Millions of jobs affected! \u00a0<\/p>\n
A white-collar bloodbath!\u00a0<\/p>\n
Massive economic disruption!\u00a0<\/p>\n
A technological spreading of Marxist propaganda!\u00a0<\/p>\n
Politicians, pundits, and C-suiters everywhere are shouting out about the epic mud stomping that artificial intelligence (AI) is going to do on our stable and beloved job market in the next few years. Or the next ten minutes, depending on whom you\u2019re listening to. They\u2019re loud. They\u2019re apocalyptic in some cases, dystopian in others. And, yeah, they\u2019re freaking people out.\u00a0\u00a0<\/p>\n
The bad news is coming from all sides, though the numbers vary widely. On the high side, Goldman Sachs predicts that AI could impact up to 300 million full-time jobs worldwide, particularly in the U.S. and Europe.\u00a0<\/p>\n
On the low end, there\u2019s the World Economic Forum, which predicts that by 2027, 83 million jobs will be lost to AI, while AI will create 69 million jobs. That would result in a net 14 million job loss.\u00a0\u00a0<\/p>\n
Then there\u2019s the guys pitching the real grabbers. Like Dario Amodei, CEO of Anthropic, who says he thinks AI could wipe out half of all entry-level white-collar roles within five years, a situation he\u2019s calling a potential \u201cwhite-collar bloodbath.\u201d\u00a0\u00a0<\/p>\n
Or Adam Dorr, a real doom-and-gloomer, who predicts AI will wipe out all (yes, all) human jobs by 2045 except for politicians, sex workers, and ethicists. I\u2019m going to take a wild guess and say that Dorr thinks of himself as an ethicist. And he probably also hasn\u2019t been to CES in the last few years, or he\u2019d have seen what the Japanese have been doing with sex robots. (Hint: disturbing.)\u00a0\u00a0<\/p>\n
By comparison, our intrepid political leaders\u2019 views on AI seem tame. Even President Trump is sticking to a relatively middle-of-the-road prediction (though he is convinced that AI might be spreading Marxist propaganda). He thinks AI will take between 10 and 15 million jobs in the next five years, but doesn\u2019t believe the government\u2019s role is to restrict AI; instead, it should give tech companies more freedom to use it. (Some of his other beliefs on AI\u2019s long-term effects are more disturbing than the Japanese sex robots, but let\u2019s stick to the topic at hand.)\u00a0<\/p>\n
Let\u2019s all just take a breath, shall we?<\/strong>\u00a0<\/p>\n What should the rest of us take from all these, let\u2019s be fair, often vague predictions? Nobody knows, that\u2019s what. If AI is going to supplant 14 million jobs, as the WEF predicts, that\u2019s certainly not great. However, by comparison, COVID-19 saw the world lose 255 million full-time jobs, or 8.8% of global working hours, according to the Sustainable Development Goal indicators website. Way worse, and we still don\u2019t see mountains of white-collar workers living in refrigerator boxes, do we?\u00a0<\/p>\n The more critical indicators are what AI has been doing to the job market in the last year and what it\u2019s doing now. Are hundreds of corporate boardrooms stuffed with slavering CEOs looking to dump flesh-and-blood employees in favor of AI?\u00a0<\/p>\n Rohit Patel, Director of Meta\u2019s Superintelligence Labs, doesn\u2019t think so. Meta Superintelligence Labs is where some of the most cutting-edge AI cookery is happening; he\u2019s in a unique position to comment.\u00a0<\/p>\n \u00a0He believes the negative impact of AI on jobs is being overhyped. \u201cIt\u2019ll have an impact for sure,\u201d he says, \u201cbut it\u2019s not going to be as damaging as is being predicted, and it\u2019s not going to happen immediately.\u201d\u00a0<\/p>\n Reason being, and Patel sums it up nicely, a job is more than just a task, and individual tasks are all AI is capable of right now. \u201cBecause AI is relegated to tasks means it can help make the people who do those tasks more efficient, but that\u2019s all you can reasonably expect right now. That\u2019s why I don\u2019t see any sort of job being replaced wholesale shortly.\u201d\u00a0<\/p>\n Patel believes that CEOs who focus solely on job consolidation are making a mistake. \u201cA task isn\u2019t a job because nothing is guaranteed,\u201d he says. \u201cA high performer today without AI isn\u2019t necessarily going to be a high performer tomorrow with AI. To me, it\u2019s less about cutting say 40% of your workforce right now, and more about taking the time to make a department more efficient with AI and then deciding whether that department needs as many people.\u201d\u00a0<\/p>\n AI in the real world<\/strong>\u00a0<\/p>\n I spoke with Jim Yu, CEO of BrightEdge, a well-known San Mateo-based provider of SEO solutions. BrightEdge has been experimenting with AI for several years, but doesn\u2019t sell an AI solution, so they\u2019re a good example of what a tech-savvy business can do with this technology today.\u00a0<\/p>\n Because SEO on a global scale requires massive amounts of data analysis that customers expect at an ever-faster rate, Yu says BrightEdge has been testing new technologies to help meet this challenge for some time.\u00a0<\/p>\n \u201cWe started down this path about 10 years ago,\u201d he explains. \u201cThat\u2019s when our customers went from wanting the data so they could create their own insights to demanding that the app do it for them. So we started testing different technologies to help evolve the solution that way. We tested a lot of different approaches, and we had a lot of failures. But that\u2019s the kind of experimentation and testing it takes to make the most of something new.\u201d\u00a0<\/p>\n To BrightEdge, generative AI has been a game changer, not just across application development, but also in the front office, particularly sales. Even so, when asked if he was considering laying off employees and replacing them with AI, Yu had a pretty strong response that jibes with Patel\u2019s.\u00a0<\/p>\n \u201cNo. And for those that are thinking that way, they\u2019re going to be disappointed. AI can only do so much.\u201d He also points out a key problem that most pundits aren\u2019t talking about: You need to hold AI\u2019s hand\u2014a lot.\u00a0<\/p>\n \u00a0\u201cWhat you\u2019re looking for with AI is to make your employees more efficient and accurate, right? But that\u2019s not plug and play at all,\u201d he warns. \u201cFor us, it\u2019s requiring endless tuning and testing across all kinds of use cases. It\u2019s not some kind of magical solution. Getting to an 80% correct answer on an individual basis and for pretty basic tasks isn\u2019t super hard, yeah. But what does that really get you? And getting beyond that 80% isn\u2019t at all easy.\u201d\u00a0\u00a0\u00a0<\/p>\n So is anyone really on the chopping block?<\/strong>\u00a0<\/p>\n Both Patel and Yu acknowledge that some jobs are at risk right now or in the near future, but neither one describes those consequences as a \u201cwhite-collar bloodbath.\u201d\u00a0\u00a0<\/p>\n \u201cCan you use AI to replace someone right now?\u201d muses Yu. \u201cSure, if you\u2019re using people for basic research, for example. But that\u2019s stuff like college interns. If you\u2019re using three interns right now, you may only need one if your AI testing is successful enough.\u201d However, he stresses that you\u2019ll need to test AI rigorously before making that sort of decision, and that you\u2019ll still need one intern to verify the AI\u2019s results.\u00a0\u00a0<\/p>\n Patel is a little more expensive. \u201cBeyond research, AI will definitely impact creatives fairly soon,\u201d he predicts. \u201cParticularly those involved in mass producing mediocre stuff. Like say you\u2019re someone who creates the pictures that hang in hotel guestrooms. AI can certainly churn those out faster than people and the quality will be the same. That kind of job is definitely in jeopardy.\u201d\u00a0<\/p>\n But in either of these examples, even the creative one, AI still can\u2019t just be turned loose to work on its own. Because AI is not reliable in any task. It\u2019s often wrong without knowing it, yet still offers up those answers with authority. It also fabricates information, which Patel and Yu refer to as \u201cAI hallucinations.\u201d\u00a0<\/p>\n \u201cAI essentially works on an auto-complete model,\u201d explains Patel. \u201cIt\u2019s got a fixed data set to work with, and it\u2019s just trying to answer your question with the information it has. It doesn\u2019t have a concept of correct and wrong. That\u2019s why sometimes you\u2019re reading an answer and you\u2019re asking, \u2018Hey why did it just make this whole thing up\u2019 Actually, I\u2019m surprised it doesn\u2019t do that more often.\u201d\u00a0\u00a0\u00a0\u00a0<\/p>\n So an AI might turn up a wrong result that an intern wouldn\u2019t have and deliver it with enough confidence that a reader might simply accept it if they didn\u2019t take the time to check it out first. Or a creative AI might decide to simply reproduce an existing painting if it couldn\u2019t come up with a variation of its own.\u00a0\u00a0<\/p>\n