technologies to improve our way of life.