The promise of artificial intelligence is that it will make our lives easier. The fear about AI is that it will render us unnecessary and then make a tasty stew of us. Rest assured this introduction is not written with help from ChatGPT, because I for one will not be the architect of my own demise.
Which brings us to Callaway’s newest irons: the Apex Ai200, Apex Ai300 and Apex TiFusion. They continue the company’s cutting-edge endeavors to employ AI to help create the leading-edge designs — here, in the forged chasses long appreciated by better players.
But what, exactly, does using AI in club design entail, and what are the benefits? GOLF.com spoke with Brian Williams, Callaway Golf’s VP of R&D, to find out.
GOLF: It feels like everyone is selling AI in some form. What does it do for Callaway to help you make better products?
WILLIAMS: AI helps us make custom and tailored, fully optimized solutions. Our platform has been being built for 15 years by a team of a dozen data scientists who are continuously evolving and expanding on it, to teach it to do new things. It’s not at a point where it’s autonomous enough to say, “I’ll go figure out now how to make the next thing better, I’ll take it from here.”
We are having to teach it how to do those things, what the next steps to take might be, and how to take them. And we’re continuously building to be, I think, a step ahead of what everyone else can use it for. We have competitors that are talking about, “We use it to optimize for CG location.” That’s work that we would have done seven or eight years ago. We’re now into this space of modeling dynamic collisions. I see us having a leadership position in the science that goes behind it, and in the platform that it runs on.
GOLF: You can’t just go to ChatGPT and tell it to build a faster club….
WILLIAMS: It’ll tell you, “Look at mass properties. Look at your static spec package.” The AI can use those kinds of informational inputs, but it can’t know how to model a golf club unless you’re building a platform that shows it how we do it, how we simulate it, and what we’re looking for as results — and then build that iterative looping process that allows it to learn. That’s where it comes in, as a learning mechanism of, “Okay, I tried this, and this worked but this didn’t. Now I’m going to try something else.”
It’s connecting these dots in a complex way that our engineers just couldn’t do. Fundamentally, the inside of our face of our driver looks the same today as it did five years ago. We have varying topography. You would never intuitively know that the face that we have today is going to be better than the face that we had five years ago. It’s better because AI is solving for these details in construction that matter in a way that’s not intuitive, and that a designer just wouldn’t be able to do.
$200
View Product
GOLF: You mentioned dynamic collisions — actual golfers’ ball-club contact rather than robot testing, theoretical/neutral impacts, and so on. This real-life data that your AI is using about where players strike the ball on the face, club path, and all that — where do you get it from?
WILLIAMS: Thank you for asking that — that’s part of our advantage as well. We’ve connected data systems from all over the world. We’re using local data from our ECPC, our Ely Callaway Performance Center, with people who come in for fittings. We’re using that from tour staffers we’re fitting. We’re using it from elite amateurs, what we call our “Gold Panel.” All that is local data.
Outside of that, we have Callaway Performance Centers and fitting studios all over the nation. We have a data system that’s bringing fitting data back to R&D from Callaway fittings, and we’ve now expanded that internationally. So, we have significant reach now, millions of data points on delivery that’s coming from tens of thousands of players. And we can segment that to talk about, “What players do we want to go look at?” “Are we looking at a universe of players?”
That’s where we started. Now we’re going to where we have targeted, segmented player types, to be able to say, “For this product, and specifically in this model within this product platform, here are people that have been fitted into a similar thing in the past.” So, we know who the Apex player is, and we know who purchased a DCB iron and would be a great candidate for an Ai300. We can have really targeted data that flows into our models.
GOLF: It seems like we keep seeing the term swing code as part and parcel of dynamic collisions. It’s essentially an overarching term for swing speed, club delivery, face orientation and so on?
WILLIAMS: It’s kind of the building blocks of a swing. The fundamental seven or eight things that you can capture and categorize with today’s fitting technology. We can capture your head speed, your path, your angle of attack, your face angle, your face relative to path. We can look at dynamic loft, dynamic lie. If we have those kinds of inputs, we’re able to say, “We know how your face is coming into the ball at impact. We know if it’s a glancing-across blow. We know if it’s down and across. We know if it’s up on the golf ball.”
What I learned getting fit for Callaway’s Paradym Ai Smoke irons | ClubTest 2024
All those things matter for how the golf ball comes off the face. So, swing code for us is talking about how that club is delivered at impact, or just prior to impact, so that we can give AI a more intelligent starting point for the forces in play between the face and the golf ball. By doing that, it can decide how it wants to flex and what’s going to be best for that kind of player.
That’s a big thing we’ve seen in our benefit for, say, our Ai Smoke MAX D driver. It’s helping a player turn the ball over. Fundamentally, this is a player who’s going to buy this club because they don’t want to slice it anymore. They want to be able to turn it over. Well, everybody’s done that in the past with mass properties. Put a lot of weight in the heel, make it draw biased, help them close the face, and that’s the recipe. With AI, we’re now talking about, “Yes, we still want to be draw biased, we still want to have an optimized mass property package — but we know this player’s coming in down and across with a face open to path.” The face is not catching the ball squarely all the time. That’s why the player slices.
So, let’s not model for a square, neutral impact. Let’s model for a down-and-across glancing blow. What’s going to happen with sidespin, and how can we use face deflection to resist the natural tendency to have a left-to-right sidespin? If we can reduce that across the face, they’re going to hit straighter shots and benefit from it. It’s this additive effect of not just a traditional mass property spec-driven package, but we’re now able to model a dynamic collision and talk about how we can further improve launch trajectory and the conditions that your golf ball comes off the face.
GOLF: In terms of some of the specific AI-driven technologies in the new clubs, can you explain what “micro deflections” are?
WILLIAMS: Sure. In its purest form, when we’re talking about a golf clubface, we’re talking about tens of thousands of tiny points. We can model for each of those points to be as thick or thin as we want to allow. We used AI to create a topography, and what we’re doing is in changing the topography of the face, and in varying thick-to-thin transitions, we’re able to get the face to deflect in different ways all over the face. When we’re talking about micro deflections, we’re talking about at a specific point on that face, where your golf ball impacts, the center point of your impact location, that face can flex differently and in a way that we can control from a different point on the face.
Prior to our use of AI, you would have very kind of generic VFTs, and when you have a really thick or really thin section, you’re going to get some difference in ball speed and ball speed robustness. What we’re doing is allowing that material to change the way that the club deflects in a localized region. It’s allowing for, effectively, a resistance to gear effect…. A golf ball [struck] off-center, it’s going to have gear effect, and it’s going to come off with a certain type of spin, sidespin or drawn-bias sidespin.
A micro deflection is a way of that face deflecting differently in that location, in a way that we want to, so that it’s resisting gear effect and imparting less sidespin than it normally would. It’s almost like we’re flexing within an overall pattern of flexion. The overall head is rocking a certain direction, and the face in a localized sense is also deflecting in a different type of plane. And that’s giving a smaller gear effect condition on the golf ball — one that cannot overpower gear effect but resist it and control it in a subtle way. As we know, though, a subtle way can be important. It can be 100 or 200 rpm of sidespin, which would be significant and help keep a ball more online to target.
$200
View Product
GOLF: How do golfers know they’re getting the benefits of AI?
WILLIAMS: A lot of the stuff we’re talking about are the kinds of things you don’t experience just from the first pure strike of hitting the golf club. It’s stuff you see over time, either in the course of a fitting or the course of your ownership. That’s when you’re going to see, “I’ve hit enough shots now to know that something’s different with this club.” It’s consistent, it’s repeatable. Around the face, “My misses are better.” That’s where we’re using AI to differentiate ourselves.
GOLF: Are there misperceptions that consumers have around AI products? Where they think that it’s built a certain way or does a certain thing, and they just don’t really get it?
WILLIAMS: As we discussed earlier, my immediate reaction to that is that “it takes a designer out of play,” and that’s not the case. Designers remain critical. One of the things we hear is, “Does this mean all of my golf clubs are designed by AI, and we’ve lost the human interaction in the engineering?” Or are we killing jobs because of AI? That’s not the case at all for how we operate.
Our designers are still critical for a number of reasons. We’re still building in all the traditional chassis, and envelope, and spec packages. We still are aligned on our team to talk about goals and objectives for the product and how we achieve them in the physical sense. So, we’re still using designers to build our base models. We’re using engineers to set up the AI optimization runs so that they have the right kinds of inputs. And then we’re using AI as what we call the solver, or the optimizing tool.
We use it to take something that we think is the right starting point and say, “Now you help me find the details that I can’t find on my own.” That is where we are seeing this additive incremental benefit that we couldn’t get by just relying on our human capital and our engineering talent. But without that, we’re in no kind of place to operate in any kind of an autonomous sense with AI. It’s still very much a partnership between golf designers who have years of experience as players, and using our products, and knowing what matters to golfers, and then these new breakthrough types of tools to make them better.
So, yeah, the machines are not yet running golf club design at Callaway Golf. Designers and engineers still are the core of our golf research and development. But now we have more powerful tools — and a very firm conviction that they’re helping us to make better golf clubs.
The post How AI is helping Callaway make better golf clubs appeared first on Golf.
