McDonald's ended its partnership with IBM after their AI tech botched drive-thru orders
Despite the AI failure, IBM's stock rose this month as analysts praised its AI advancements
The incident highlights current AI limitations in complex tasks like order-taking but suggests future improvements are still likely
Is anyone else still thinking about IBM’s artificial intelligence (AI) drive-thru fail, or is it just us?
In case you missed it, McDonald’s decided last week to end its global partnership with IBM after the latter’s AI tech messed up drive-thru orders and created a PR nightmare for both companies. The fast-food giant had been working with IBM on AI technology at more than 100 drive-thrus since 2021.
We thought this might be a black eye for IBM, but to our surprise the tech giant seems to have come out of the AI failure unscathed. Actually, IBM’s stock soared this month as analysts continued to tout its spending and progress on AI development.
IBM's Automated Order Taker (AOT) technology, which includes its WatsonX generative AI, was essentially a voice-activated AI bot in Mickey D's drive-thrus.
We reached out to IBM for comment about the failure, but the company declined to provide specifics and instead sent us the same statement it gave to other media outlets. Sources familiar with the technology told CNBC that the AI had difficulty interpreting different accents and dialects, among other challenges affecting order accuracy.
McDonald's confirmed that it would shut off the AOT no later than July 26.
Par for the course
Failure might just be par for the course in developing new AI use cases because as Futuriom analyst Scott Raynovich told Fierce Network, in general, AI has still proven to be “relatively immature technology.”
“So far, AI use cases have proven to have many safety and accuracy issues, so it does not surprise me that this experiment ended at all,” he said. “When is the last time you got Siri to work well, or a phone menu was better than a human? Phone menus and ordering systems may indeed be a litmus test for when AI starts to work in common applications.”
McDonald’s is not alone in testing the possibilities of AI. For example, Chipotle has invested in AI and robotics through collaborations with various companies. It is testing an automated digital makeline with Hyphen and a collaborative robot named Autocado developed with Vebu for guacamole preparation (yes, you heard that right. AI is now making guac.)
In the U.S., Wendy’s partnered with Google Cloud to develop the “Wendy’s FreshAI” chatbot. White Castle teamed up with SoundHound AI to bring voice-powered AI technology to more than 100 restaurants by the end of 2024.
Meanwhile, Panera, Arby’s and Popeyes have implemented OpenCity’s “Tori” voice assistant in their order lanes.
For its part, McDonald's in December launched a separate, multi-year partnership with Google Cloud, moving restaurant computations from servers into the cloud and applying generative AI “across a number of key business priorities” in restaurants globally.
AI culture of forgiveness
It might be a false assumption that failures like IBM's have the potential to destroy confidence in AI, or its vendors. Consumers have been generally forgiving of AI mistakes (i.e. ChatGPT made some pretty glaring errors in its early days, but still has grown to over 200 million monthly active users worldwide.)
GlobalData analyst Charlotte Dunlap said because IBM, like Google and OpenAI, have been at the forefront of the GenAI market segment, any glitches that “naturally come with emerging technologies are going to be high-profile as well, there’s no getting around that.”
“GenAI technology is in its early days, but it will only continue to improve, and we’ll increasingly see successful use cases unfold,” she said.
Not dire straits
Maybe the reason no one is really up in arms about this whole McDonald’s debacle is that a botched hamburger order isn't all that dire? After all, drive-thru order mistakes are fairly common at the hands of humans.
Avasant analyst Dave Wagner noted that taking a complicated order in real-time could be one of the hardest things you can ask an AI to do, because parsing language the way humans order is difficult.
“It is one thing if I order a Big Mac meal with no substitutions. It is an entirely other thing if I’m ordering for a family of four and someone wants no pickles and another wants extra cheese,” he told Fierce.
In slightly more “controllable environments,” Wagner said Gen AI is having a more positive impact: career discovery, assisting in content creation and triage within a customer service environment. Additionally, most successful generative AI projects are either led by AI with human oversight or led by humans with AI assistance.
However, a lot of customer-facing AI “will quickly be table stakes,” Wagner added.
“Some version of this is inevitable. It will happen. But again, how much will it matter? Once perfected by IBM, Google, ChatGPT, Microsoft or some other player, it will be replicated across the entire industry."
Until then, it might be best to let the humans do the cooking.