Lecture by Dr. Jorge Grünberg, from the " rector " at the " Universidad ORT Uruguay," aimed at ORT Argentina students who are graduating from high school.
https://youtu.be/7yVxpRm_Ovc
I would like to thank President Guillermo Feldberg and all our friends at ORT Argentina for inviting me to meet with you today. It is a pleasure for me to be at this sister institution, which serves as a model for ORT organizations around the world.
I imagine that many of you are wondering, in the midst of this artificial intelligence revolution, what we humans do better than artificial intelligence. I imagine you’re also wondering which major to choose that will still be relevant in 2030 when you graduate. I’d like to invite you to think about what you should look for in a university to ensure that it’s “AI-ready”—that is, prepared for artificial intelligence.
First, we are experiencing a massive shift. This isn't the cloud, it isn't Web 2.0, and it isn't cryptocurrency. This is a much bigger shift.
Second, artificial intelligence will have an impact on jobs, but not on all of them equally. Of particular relevance to you, the jobs most at risk are junior and entry-level positions. Statistics show that there are fewer entry-level jobs, and this is a global trend.
Third, knowing how to use artificial intelligence will be essential in any profession. Recent studies show that employers are seeking candidates for a wide variety of jobs—from truck drivers to psychotherapists to financial advisors—across the socioeconomic spectrum, and job postings themselves require knowledge of artificial intelligence. In other words, it will be important to be literate in artificial intelligence in almost any profession you choose.
Keep in mind that you won't be competing with artificial intelligence—you'll be competing with those who use it better than you do. That's why you need to learn how to use it.
There have been significant technological changes in the past that sparked similar fears that machines would replace humans.
Historically, this did not happen, at least in the medium term. When ATMs were introduced in banks, it actually led to an increase in the number of bank employees. When AutoCAD and other tools appeared, the workload for designers actually increased. More recently, despite the introduction of effective image recognition systems, the number of radiologist positions has increased.
The question is whether artificial intelligence will be a historical exception.
There are many aspects that are unprecedented. The rate of adoption is much faster than before. It took three or four years for cell phones to be used by a large percentage of the population. It took ChatGPT only a few weeks to be used by tens or hundreds of millions of people.
What sets artificial intelligence apart from previous major technological changes? The difference is that, throughout human history, the only truly scarce resource has always been intelligence. Capital is reproducible; if I can’t get it in one place, I can get it somewhere else. Natural resources are finite, but replaceable. Intelligence, however, has never been scalable. It has never been possible to encapsulate intelligence and reproduce it at scale. Now, for the first time, there are automated systems that successfully replace the intelligent activities of people.
For now, there is no evidence of a “job apocalypse”—as the press has called it—that is, a large-scale replacement of human jobs. But there is, as I mentioned at the beginning, a reduction in job opportunities at the entry level. Younger college graduates are the ones facing the greatest challenges when trying to enter the job market. But this is not the case equally across all professions. One of the most affected professions, for example, is computer science. Entry-level programming jobs, which used to be a guarantee of job security, have declined in almost every country. In other professions where human intervention is intrinsically valuable—such as medicine, nursing, or teaching—this is not the case. In short, there is a decline in entry-level jobs, but it is not uniform.
To fully understand the impact of artificial intelligence, it is important to know that there are two possible strategies.
One involves companies that implement artificial intelligence to replace people; the other involves companies that implement artificial intelligence to enhance people’s capabilities. The latter is called an augmentative strategy. In general, when augmentative strategies are used, jobs do not decrease; rather, they become more specialized. Therefore, a desirable public policy might be to encourage a smart business community that seeks to enhance capabilities rather than replace people with robots.
It is also important to distinguish between companies that produce technology and those that use it. Today, the impact is greater on companies that produce technology—not necessarily because they are cutting jobs, but because traditional programming is being automated more and more effectively.
On the other hand, there are other activities—such as trying to convince someone or determining whether someone is being honest with me—that are difficult to automate. For those interested in computer science as a career, keep in mind that the future isn’t all bleak; there are jobs for systems engineers, and there will be many more in the future. And there will be new tech professions such as agent designers, workflow orchestrators, agent-based system managers, and many others that we can’t even imagine today.
Generative artificial intelligence is not all-powerful. What are the main limitations of generative artificial intelligence? It is important to understand the limitations of generative artificial intelligence in order to understand our comparative advantages as humans.
One of the limitations of generative artificial intelligence is the cost of cumulative continuous learning. Once a model is trained, what it knows is impressive, but it doesn’t learn anything new afterward. Reinforcement learning and fine-tuning exist, but they are costly processes that, in practice, are not cost-effective for the continuous learning required by certain tasks in changing contexts. Therefore, our main competitive advantage is that we learn much better and faster than generative AI models.
Consider that a teenager like you can learn to drive a car. After twenty or thirty hours of driving lessons, you can already pass the test and drive—and, what’s more, improve your driving with experience. In contrast, Tesla or Waymo, for example, need millions of hours and miles for their cars to learn to drive autonomously.
Another limitation of generative artificial intelligence is that it has limited abductive reasoning ability. Deduction and induction are two skills that generative artificial intelligence performs very well. Abduction is the ability to understand, based on very little observed behavior, the relationships within a group or a person’s intentions, for example. In other words, if I see someone acting in a certain way and want to know their motives and intentions, this is very difficult for artificial intelligence, whereas it is much easier for humans. Humans can reach reliable conclusions based on small samples.
For those of you who are about to choose a college major, ask yourselves: What should you look for in a college to know if it’s AI-ready?
Keep in mind that the goal of a university is very different from that of a company. In a company, what we want is to reduce people’s effort. We’re happy if artificial intelligence allows people to put in less effort to sign the same number of contracts or to review the same amount of legal documents or X-rays. At a university, it’s different. We don’t seek to eliminate students’ effort. At the university, we seek to create effort for students. Effort is the mission. It’s like a bicycle. If I want a bicycle to get from one place to another, I prefer an electric bicycle, but if I want a bicycle to exercise, I prefer a regular bicycle. The same thing applies to our cognitive effort.
All universities are now facing the challenge of how to incorporate AI into our teaching and assessment. Some have chosen to ban its use entirely or in part. For example, this week the University of Chicago decided to ban artificial intelligence in its social science courses. That is one extreme. They are called “analog courses.”
Is it advisable to ban or restrict the use of AI—which graduates will later have to use in their professional lives? Is it reasonable for us to go back decades in our assessments to handwritten exams and oral exams?
Let me tell you what we’re doing and what we plan to do at our university. We promote the use of artificial intelligence. Everyone can use whatever they want, but they must give credit. When students turn in assignments, they must disclose whether and how they used artificial intelligence, especially if any content was generated automatically.
We have a required artificial intelligence course for all degree programs. We support faculty through competitive grants so they can experiment with AI tools in their disciplines.
We’re working specifically to adapt assessments to account for students’ use of AI. Until now, assessments were always based on the product the student submitted—the essay, the homework assignments. Now, evaluating the final product doesn’t make sense because it may have been generated automatically. So we need to evaluate the process; we need to be much more closely involved with the students.
Furthermore, in order to have a better chance in the job market, they need to try to enter the job market as close as possible to a senior-level position. There is currently no decline in demand for seniors, but there is a decline in demand for juniors.
Try to enter the job market with concrete portfolios that show companies you have real, demonstrable, and transferable experience. Another thing we’re doing, for example, is making clinical education a standard part of all degree programs. We try to ensure that students spend an extended period of time working at a real company.
To wrap up, here’s some advice for you on how to use artificial intelligence as high school students—and soon-to-be college students. First, use these models as a learning aid, not as a substitute for cognitive effort. In other words, when you ask artificial intelligence to write your essay on the French Revolution, you’re fooling yourselves. You won’t be able to reproduce that on a test, and you certainly won’t be able to do so in the job market.
Use artificial intelligence as a tutor or a sparring partner; it's very productive. Always verify the "hallucinations," because artificial intelligence is designed to please the user. The problem with that is that, if what I say is indeed nonsense, it will look for an explanation that shows I'm right—but based on what are called "hallucinations."
The great challenge for you—for your generation—will be how to coexist with new forms of artificial intelligence. Until now, intelligence has been the exclusive domain of humans. Don’t try to do better what machines already do very well. Cultivate your humanity; that will be what sets you apart.
Thank you very much.