This is a MarketScale Software & Technology Podcast Series, hosted by Daniel Litwin. This is the first episode of a three part series titled Bringing AI to Businesses with Ben Taylor, Chief AI Officer & Co-founder of ZIFF Inc. In each episode, we’ll explore different aspects of AI’s push into business-operations ubiquity, from its most useful applications to the surprising business ethics that come with implementation. Each episode will also feature a short article from Taylor, which you can read below.
How Do You Deliver on an AI Project, Both As an Executive and a Data Scientist?
Many executives are naive when it comes to AI capabilities and navigating where AI might provide value to their business. The data scientists aren’t helping either, where most struggle to communicate value to the business representatives. Most data scientists also lack urgency; they have no pressure to last-mile AI into production. Funding science projects will accomplish one thing: covering the tuition of your data science team so they can land better jobs at Facebook or Google. So, once the trigger is pulled and you have a team prepped and passionate about bringing AI to your business…how do you make sure everyone delivers so you avoid wasted time and money?
Avoiding The Science Project Landmines
If you are an internal advocate for AI, do everything you can to constrain the timeline. Ask yourself: Is there anyway to do an internal proof of concept in 60 days instead of 12 months? What can I do to reduce internal budget? What can I do to reduce the number of people required? The more you reduce these variables, the more likely you are to get buy-in from the internal business units.
I’ve always been a fan of leveraging outside hardware companies, consulting groups, or AI platforms to shorten do-ability tests.
Crawl, Walk, Run
Some AI projects fail because they are too ambitious. They don’t have a short-term proof-point, and the complexity comes tumbling down like a house of cards, revealing a project that had no clear goals, tangible value or structure. This flaw can come from inexperienced data science teams that are too “academic,” where they are more interested in a challenging thought experiment than a Bayesian method in production. If you can carve your project up into bite size milestones, your chances of success are higher. It shouldn’t be ignored that AI projects aren’t a one-and-done either; you have the advantage of improving on your algorithms. Just look at the evolution between AlphaGo and AlphaGo Zero, and how they would’ve never achieved such a grand level of “reinforcement learning” without trying a few, more tangible methods first. Get some novice wins into production and then level up on subsequent versions.
It Is Harder Than It Looks
Getting a successful AI project to value is much harder than it looks. Most major wins for AI are behind six to 10 iterations on the same problem. We see successful companies solving the same problem multiple times, where each time they solve it they understand the data set and problem a little better. Once a project has crossed a predefined criteria for success, taking that AI project into production can create additional problems. Supporting AI in production requires quality monitoring (e.g. did your incoming features drift) to ensure models are behaving as designed. This requires an involved data science team. And yes, I said team. Collaborate, get multiple eyes on the project, and make sure everyone is on the same page before launching something into production. You don’t want your AI project to end up like the Mars Climate Orbiter: dead in the air because of a unit conversion mistake. Double check, triple check, and then check again that the final product is in line with the initial vision you set up for success. Feels a lot like simple project management, huh?
Highlights from the Episode
Daniel Litwin: That brings me to my next point, which is, looking at some of these big projects, ones that don’t really have anything to do with AI for a business and how data scientists might use those to help convince executives to pull the trigger on implementing AI. And, the main one I want to talk about is AlphaGo and AlphaGo Zero, which I was looking into and it’s pretty amazing. You have this AI program that learns how to play the classic game of Go, the Chinese game. AlphaGo, the original, would learn from humans. It was given historical data from some of the masters of the game; “this is what they did, this is how they pulled off their strategy,” and then the AI obviously surpassed that.
Then, it was introduced to the idea of AlphaGo Zero. This new iteration of the AlphaGo Zero demonstrated that an AI can learn and surpass basically anything a human has been able to do without even using humans as the starting point; it learned from scratch. It was just playing itself the entire time and within 40 days it had already completely destroyed any record of any human player of Go and any other AI player of Go, it just became the super champion. It’s pretty crazy how much technology can improve from one iteration to the next of a project.
I think this project really exemplifies something that a data scientist might be able to show to an executive to maybe better comprehend the timeline of creating an AI project. The astounding capabilities of an AI project. Walk me through a little how a data scientist might use something like this to help an executive pull the trigger on something new and fresh for AI in their company.
Ben Taylor: Yeah, this is a great example, there’s actually a lot going on here for us to pull on. To just drop this example on an executive’s lap isn’t really going to do anything for the data scientists. But there are some certain things they can pull on. One of them that I really like bringing up: AlphaGo Zero had no human insight, there was no human features that humans had engineered or thought about to really assist with that AI to do as well as it did. We’re seeing this in business, where a lot of times in business we respect these human features because we’ve spent two decades coming up with them, that a business executive might think that the human features and insight into there are superior to what an AI might provide. But the thing with humans and the way that we think, we have to bucket complex data sets. You’ve heard of Big Five Personalities if I’m doing some type of persona or hiring assessment, I might try to assess you for these five personalities. The reason there’s five and not 5,000 is because humans were constrained, you can only do so much research and it’s really hard for you to have a continuous number of personalities. But for AI that’s possible. We have the exact same thing with emotions on the face and then also phonemes and parts of sound. Humans have discrete ways of understanding data where AI can come in, like AlphaGo Zero, and find new features that the humans would never be able to find and provide a lot of value for business.
There’s one other thing too that’s happening there. A business executive, you really need to give them a case study that is so close to home, it’s essentially their case study. The problem with that is if you’re waiting around for that to happen, I would say you’ve waited a little too long. Because if you have this great clickthrough rate case studies falling on your lap, that means most of your competitors already have that in production. Sure, every year the bar to get into AI keeps going down, but do you want to wait until it’s all the way down to the ground floor where you were the last one to adopt? I think you’ll see some businesses will and they’ll get into trouble for that because they’ll be dealing with market disruption before they have to get AI into production.
DL: Right. Do you really think that these…I don’t want to call them a toy project in a way to demean them, because what AlphaGo Zero accomplished is really, really incredible. But, at the same time, I think for a business executive, they may not see it as, “how is this going to apply to my business?” But, at the same time, it’s this strange counterintuitive thing where actually showing them AlphaGo Zero and the capabilities of what that AI did, you can at least demonstrate the unlimited capabilities of what AI could bring to your company. Do you really see toy projects like that helping convince executives to pull the trigger?
BT: We have, which has been really surprising. I think a year ago I would have told you, “no, that would not be a useful conversation piece.” But, for us, for some of the marketing pieces we’ve done, they have no business application at all. So, we did an ABC Bachelor/Bachelorette ranking using AI on the new season. We’ve had executives react to that because it seems like that should be impossible. How on earth could you look at a single face and then predict the season rank and predict the winner?
The other one that people have reacted to, we are showing off some of these genetic GANs where you can…what we do is we have target models where AI has been pretrained to recognized gender, race, age, emotion, attraction, all of these different things about a face. We can now manufacture these faces to order and we’ve had big reactions from that. We’ve also had reactions from Robot Apocalypse pieces from executives, where you’d think this has absolutely nothing with business. Hopefully it never does, but it may in the future.
Even though these executives are smart, they’re not going to do a science project, they’re still this emotional geek, a nerve that you can hit on. It doesn’t mean that they’re going to sign up for a year-long science project, but you can at least begin to have a conversation about a short-term AI win. I would say across the board, most business executives are naïve with the current AI capabilities, because they see it at a high level. Maybe they’ve heard of AlphaGo, I’d say most of them probably have not. You have self-driving cars that are still having issues. You have some cancer research that’s not really prime time yet. They kind of see these points and I think they take the stance that AI is not ready, it’s not prime time. But, if you can show them a few of these examples to startle them and shock them, that opens up the conversation.
DL: What’s also great about showing them these kinds of projects, especially with AlphaGo Zero, is it’s a testament to the power of iterations, of versions of AI. That even though we are going to deliver on the first one, that doesn’t necessarily mean that is the end all, be all. Often times we find something, we change our perspective on how we approach implementing this AI to your business and that one small change ends up producing magnanimous returns for your company. It’s that kind of conversation, knowing that AI isn’t, “you implement it and it’s done,” but it’s a fluid process and there are versions you need to continue to find the ways that it’s going to benefit you uniquely that I think help deliver on the project too. Because it helps executives understand that it’s something that can grow with the company.
BT: Yeah. I’m really glad you brought that up. We have definitely found in the last couple of years that iterations are key. I think a lot of people think with an AI project, if you’re going to go and do Project A, it’s a one and done attempt. You get the data set, you build an AI model, you ship it to production and you on to the next. Our recommendation now, we tell people you are looking at six to seven iterations on the same problems. Let’s just round up to ten and say, if you have a Project A, for it to be incredibly successful, you should be ready to do ten iterations on that project where you’re getting different versions of the data set and building models reacting to it. For the partners that we’ve seen that have done that, they’ve had tremendous success. But, if you looked at their first, second and third iterations, the results were marginal. They were okay, they weren’t transforming their business.
For the latest news, videos, and podcasts in the Software & Electronics Industry, be sure to subscribe to our industry publication.