ChatGPT attests to the advancement of AI, but it has limitations, points out an expert

We are fascinated by evolution. With each sign that we are moving quickly towards the new, comes the frisson. This is what is happening with ChatGPT, which has not left the news for the last few weeks. In fact, Artificial Intelligence (AI) is one of the most disruptive technologies of recent decades, with proven potential to reduce costs and losses, automate processes and increase productivity. But models such as ChatGPT – a prototype of an AI chatbot developed by OpenAI and specialized in dialogue – still need to be parsimony analyzed. The warning comes from an expert in AI and strategy and professor at the Federal University of Rio Grande do Sul (Ufrgs), Luís Lamb. “These models have evolved precisely because the processing capacity of computers is increasing, but they still invent facts, that is, they still do not have a great degree of confidence”, he explains, who is currently in the United States studying an MBA in Innovation and Technology at the Massachusetts Institute of Technology (MIT).

Digital Market – Artificial Intelligence is already a forceful reality. How do you see all this movement around ChatGPT?

Louis Lamb – In fact, there is a lot of marketing and discussion about ChatGPT and other technologies called Large Language Models (LLM). But, in fact, these models have a number of limitations, and the CEO of OpenAI himself, Sam Altman, has already confirmed this. Still, they have made the impression of being grandiose, problem-solving, thinking role models. And actually, it’s not like that.

MD – How do LMM models work?

lamb – LLM models have this terminology because they are trained, that is, they use machine learning algorithms precisely to process a large volume of text. And, from that, and using some Artificial Intelligence techniques, they make predictions of the next sequences. These models have evolved precisely because the processing capacity of computers is increasing. But, there has to be an issue to be understood. These systems still invent facts, that is, they still don’t have a great degree of trust. And there is a semantic problem, that is, they do not base their inferences on a semantic correspondence with reality. They are still not models that we can say offer a great guarantee of producing correct results and that, therefore, can be used for tasks that have a very significant critical issue.

MD – Why was ChatGPT launched before similar solutions from big tech players?

lamb – The technologies behind ChatGPT are also dominated by large research centers and companies such as Meta, Nvidia, among others. OpenAi made this impact, first, because these products are constantly being refined with the interaction they have with humans. And what explains having launched this solution before the giants? That’s a business issue. A company that has a search engine as its core business, for example, can’t offer wrong answers, can it? If you do, you could lose a lot of customers. So, the truth is that big techs also have this solution, but at the same time they have a great responsibility towards their billions of users. If they make a mistake, they will tear down their business model. Is OpenAi the only one that masters this technology? No, it’s not the only one. But it is a company founded precisely to carry out major research and investments in disruptive technologies and, without a doubt, it is generating innovation.

MD – What are the positive consequences of tools like this in our lives?

lamb “We’re going to have a series of consequences on our activities, no doubt. Artificial Intelligence tools are getting better and better at automating strenuous, repetitive and trivial tasks that don’t make sense to be performed by humans. I always like to give the example of when the calculator appeared. At the time, teachers were concerned, thinking that students would just use this tool and that would end math classes. However, from the moment we have access to better technology, we can use it precisely to qualify teaching. The calculator did not end math teachers, math students, or engineers. Computers, as good as they are today, didn’t do away with engineers or programmers. In the specific case of solutions like ChatGPT, it’s the same thing. High quality texts, which require an in-depth analysis capacity and in which the author expresses his reasoning, will still be produced by human beings. What is becoming increasingly clear about the present and the future is the importance of our ability to think, not memorize.

MD – And what are the risks of using it without greater care?

lamb – These systems still have a very limited vision because they do not base their references on reality. This means that the computer does not make a semantic association of meaning when writing a certain text. It is also along this line that the idea of ​​what is called here hallucination arises. What is that? They are ideas, inventions that these language models produce that don’t make much sense or, sometimes, don’t make any sense. This is one of the issues among many others that have to be resolved before these models are seen as capable of replacing all the work of students and certain professions that do text production. These models will get better and better, but there are AI issues that will still have to be resolved before these models can be used. The trend we are noticing at the moment, in terms of the evolution of the economy, is the dominance of data, reasoning and people who have the ability to generate innovative solutions.