Examples of Failure in Artificial Intelligence
Artificial intelligence is groundbreaking and, at times, still quite mind blowing. We’re constantly peppered with amazing stories of efficiency, automation, and intelligent prognostication. But AI isn’t perfect. And for every story of success, there’s another tale of a mess up or mistake – a situation where something didn’t go as planned.
While I’m a huge believer in AI and have seen the power of it in my own businesses, sometimes it’s nice to see the other side of the coin, have a couple of laughs, and remember that we’re all just pushing for bigger and better things. But along that path, there will be friction and interruptions. It’s how we respond to these anomalies and shortcomings that ultimately defines where we go from here.
6 Examples of AI Failures
We don’t want AI to be perfect. At least not yet. The fact that AI can still experience failures and lead to unintended consequences is somewhat refreshing (when you’re a couple of steps removed from the failure). But even for those close to the fire, these mistakes serve as lessons in growth and improvement. And the more failure there is today, the more growth and improvement will occur – ultimately leading the way to a more perfect and complete industry.
Okay, enough with the preamble. Let’s deliver the goods. Here are some of the top examples of AI failure over the past few years:
Microsoft AI Chatbot Learns Some Unbecoming Language
AI chatbots have sort of become the norm on social media and other websites. Facebook has a really good one built into Messenger and it’s leveraged as a powerful business tool for providing customer service and jumpstarting conversations with prospects. But AI chatbots aren’t perfect, as exemplified by Microsoft’s AI chatbot, which briefly went by the name of “Tay.”
Released in March 2016 and deployed for Twitter users, Tay was programmed to have casual, natural conversations in the language of typical millennials. But it only lasted 24 hours. What happened? Well, a group of trolls on the site targeted its vulnerabilities and manipulated Tay into making very sexist and racist statements.
Peter Lee, the VP for AI and research at Microsoft, had to issue a public apology for not foreseeing this possibility ahead of time.
Athlete or Felon?
Amazon has a project they call Rekognition. It’s an AI-based facial recognition software that’s marketed to police agencies for use in investigations. It’s essentially supposed to cross analyze images and direct law enforcement officers to possible suspects. The problem is that it’s not very accurate.
In a study by the Massachusetts chapter of the ACLU, dozens of Boston-area athletes’ pictures were run through the system. At least 27 of these athletes – or roughly one-in-six – were falsely matched with mugshots. This included three-time Super Bowl champion Duron Harmon of the New England Patriots.
Can you say, not a good look?
Users Find Flaws in Apple’s Face ID
Apple is always coming up with cutting edge technology. They’ve set the standards in the smartphone and mobile device industry for years. For the most part, they get things right. But sometimes they can be a bit too brash in their marketing. In other words, they like to flex their muscles. As you might expect, this invites haters, trolls, and skeptics to challenge their claims.
One recent example occurred with the release of the iPhone X. Leading up to the launch, Apple had invested a lot of time and marketing dollars into their front-facing facial recognition system that replaced the fingerprint reader as the primary method of accessing the phone. The claim was that the AI component was so smart readers could wear glasses, makeup, etc. without compromising functionality. And that’s essentially true. The problem is that Apple also clearly stated the Face ID technology can’t be spoofed by masks or other techniques.
One Vietnam-based security firm took this as a challenge. And with just $200, they made a mask out of stone powder, glued on some printed 2D “eyes,” and unlocked a phone. This is just a reminder that bold claims can sometimes come back to bite!
Robot Dog Meets Fatal Ending
Who doesn’t love the idea of a robot puppy? You get a cute little machine without the barking, walking, pooping, eating, or expensive vet bills. But if you’re looking for a life partner, you might not want this robodog.
In 2019, a Boston Robotics’ robodog named Spot met a dramatic and untimely onstage death while he was being demoed by the company CEO at a conference in Las Vegas. Tasked with walking, he slowly started to stumble and eventually collapsed to the floor as the audience uncomfortably gasped and chuckled.
Watson Is Not a Doctor
IBM’s Waston is a pretty amazing piece of technology. This smart supercomputer has many accomplishments under his belt, including defeating some of the world’s smartest people in a game of televised Jeopardy. But as much as Watson knows, he’s not to be trusted as a doctor – yet.
In 2018, IBM Watson attempted to launch a medical AI system to make suggestions for treating cancer patients. IBM’s objective was nothing less than to “eradicate cancer.” But it didn’t take long for hospitals and oncologists to see major flaws. At one point, Watson suggested putting a patient with excessive bleeding on a medication that would cause even more bleeding – possibly killing the patient in the process!
IBM has blamed its engineers, stating they programmed Watson with hypotheticals and fictional cases, rather than relying on actual patient data and historical medical charts. Either way, it’s not a good look for Watson. Perhaps he’ll stick to gameshows.
Voice-Spoofing AI Software Cons CEO
Deepfakes are becoming a serious (and alarming) problem. Hackers have found ways to fake voices, pictures, and even video. And in certain cases, the effects are disastrous.
In March 2019, the CEO of a UK-based company got a phone call from his boss over at the German parent company. He was instructed to transfer the equivalent of $243,000 to a Hungarian supplier. The request was marked as urgent and the CEO was told to carry it out right away. The only problem with the request was that it wasn’t his boss on the other end of the line. It was an AI-based software made to mimic the boss’s voice.
While we’re calling this an AI failure, the reality is that the AI software won. It was the humans who got played to the tune of a quarter of a million dollars!
AI: Challenges and Opportunities
As we can clearly see, AI isn’t without its issues. As it pertains to business, AI implementation still faces several challenges. They include:
- Limited processing power. While AI and ML have great potential, they utilize a ton of processing power. Most computing simply isn’t that advanced. As a result, it’s difficult to fully utilize these technologies outside of very specific environments.
- Limited knowledge. Only a handful of people truly understand AI well enough to explain it to the marketplace. This has kept adoption rates from being where they should be and is slowing down growth.
- Lack of trust. There will always be a degree of mistrust between people and computers. And while relations have improved over the past few years, failures like the ones outlined in this article don’t help much.
- Data security. In order for AI applications to work, these systems need access to millions of data points. This creates possible weak spots and vulnerabilities for hackers to target and compromise.
Despite these challenges, the beauty of AI lies in its intelligence and brilliance. Not only are many of the world’s greatest minds dedicating their lives to improving and refining technology, but when you couple it with machine learning (ML), you get a self-fueling cycle of growth where every problem and mistakes ultimately paves the way for greater efficiency, accuracy, and opportunity. Speaking of opportunities, here are some of the top ones I see moving forward:
- Makes big data easy. One of the issues businesses have with big data is finding ways to make sense of it. With so much information to sort through, discovering how to interpret and apply findings is difficult. AI can streamline this and lead to better, more efficient processes.
- Superior analysis. Certain AI systems can be used to monitor, detect, and analyze changes in various settings. Businesses can use it to keep an eye on competitors and understand things like price changes, PR activities, social engagement, etc.
- Smarter searches. The internet makes the world go around. In particular, search engines shape trends and control the flow of information. Search engine companies like Google are using AI and ML to transform the industry and deliver more accurate and timely results to customer searches.
Is AI perfect? As the tales in this article show, that would be a resounding no! But is it powerful and ever-improving? Yes on both. So as we look ahead to the next one, three, and five years, let’s remember that AI is a work in progress. And to get where we want to be, we have to deal with a bit of friction along the way. It comes with the territory.
The post Examples of Failure in Artificial Intelligence appeared first on ReadWrite.