Generative AI could create more jobs than it destroys…


Everyone agrees that generative artificial intelligence (AI) tools can save time and increase productivity.

But while these technologies make it easier to run code or produce reports quickly, the work of building and maintaining large language models (LLM) can require more human effort than the effort saved downstream.

Additionally, many tasks don’t necessarily require AI firepower. Standard automation is sufficient.

LLM is complicated to implement

That’s what it says Petar Cappelliprofessor of management at the Wharton School of the University of Pennsylvania, who spoke at a recent MIT event. Simply put, generative AI and LLM can create more work for humans than make tasks easier.

Why ? Because LLM is complicated to implement. And “it turns out that generative artificial intelligence could do a lot of things that…we don’t need,” Mr. Cappelli said.

Certainly, AI has been presented as a game-changing technology. “Technological projections are often wrong,” the professor puts into perspective. “Most birth predictions have been proven wrong over time.” The wave of driverless trucks and cars expected in 2018 is an example of these optimistic predictions, which have yet to materialize.

An example of the imminent arrival of an autonomous vehicle

Above all, grand visions of digital transformation often get lost in utopian details.

Proponents of autonomous vehicles have focused on what “driverless trucks could do, rather than what needs to be done and needed to avoid regulations — insurance issues, software issues, and all of those issues,” the MIT professor points out. And.

In addition, Mr. Cappelli added: “If you look at their actual work, truck drivers do a lot of things other than driving trucks, even long distances. »

Programmers “spend most of their time doing things that have nothing to do with computer programming”

A similar analogy can be made with the use of generative artificial intelligence.

Programmers “spend most of their time doing things that have nothing to do with computer programming,” he says. “They talk to people, negotiate budgets, etc. Even on the programming side, all of this is not programming, strictly speaking.”

The technological possibilities of innovation are fascinating, but their implementation is slowed down by the reality on the ground.

Generative AI and operational AI “generate new business”

In the case of generative AI, the labor savings and productivity benefits can be offset by the amount of upfront work required to build and maintain the LLMs and algorithms.

Both generative AI and operational AI “are generating new business,” Cappelli noted. “People have to manage databases, organize documents, sort out these issues of conflicting reports, validity, etc. That’s going to generate a lot of new tasks, and somebody’s going to have to do them.”

According to him, operational artificial intelligence, which has been around for some time, is still being developed. “Machine learning is significantly underutilized. This is partly about database management issues. It takes a lot of effort to collect data to be able to analyze it. Data is often in different silos within different organizations, which is politically and technically difficult to achieve.”

5 problems that generative artificial intelligence in business must overcome

Mr. Cappelli lists several challenges in moving toward generative artificial intelligence and the LLM:

Solving a problem or opportunity using generative artificial intelligence and LLM can be overwhelming

“There are many things that large language models can do that probably don’t need to be done,” he said.

For example, business correspondence is considered a use case, but most tasks are already done through letter forms. Add to that the fact that “lawyers have already approved the letter form, and anything written in big language forms probably needs to be seen by a lawyer. And that’s not going to save time.”

It will be increasingly expensive to replace automation with artificial intelligence

“It is not clear whether large language models will be as cheap as they are today,” warns Cappelli.

“As more and more people use them, the infrastructure must grow, and thus the demand for electricity. Someone has to pay for this.”

Humans are needed to verify the generative results of artificial intelligence

The results of generative artificial intelligence can be suitable for relatively simple things like email. But for more complex reports or projects, you need to confirm that everything is correct. “If you want to use it for something important, you have to make sure it’s accurate.

And how do you know if this is the case? It helps to have an expert, someone who can independently confirm and is knowledgeable about the subject.

It is important to ensure that there are no hallucinations or bizarre results and that the data is up to date. Some say that other major language models could be used to assess this, but this is more a question of reliability than validity. We have to check it somehow, and it’s not necessarily easy or cheap.”

Generative artificial intelligence will drown us in a flood of sometimes contradictory information

“Because it’s so easy to generate reports and results, you’ll get a lot of responses,” Cappelli said. In addition, the LLM can provide different answers to the same question. “This is a question of reliability – what would you do with your report?

You produce one that gives a better picture of your team and you give it to your boss.”

In addition, he warns: “Even the people who make these models can’t give you clear answers.” Are we going to drown people in evaluating the differences between these results?

People always prefer to make decisions based on their intuition or personal preferences

Machines will struggle to overcome this problem. Companies can invest significant sums in creating and managing LLMs to select job candidates. But study after study shows that people tend to hire people they like.

“If you build an AI model on this, you’ll find that your superiors, who are already making decisions, don’t want to use it.”

According to Cappelli, the most useful application of generative artificial intelligence in the near future is to search data stores and provide analytics to support decision-making processes.

“Right now we’re washing data that we couldn’t analyze ourselves,” he said. “AI will be much better than us at this,” he added. In addition to database management, “someone has to worry about firewall and data pollution issues.”



Source link

Leave a Comment