Five companies stumble with AI deployments, revealing gap between promise and reality
Article excerpt
Air Canada's chatbot dispensed false refund information to passengers. Zillow's algorithm-driven home valuations went so badly wrong the company shut down its entire buying program. Samsung leaked confidential data through its AI tools. IBM and CNET each had their own public stumbles. These aren't isolated glitches or minor engineering hiccups, they're systemic failures that rippled through real businesses and real customers, revealing how quickly AI systems can amplify mistakes at massive scale. The pattern across these cases points to a gap that hasn't closed despite years of hype: the difference between what AI can technically do and what it should do in production environments where real money and real trust are on the line. As artificial intelligence advances, the real question isn't whether machines can perform specific tasks, they often can, but whether it makes financial sense to deploy them, considering costs, disruption, and what humans uniquely contribute. These five failures suggest the answer, at least so far, is more complicated than the vendor pitches make it sound.
Elections are akin to horse races. Like a bettor with a large bankroll, a political party has a vested interest in backing the right horse. And while political consultants are thick on the ground in Washington, and while the sophistication of their tools, polls, demographic analyses, social media operations, trumps that of touts at the track, one has to wonder what happened this year in Maine, where the Democrats chose to back an odd horse.
My firm conducts background investigations, so I know first-hand that everybody wants to back a winner. While there are no guarantees in horse racing, a thoroughbred may pull up lame in the far turn, handicappers try to maximize their odds by maximizing their knowledge of each horse. Their sources of data and methods of analysis may vary, but serious parimutuel bettors always conduct due diligence. Why don’t political parties?
The candidate in Maine is an untraditional choice for a national party, so untraditional that one wonders what, if any, vetting was done before the party lent him support. Of the many tourists who visit Croatia each year, one suspects that vanishingly few of them choose to commemorate the trip by getting a chest tattoo of the Totenkopf, or death’s head, the skull motif of Nazi SS units, the shock troops of the Holocaust. To be clear, national political parties in America have traditionally not chosen candidates with Nazi chest tattoos. Also, while some relationships are fraught with difficulty, comparatively few men running for the Senate have been accused of referring to women with a derogatory term related to female genitalia.
Regardless of the level of due diligence conducted in Maine, what type of background research should a political organization undertake before giving a candidate its imprimatur?
One might argue a political party could have missed the tattoo as it is not common practice to subject a candidate to a physical examination. Looking into past relationships, however, is standard operating procedure and, as any investigative professional will tell you, failure to contact current or former wives or girlfriends is negligence. There is also no excuse for not looking at the candidate’s social media profile. Clearly, the Democrats missed what one news outlet characterized as “a slew of horrifying posts made on Reddit [in which he disparaged] women, rape victims, minorities, veterans, cops and working-class voters.” Moreover, might it not have seemed off-putting to voters that the candidate sent sexual text messages to numerous women while he was married, as reported nationally? Finally, did the party know about or just expect the electorate to ignore the decade he maintained an account on “a seedy hookup app” called Kik, even while he was married? Keep in mind, this spurred a senator, in the candidate’s own party, to challenge the candidate to prove he did not send sexual images to minors via this application.
Proper due diligence should include a comprehensive review of a candidate’s personal and professional background, including research into criminal records, litigation and regulatory actions, sanctions exposure, and media profile. The purpose is to find potentially adverse information, such as material misstatements.
A Google search of a candidate’s name is not enough. The subject is a prospective holder of a national office, not a first date. It is critical to go beyond traditional social media platforms, beyond the candidate’s Facebook or LinkedIn profile, to identify and analyze the content of accounts on alternative platforms and messaging sites. Then researchers must go further, to search the deep web, the unindexed part of the internet that is inaccessible to conventional search engines, as well as the dark web, the portion of the internet that hosts online criminal communities, and other online locations, for reputationally significant content written by or about the candidate.
Perhaps most importantly, a researcher must interview those who know the candidate, those who have lived with or worked with the individual. These are the people who will have knowledge of controversial statements or behavior by the subject. These are the people who will have witnessed any aberrant behavior and who can provide context for known controversies.
GRAHAM PLATNER SAID SOLDIER SHOT FOUR TIMES ‘DIDN’T DESERVE TO LIVE’, DEMOCRATS WILL VOTE FOR HIM ANYWAY
These methods are not radical. They are methods my firm employs every day on behalf of organizations that care about the quality of the people they recruit, from a professional sports league vetting a potential new team owner to a Fortune 500 company assessing the bona fides of a prospective director. These methods are so well known and often practiced that one wonders what went wrong in Maine.
By not thoroughly vetting the candidate, internally or with the help of a professional investigative firm, one national political party finds itself having to explain how it missed serious “red flags.” Worse, the party is in the position of hoping no other damning information comes to light in the next five months. There is no indication that the party knows what, if anything, more is out there and thus is unable to guard against its revelation. Instead of learning things that could be reasonably ascertained about the candidate, which would have given the party time to either control the damage or choose another contender, the party went all in on a strange horse. With the horses not yet at the post for the general election, one wonders if there will be a scratch in the race before November.
Thomas Feeney is managing director with global investigations firm Nardello & Co.