AI CAN'T IMPROVE PRODUCTIVITY
Myth or fact?
6/6/20266 min read
Some research released in recent days reports that 90% of (American) companies claim not to have achieved increased productivity with AI, which reinforces the claims that AI is just a bubble or merely a marketing product.
More recently, Uber's COO stated that investments in AI can no longer be justified due to a lack of concrete results in terms of service functionality. At the same time, Tesla unveiled its new engine, demonstrating that it has reached a level of powertrain production costs for its vehicles that was considered impossible until recently, thanks to the massive use of AI.
The CEO of DeepMind, Google's quantum computer, stated that we are at the foot of the singularity and should reach the top by 2029 or 2030.
So, who is right?
I have no doubt that AI is a great marketing tool and, above all, a source of financial speculation. Look at the value of AI companies on the stock market and do the math. What this proves to me is that we do have a bubble, and at some point, this will have to be adjusted. Nobody lives eternally on an overestimated future.
But saying that AI hasn't increased productivity or added features makes me believe that, if it's true, these companies don't know why they're employing AI or are reversing the way they use the tool.
This reminds me of when the computer began to dominate workstations. In the beginning, it was used simply as a luxury typewriter with a built-in fax machine. Secretaries used the computer to write texts and memos dictated by their bosses, send emails, schedule meetings, make purchases and personal orders, etc. In some companies, in addition to this practice, it was common to print emails for discussion of the topics in meetings. Zero increase in productivity, because these machines were expensive and required much more maintenance than a mechanical machine.
In practical terms, these companies were only improving the equipment but not the way the work was done. I saw this firsthand at an agency: a creative department with 13 people and only one computer, in the hands of the secretary. Although it wasn't expensive, the computer was a foreign object not integrated into most of the work done; so it didn't make much difference. At that same agency, I was there when they bought a beautiful MacIntosh with a scanner and wax sublimation printer, costing $40,000, which was then kept in a display case. Despite only one of the art directors knowing how to use it and the wait of up to 4 hours to scan an image or print a layout, the result was not only more realistic: it came out ready to send to the magazine, meaning it didn't need to go through a finishing process, printing the text on photo paper, or scanning the slides at a print shop.
Ah, but then it was the $40,000 MacIntosh that changed everything? No. A little later, I had a CompaQ that cost R$3,000, a regular HP scanner, a DeskJet printer, and besides already producing layouts that became final artwork, I could also send them via email to the print shop, eliminating another huge step, which included saving the file to a floppy disk, waiting for the courier to arrive to take it to the print shop, hoping the file wouldn't get corrupted, waiting for the file to be returned, and then sending it to the printing company.
I'm giving this example to show that old processes rarely benefit from new technologies, and the same is true for AI. So, the issue isn't adding technology, but understanding what role it will play. The point isn't to include new tools, much less adapt processes to accommodate these tools, but to rethink what is done and how it is done. This reminds me of Steve Jobs's quote about asking consumers what they wanted; according to him, at the end of the 19th century, if you asked most people what they thought of better transportation, they would choose faster horses, not cars.
I'll give another example. When the cell phone arrived, it was just a portable phone (though not entirely), connecting people, not places. A good, but modest, advance. Then, it started receiving and sending texts, but it was still a pager combined with a payphone. After that, it started receiving emails and began to replace the computer in many cases. Finally, the iPhone arrived, and the rest is history. It replaced the computer, the digital camera, the GPS, the secretary, the paper agenda, the credit card, the camcorder, the voice recorder, the business card, the flashlight, the tape measure, the level, the glasses, the music player, the TV, the internet router, the photo album, the radio, the clock, the alarm clock, even the telephone—just to mention the top 20 uses.
The same thing is happening with AI, and the same kind of people who said that the cell phone wasn't improving anything (or the computer, before that), are now talking about AI. My apologies to Uber's COO, but if they aren't AI to predict where there will be more demand for cars in the future or in the next 10 minutes, if they're not using it to understand the type of car and driver each neighborhood prefers, if they're not using it to predict where robberies or harassment cases occur most frequently, to better understand the habits and interests of service users, to build predictive models of driver behavior, or to predict vehicle maintenance needs, then they really don't know what AI is for.
That's what Tesla is doing. By combining proprietary hardware and software, and plugging everything into AI, Tesla has begun to extract information from every part of the process, from production to sales, and from sales to use. It has images and routes of its cars, maintenance histories, breakdowns and usage habits, temperatures encountered, problems found, average speeds. It knows when and how much people refuel, where they do it, and how they do it. Furthermore, it knows when and how competitors do it, since they have made their charging network open to competition. Tesla knows exactly what's happening with its entire fleet worldwide, in real time, and can predict what to do in a day, a month, a year. With this information, it works to improve what will meet customer needs, even before the customers know what they will need.
Uber could have most of this data: driver and passenger profiles, trip profiles, maintenance habits, vehicle demand forecasts for events, days and times, when it rains, when it's sunny. It could help its drivers drive better, take better care of their cars, avoid breakdowns, and provide information about complicated routes and dangerous roads. It could create different types of customers, different services, different fares… Integrated with Waze or Google Maps, with AI and incentives for customers and drivers, Uber should know more about traffic and vehicle behavior than the CET (São Paulo's traffic authority). It could sell this data to city halls, car manufacturers, pension plan companies, insurance companies… In short, it could be the expert in urban mobility. It could even create a business model focused on buses and offer a specific control version for taxis. I'll go even further. The COO's complaint is about the cost of tokens generated by using AI. This shows, in my opinion, that they are applying AI in the wrong way, the opposite of what they should be doing. They are not an AI company, although they could be; they are a consumer of this service. If they are spending too much on tokens, it's because they are offering this service as an intermediary in the hope of gaining insights, or they are letting their IT and operations team use AI to try to figure out what to do. Token consumption needs to be directly linked to remuneration or results; that is, either you charge for providing the AI (which, on top of that, is from another company) or the investment must be backed by lower costs and/or better results.
Exactly what Tesla is doing. Each piece of data collected becomes an improvement in a process, a cost reduction, or a technological advancement that will translate into results later on. And look, Tesla has a sister company that's in AI, meaning it could use these tokens to help with the group's revenue, but that's not what it does, at least as far as we know. It uses them for its own benefit, although helping with the group's revenue is an adjacent benefit.
In short, if your company only uses AI tokens so that people have more free time or can do the same work with less effort, it's not using AI. If AI serves to automate processes that shouldn't exist, it's not using AI. If it uses AI only to avoid hiring certain types of services or to pay less, you're just using AI packaging.
AI completely changes how you do your work, how the company operates, and how it improves the development of products and services with less time and fewer errors. It needs to be part of the company's core to produce real results. And your company doesn't need to be a Tesla to measure the cost of every inch traveled. Using intelligence to reduce deadlines, costs, and resources is human intelligence and a necessity. It only becomes artificial because we use digital resources to achieve that. If you expect AI to reinvent your business, forget it. You need to reinvent yourself and use AI to do the heavy lifting.
If you don't know how, ask me.