Is McDonald’s AI Safe?

Forbes

On this episode of Forbes Talks, Breaking News Reporter Brittany Lewis talks with James White, the chief technology officer at CalypsoAI to discuss how McDonald's is integrating AI into its 43,000 locations.   They delve into the specifics of McDonald's AI plans, including internet-connected kitchen equipment, AI-enabled drive-throughs, and AI-powered tools for managers. They also cover the potential security and privacy concerns related to these new technologies, like data protection, voice privacy, and the use of computer vision.  Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1  0:00 Introduction 1:12 Why McDonalds Is Testing New AI In 43,000 Locations 8:31 Will AI Speed Up Production In Fast Food 13:02 Future Of AI In Food Production  Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:  https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript  Stay Connected Forbes newsletters: https://newsletters.editorial.forbes.com Forbes on Facebook: http://fb.com/forbes Forbes Video on Twitter: http://www.twitter.com/forbes Forbes Video on Instagram: http://instagram.com/forbes More From Forbes:  http://forbes.com  Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.

Transcript

00:00Hi, everybody. I'm Brittany Lewis, a breaking news reporter here at Forbes. Joining me now

00:07is James White, Chief Technology Officer at Calypso AI. James, thanks so much for joining

00:12me.

00:13Thanks so much for having me, Brittany. Love to be here.

00:16We have a really interesting conversation today that I would love to get your perspective

00:21on because McDonald's is reportedly integrating artificial intelligence into its 43,000 locations

00:27and the revamp will be complete with, quote, internet-connected kitchen equipment, artificial

00:32intelligence-enabled drive-thrus, and AI-powered tools for managers, according to reporting

00:37from The Wall Street Journal. As someone who's an expert in the AI space, what's your reaction

00:42to the news?

00:43My initial reaction is that it's great. It's great to see a company like McDonald's that

00:48is so large globally and expanding into their digital future with AI technology. It's a

00:54company that operates, obviously, in two spaces, digital and analog, and they have

00:59a huge presence in the real world with brick-and-mortar locations all over the U.S. and globally.

01:05So it's fantastic to see them embrace AI. And the use case they're targeting really

01:09seemed like an apt usage for AI.

01:12And the McDonald's CIO told The Wall Street Journal that this shift is due because the

01:18restaurants are pretty stressful. It's a fast-paced environment. Also, there's things like there's

01:22the running joke that the ice cream machine in McDonald's is always broken, and they want

01:27to alleviate that. They want to fix that problem. And I think that this brings up a larger conversation

01:32in AI about the pros and cons of this paradigm shift to the technology, pros obviously being

01:38it can help alleviate stress, like McDonald's is trying to do. They want things faster and

01:42more accurate. Cons can be some people think that, hey, my job might be replaced with a

01:47technology, as well as there are opening a host of cybersecurity concerns if you use AI.

01:54So as someone who's the cybersecurity expert when it comes to AI, when you hear the news

01:59about McDonald's from the security lens, what are you thinking?

02:04Yeah, I guess the very first thing is that thinking through the apt use cases is the

02:09first. So is AI the right solution for the problem? And if it is, then how do we provide

02:14that safety and security? So there are a couple of use cases that McDonald's are approaching.

02:20One is bringing in IoT sensors into their devices in each retail outlet. And that's

02:27really interesting because it can help identify things ahead of time, like if a flurry machine

02:31is about to go on the blink or something like that, or also things like identifying spikes

02:36that happen weekly. Maybe it's right after school gets out, there's a spike in the flurries or

02:41something like that. But then you have to understand what that data, if it's not protected,

02:46can help ne'er-do-wells use that for malicious purposes. So for example, if you knew when a

02:53store was getting its deliveries, that's a prime time where you could pick theft or something

02:59like that. Or if there is a device that is particularly well used and get to the supply

03:08chain of that device to somehow manipulate that device in some way, that's another way you

03:13can identify where chains are using machinery in what areas and how they can be impacted.

03:19But the use cases where they're using generative AI are very, very interesting. So one of the

03:24purported use cases is to automate the bureaucracy that is on managers' shoulders today. So moving

03:32around shift schedules, et cetera. And that obviously can be a nightmare day-to-day if people

03:37are ill or if there's some sort of dramatic shift change that has knock-on effects to the rest of

03:42the staff. So introducing generative AI there is a really, really good use case. However,

03:48if it's not protected, it could fall under attack from, if people identify that all it requires an

03:55email to send in a request for a leave of absence or a change in shift, then that could be manipulated.

04:02Or as we're well aware, there are things like injection attacks where you put in

04:07content into an email that is not visible to the regular user, but is embedded in the HTML,

04:14so the complex content sent in an email these days, and that can cause impact to the generative

04:19AI system. So it has to be monitored, it has to be protected. And if those things happen,

04:24it really will realize the benefits that are intended from these use cases.

04:28I'm sure anyone who's been into a McDonald's knows just how fast paced it is. It's busy,

04:34there's a lot going on. You want your food, you want it fast, you want to be in and out.

04:38And as a customer, from a customer's perspective, I would love to get your thoughts on some AI

04:43potential questions that they could have, because the McDonald's CIO said that they want to implement

04:50AI in drive-thrus. So I'm thinking when the person's talking, they're going to be using

04:55voice AI, some technology around that. What are some security concerns you have about that? And

05:01then how do you fix those? Yeah, so I guess there's both security and privacy concerns.

05:08So let's go with security first. When you're driving up to some sort of machine and you

05:13usually wait for a human being to interact with you now, and presumably it will be interpreted

05:18by Gen AI or some other AI system, it will listen to your order and it will then need to do a couple

05:24of things to build the confidence of the users. So one is to replay back and what you ordered

05:29before you say, go, we don't want 10,000 McNuggets ordered by accident. There are other attacks like

05:37above and below human audible attacks. So where you play an audio that is above or below what

05:43humans can hear, often machinery can still understand it. So to leave a little

05:49Bluetooth speaker underneath the machine and play at different orders, that could have an impact

05:55without anyone hearing that audibly from a human perspective. On the privacy side, we've got

06:01people making orders and you've probably had a lot of phone calls where the automated message

06:07says this material will be used for training future staff, et cetera. Well, now we're in a

06:12situation where your audio, your voice could be used to train future versions of models. So I'm

06:18sure companies like McDonald's, you have a great reputation, will have a really strong privacy

06:24policy and tell their customers what their data will be used for, if at all. And if it's not used

06:30for anything, then it is completely safe in their fears around their data being used or their voice

06:37being used. However, they need to understand how long data is stored for and if it's stored at all,

06:42where it's stored, what region, are they using models that are US-based or in other jurisdictions?

06:47These are all valid questions any consumer should have for any company offering AI

06:52interaction services with them at the point of sale. I also want to read some more reporting

06:58from the Wall Street Journal. This is what they wrote, quote, McDonald's is also exploring the

07:02use of computer vision, the form of AI behind facial recognition in store-mounted cameras to

07:08determine whether orders are accurate before they're handed to customers. And as we know, facial

07:13recognition is a somewhat controversial technology. So what are some of the security concerns there?

07:19And how do you think McDonald's and other companies that use that can address them?

07:25Yes, so I think, you know, computer vision is the wider field and they're using it to identify

07:32if a burger has been created properly before it goes out, you know, were the pickles removed,

07:37and does it have tomato sauce, et cetera. And so those use cases are innocuous. You know,

07:41I don't think anyone's going to be too worried if a burger is scanned. However, where those get a

07:46bit tricky is if people's faces, et cetera, are also captured by the vision technology,

07:53are those faces used for any purpose? Are they stored? Are they recorded? And so this is where

07:59it gets a little bit murky. You can resolve these quite easily in a fashion that means that

08:04the camera's only ever pointed at the food preparation, it's never upwards, et cetera.

08:09But even at that, you can still have, you know, and I know this sounds funny, but you could have

08:13reflective surfaces and stainless steel in kitchens that could capture people's faces,

08:18et cetera. But in the most part, companies are quite used to computer vision now and have understood

08:24the limitations and the optimal camera angles and things like that that can be used to reduce

08:29privacy concerns. And as I said, the McDonald's CIO said that the restaurants are a stressful

08:36place. There's a lot of people in and out, there's delivery people, there's a lot of customers

08:41wanting to get in and get out, a lot of workers behind in the kitchen and the counter. A lot's

08:46going on and they want to keep it moving. In a fast-paced environment, how do you make sure

08:51even more so that AI is secure? Because there is a ton going on.

08:56Yeah. So AI is quite, as it's well reported, a GPU-intensive process. So it requires a lot of

09:04compute power or GPU power. And so it's important that each restaurant has the allocated power

09:13necessary to power these systems, but that doesn't become a bigger problem for concern.

09:17If everyone becomes reliant on an AI for drive-through or their order fulfillment, et

09:23cetera, and then something goes wrong with that system, now you're faced with a problem that

09:28maybe the old processes are not in place anymore. And so how do you fall back? What's a graceful

09:32fallback for staff in a restaurant when things go wrong? Preparing for AI systems to go wrong

09:38is just as important for them to go right. And so understanding in this situation, we hope the AI

09:44system will provide X service. Well, then if it does not, what do we do instead? And so I think

09:49that critical staff training, how to coincide with AI instead of being replaced with AI,

09:56is a critical step for McDonald's and other companies planning on doing something similar.

10:01And I know that Calypso has rated different models of AI,

10:05but when it comes to cybersecurity surrounding it, what do businesses usually miss?

10:10Yeah, it's interesting. So I think the very first thing is they forget that AI is often just

10:16doing the normal task in an automated fashion. So whatever rules and restrictions apply to the

10:20use case already, they still apply. You don't get to move away from the existing laws and

10:26regulations in your sector. If you're in a highly regulated industry, the data that you're using

10:32could be very proprietary or could be specific to human beings, either employees or customers,

10:38and the same level of control and access that are required pre-AI or required post-AI.

10:44But we often see that the focus is on how do we secure AI as a net new, forgetting that it also

10:49holds for all of the existing use case and restrictions and requirements. And then secondly,

10:55we often see that unfortunately the whole horse is bolted. So the excitement of solving a problem

11:01with AI takes over and the AI system is now in production before security is thought about.

11:09And so it's a, how do you get the genie back in the bottle type situation?

11:13So luckily, custom defenses, if you're red teaming your models or using custom defenses,

11:19fighting fire with fire effectively, that can help even if the horse is bolted, you can put those

11:24dirty controls in place and then retrospectively secure that AI system. And you brought up some

11:31interesting points that I necessarily wouldn't think about as someone who doesn't think about

11:36AI day to day, like the reflective surfaces and things like that. And we know that McDonald's,

11:42according to reporting is rolling this out in over 40,000 locations. So when they're doing

11:47a mass implementation like this, what do businesses really need to be thinking about?

11:53That's a really interesting point. The US is a huge part of the world and it has various

12:01different light aspects for different time zones, et cetera. So in one store, the artificial

12:07lighting might be different than another store. So how the camera sees the image may differ.

12:13We saw in casinos previously where a system was rolled out very successfully and a camera

12:17failed, it would replace the camera with the newer model. And that newer camera didn't work as well

12:22with the existing AI system. So there are environmental concerns are the biggest concern

12:27when you're rolling out anything that's either audio or visual and to a specific location.

12:33Think about a drive-thru that's beside a train line or beside a freeway, or that has

12:43one of those little beeping crosswalk indicators right beside it. They all cause the AI to hear

12:50and interpret those things differently. So it's critical to understand on a location by location

12:54basis, how the AI will behave. And for me, that's the most critical factor for McDonald's

12:59factor when you're rolling this out. I find conversations surrounding AI really fascinating

13:06because it really feels like the new frontier. And when you're thinking about McDonald's unveiling

13:11this in tens of thousands of locations, what do you think is the future of AI in the restaurant

13:17industry? So honestly, I don't know if you've seen that TV show Bear, but that really brings

13:25home the stress and tension in the kitchen for me. So I believe that AI will reduce the stress

13:32and workload on staff and restaurants. And a lot of speculation on replacing jobs, I think it

13:40creates a much more harmonious environment for everyone to work in. A lot of the mundane tasks

13:44are taken care of automatically by AI systems, and often before the human being realizes it needs

13:50to happen. So if ingredients for a specific menu item are not in stock, and that's been identified

13:57by an AI vision system, and it's been ordered, you might come into work the next day and you find the

14:02delivery sitting there waiting for you. And that reduces your stress for the rest of the day. Or

14:07you've got a repeat customer, and we know that they have allergies. You don't have to ask that

14:10customer every time, you already have that list of allergies, which reduces the propensity for

14:15you to accidentally give them something they shouldn't have. And also, it's a nicer experience

14:20for the customer, right? They come in, and they feel like you remembered as a human being. Really,

14:24AI is remembering the background that's important to you. I think if you've ever worked in a kitchen,

14:29if you ever were a busser or a waiter or waitress, you know just how stressful those environments are.

14:35I know, I certainly do. And I find it, I think people will be excited by the possibility of

14:41alleviating some of that stress. James White, thank you so much for the conversation. You

14:46are welcome back anytime. Thanks so much, Brittany. Really enjoyed it. It's been my pleasure.

Category

Transcript

Recommended