How AI Could Be Used to Judge the Boss’s Mental Fitness

[00:00:00] This episode is brought to you by Kia. The Kia EV3 electric experience is here, and you could be one of the first in the UK to see the new compact electric SUV at an event near you, exploring its extensive mile range and spacious modern interior. Speak to our experts about the Kia EV3 and our award-winning electric range. Secure your exclusive place today. Search Kia electric experience.

[00:00:27] Kia. Movement that inspires.

[00:00:34] Welcome to Tech News Briefing. It's Thursday, August 22nd. I'm Zoe Thomas for The Wall Street Journal.

[00:00:41] Tesla's latest driving assistance technology, which the company calls full self-driving, hasn't been approved by Chinese regulators yet.

[00:00:50] Meanwhile, Chinese rivals already have similar technology on the roads. So what does this mean for Tesla in its biggest market outside the U.S.?

[00:00:59] And then, bosses live in fear of verbal missteps. New artificial intelligence tools capable of predicting cognitive decline could make that worse.

[00:01:09] We're going to tell you about the research and AI tools that could be used to combat those workplace concerns.

[00:01:16] But first, China accounted for 20% of Tesla's revenue in the first half of this year.

[00:01:26] But like other U.S. and European car makers, the company's sales in China are falling.

[00:01:32] Tesla is hoping its latest driving assistance technology can help spur demand as soon as it gets regulatory approval.

[00:01:39] Here to tell us more is our reporter, Rafael Huang.

[00:01:43] So, Rafael, what is Tesla's advanced driving assistance technology, which it calls full self-driving, or FSD, incapable of doing?

[00:01:52] So, the latest version of FSD, the so-called version 12, is based on a technology called end-to-end neural network, which we can roughly understand as AI-based.

[00:02:03] And this latest software provides advanced features, including cruising on city streets or reacting to traffic signs.

[00:02:11] Cars with version 12 can generally drive on their own without a human turning the wheel or pressing the accelerator,

[00:02:19] although drivers are told to stay alert at all times.

[00:02:23] Beijing is concerned about Tesla's latest driving assistance technology, according to people familiar with official thinking,

[00:02:29] including the fact that it needs to send data from China back to the U.S. to train the system.

[00:02:35] Can you tell us more about those concerns?

[00:02:38] Tesla needs data from China to train its technology so that it can better handle China's traffic.

[00:02:44] It has long wanted to bring its advanced driver assistance system to China,

[00:02:48] but only until recently, it hadn't been able to test the software with Chinese data

[00:02:54] because China has strict controls on the collection and transfer of auto data.

[00:02:59] The authorities are concerned that video and location data collected by the vehicles

[00:03:06] could give the U.S. information about military sites or other sensitive facilities.

[00:03:11] From the regulators' perspective, the accident reports involving Tesla's driver assistance software in the U.S.

[00:03:19] have also met Chinese regulators concerned about its safety.

[00:03:23] So now Tesla has to collect sufficient tax data from China to prove that it is safe enough to operate on Chinese roads.

[00:03:32] What has Tesla said about the approval process?

[00:03:34] Elon Musk said last month that he expected Chinese regulators to approve the features by the end of this year.

[00:03:41] And that's why Tesla is now testing the software on some cars in China to get approval.

[00:03:46] So what is Tesla doing to respond to those concerns?

[00:03:50] Tesla is now testing its technology to collect data in the hope to convince the regulators

[00:03:55] that its software is safe and capable of operating on Chinese roads.

[00:04:02] It is also working on how to filter out sensitive information so that the data they collect

[00:04:08] don't contain any cybersecurity or national security risk for China.

[00:04:13] It is also exploring the possibility to train its driver assistance algorithms in China with local data.

[00:04:21] Let's talk about the bigger picture.

[00:04:23] So as Tesla is trying to get its approval, some of its Chinese rivals already have advanced driver assistance technology on the roads.

[00:04:31] How does Tesla's FSD software compare to those?

[00:04:35] Tesla's latest FSD software is based on this AI technology.

[00:04:41] And a former Tesla executive said Tesla is up to two years ahead of Chinese companies in driver assistance technology.

[00:04:48] However, we cannot neglect the adoption.

[00:04:52] Analysts point to the affordability of driver assistance systems.

[00:04:57] Tesla's Chinese rivals generally charge less or even provide the software free to car buyers.

[00:05:03] Tesla wants to leverage on the FSD features to boost its slowing China sales.

[00:05:10] But analysts say that the price difference has made Tesla software less attractive to Chinese drivers.

[00:05:18] And Tesla is always slow in rolling out the service.

[00:05:22] That was our reporter, Rafael Huang.

[00:05:24] Coming up, the health of executives is important for the well-being of the businesses they lead.

[00:05:30] New AI tools could help raise and ward off scrutiny.

[00:05:35] We'll have more on that after the break.

[00:05:37] The Admiral spots the Greens.

[00:05:45] They're doing up their house.

[00:05:46] Need a hand with anything?

[00:05:48] No, no.

[00:05:49] Once the solar panels are up, our eco-transformation is complete.

[00:05:53] We've even bought a new electric car.

[00:05:55] Green.

[00:05:56] To match the house.

[00:05:57] Yeah, it looks fab.

[00:05:59] But why are we whispering?

[00:06:00] We're trying to reduce our noise pollution.

[00:06:03] Admiral is as committed as the Greens in covering your switch to electric.

[00:06:07] Get a quote now for your electric car and more.

[00:06:10] Admiral Insurance.

[00:06:11] Always looking out for you.

[00:06:17] Earlier this summer, a halting debate performance by President Biden shattered public confidence in his acuity.

[00:06:24] He's not the first leader to face such questions after delivering a shaky performance.

[00:06:29] And as AI gets better at analyzing rhetorical stumbles, it could be used to suggest business leaders are no longer sharp enough to lead.

[00:06:38] A study published in June in the journal Alzheimer's and dementia shows how AI can use a person's speech patterns to forecast Alzheimer's disease with roughly 80% accuracy, six years before the condition becomes diagnosable.

[00:06:53] Here to tell us more about this is our columnist Callum Borshers, who writes about careers and work lives.

[00:06:58] So Callum, this study forecasting Alzheimer's disease, how does the technology work to make this prediction?

[00:07:04] It's basically using recordings from patients who you know eventually went on to develop Alzheimer's and then looking backwards to see, okay, how did they sound years before they had a diagnosable level of cognitive decline that would rise to the level of dementia or Alzheimer's disease?

[00:07:21] And can the machine learn to find speech patterns that would indicate today if you sat down for an oral test that you could be on the path to dementia?

[00:07:30] So that's what the model is trying to do.

[00:07:32] But as you said, it's imperfect.

[00:07:33] So for patients, it can be a really useful tool.

[00:07:36] But there's also concern for false positives and whether it could unfairly cast doubt on somebody's mental acuity.

[00:07:42] So you mentioned an oral test.

[00:07:44] What kinds of things do people need to be saying in order to get a reading?

[00:07:47] Well, I talked to the lead researcher on this study, a distinguished professor of engineering at Boston University named Giannis Paschalaitis.

[00:07:53] And he says you'd sit with a medical professional.

[00:07:56] This would be a fairly long oral test.

[00:07:57] And they might ask you some questions and then come back to that same topic, say, a half hour later in the same interview.

[00:08:05] And it would sort of test your short-term memory loss.

[00:08:08] Maybe you had a hard time recalling what you had just talked about earlier on.

[00:08:11] And so in a business context, that might be something that people notice with coworkers.

[00:08:15] And that's just one example of how the patterns show up.

[00:08:18] I've been cautioned by researchers, by the way, that right now the model needs the structure of a written test because the model is comparing a new person's responses today to what showed up in hundreds of patients' responses to those same kinds of questions over time.

[00:08:34] The future outlook could be something more like casual conversation that you would record with a smartphone.

[00:08:39] Right. And as we've mentioned, you know, these tests are not 100 percent accurate.

[00:08:45] So, Callum, how could AI tools for detecting diminishing competence be countered?

[00:08:49] The good news for executives who might be worried about AI sounding the alarm bells about cognitive decline is that there are also AI tools that could help cover that up.

[00:08:58] And some of them are already at our fingertips today.

[00:09:01] You take something like ChatGPT or Gemini that can generate drafts of answers to employees' questions in a split second.

[00:09:10] And then there are even more sophisticated things like voice cloning tools.

[00:09:13] For example, if you were recording a PowerPoint presentation and you had a hard time delivering that message clearly,

[00:09:20] you can now go back, feed software a corrected script, what you wanted it to say, and the software has learned your voice and it can mimic your voice to make it sound as if you read it perfectly.

[00:09:34] There's even a feature on one of these software tools that will move your eyes.

[00:09:37] So it looks like you had good direct-to-camera eye contact if you were glued to your script.

[00:09:42] If someone receives a prediction from AI that they could have Alzheimer's in the future, what did the experts you spoke to say about how they should use that information?

[00:09:51] Well, they universally said you should not take it at face value.

[00:09:55] It is a red flag that you should go to see a neurologist, go see a specialist.

[00:09:59] That's true for a lot of at-home diagnostics that we already use today.

[00:10:05] The Boston University researchers who are working on this model with speech patterns, for example, tell me they don't think that the role of the physician ever will or should go away.

[00:10:15] But the idea in their minds is that we have some healthcare deserts in this country and certainly around the world where it can be hard to go get screened.

[00:10:23] Or if you do get screened, maybe it's too late because your disease has progressed to a point where it's just become obvious to loved ones and then they finally bring you to the doctor.

[00:10:31] So the idea is can we put a diagnostic tool, an at-home diagnostic tool, on an app in your phone and at least give you a signal, hey, you should really go get checked out.

[00:10:41] Callum, in the context of work, this seems really invasive.

[00:10:46] Are there examples of health information being used by company boards to judge the fitness of their leaders now?

[00:10:52] For people who want to be a chief executive, a lot of them are already used to forking over a lot of personal health data.

[00:10:59] It would not be uncommon for a wannabe CEO to sit through an extensive personality or psychiatric evaluation to try to assess whether they're fit or a cultural fit even for the job.

[00:11:12] And then on an ongoing basis, executives sometimes submit reports from their annual physicals.

[00:11:18] Could be blood work.

[00:11:18] Could be EKG results.

[00:11:20] All of this can be worked out in a CEO's contract.

[00:11:23] But of course, an executive's personal health is often closely tied to the well-being of the company.

[00:11:29] And so that's the rationale.

[00:11:30] And executives I talked to said, look, we understand that.

[00:11:32] And it's appropriate for boards of directors to want to know that we're still up to the job because it is physically demanding.

[00:11:41] At the same time, there has to be some line, they say, where you have a measure of personal privacy.

[00:11:46] And once you get into forecasting what's coming down the pike, that's a whole other level.

[00:11:52] I talked to one CEO who said, look, if it gets to that point, I'm just going to quit.

[00:11:56] I won't put up with that level of invasiveness.

[00:12:00] But he also said it would be wise, though, for executives today, if you're negotiating your next contract, try to get in writing some language in your contract about what can and cannot be used when it comes to AI health models.

[00:12:11] That was Callum Borscher's, our on-the-clock careers columnist.

[00:12:15] And that's it for Tech News Briefing.

[00:12:17] Today's show was produced by Julie Chang with supervising producer Catherine Millsop.

[00:12:22] I'm Zoe Thomas for The Wall Street Journal.

[00:12:24] We'll be back this afternoon with TNB Tech Minute.

[00:12:27] Thanks for listening.