Artificial intelligence is sinning from arrogance. It’s not that he’s doing it on purpose, because he doesn’t think or reason, but the algorithms that power tools like ChatGPT o DALL-E They are calibrated in such a way that, when a request is made or a question is asked, they end up providing very detailed answers even though they have no clear basis for generating them. “Excess of trust”, that is the diagnosis of Pablo Martínez Olmos (Granada, 40 years old), a doctor in Telecommunications Engineering who proposes the development of a “humble” artificial intelligence, capable of declaring the level of uncertainty about the results it produces.
“I liked the idea of humble because we think that a person with this virtue is someone who is aware of their limitations and stops when it comes to responding when they have no idea about a topic,” says Martínez, sitting in his office in the Department of Signal Theory and Communications from the Carlos III University of Madrid. The engineer has just been recognized with one of the Leonardo Scholarships that the BBVA Foundation delivers to promote “innovative projects” in areas of science and culture. Humility is, therefore, knowing not to give a very detailed or elaborate solution when you really don’t know the answer. Something that artificial intelligence is failing at.
It is enough to ask DALL-E—the artificial intelligence of OpenIA that creates images from textual descriptions—that generates a detailed map of Spain so that this bias is evident. Martínez has proven it. “The shape of the country is well captured, but then the map includes very detailed aspects of geography, such as mountains, rivers or names of regions, which are totally false,” he points out. Then he adds: “Algorithms are trained with certain design biases that lead them to commit what we call hallucinationsvery detailed outputs that have no basis in reality.” Are hallucinations They mean that, when faced with the same request, in this case, a map of Spain, the tool generates very different results. “In isolation, each map seems to make sense, because they are so finely constructed, but if you look closely they are riddled with errors.”
This causes several problems. One of them is the lack of rigor in the solutions offered by AI. Another, that generative tools are more exposed to malicious use. The algorithms on these platforms have several control mechanisms to limit some responses that could provide potentially dangerous or private information, such as instructions for making a homemade explosive. However, Martínez details that “the heart of the algorithm remains unreliable when it comes to hallucinations and is therefore very confident in the level of detail in his answers.” It works the same as with people: the more confident an algorithm has in itself, the easier it will be to find “the tickle and trick it.” If you ask him directly he won’t tell you, but with a little skill, ChatGPT can end up explaining how to cook meth at home.
Reformulate the algorithms
Martínez points out that his line of research towards a humble artificial intelligence It is based on thinking about how to reformulate algorithms to restrict their ability to provide arbitrarily detailed solutions that are not based on reliable information. The ultimate goal is to make it more useful, reliable and secure. Achieving this involves several computational challenges. “One of the ideas we propose is to make life more complicated for the neural network during its training, using malicious and contradictory attacks,” he points out. It is also important to increase the restrictions that the algorithm has to store all the information on which it bases its responses so that there is “a certain order.”
Martínez gives the example of a small child’s toy room. There Lego pieces, action figures, paintings and decks of cards coexist chaotically. At first glance everything is scattered around the room without any criteria, but the child, with a great ability to remember details, knows perfectly where everything is, so when he wants to put together a particular puzzle, he will know where to go to look for the parts. Only that, due to the disorder, he will use parts from other games that, on this occasion, will not make sense for the puzzle he wants to put together. However, if each game was stored in a specific drawer or case, elements from other games would be less likely to sneak in. The same thing happens with the information stored by algorithms, a bit of order is needed to shorten the margin of error. Just as it is difficult to educate a child to maintain this structure, something similar happens with the neural network of a generative artificial intelligence in computational and mathematical terms.
The engineer knows that he cannot compete with the development of large technology companies, but his project tries to propose a more robust algorithm training methodology, which leads to the development of more reliable tools. Although no solution is magic. “The word ‘intelligence’ in all of this is very confusing. “I prefer to talk about mathematical functions that we are interpolating and that, for now, cannot do without the critical capacity or supervision of the human being,” he says.
This idea of a humble, safe, traceable and controlled artificial intelligence would have, Martínez believes, a concrete application in the medical field. Detecting unknown biomarkers, designing personalized therapeutic effects, reducing the negative consequences of certain treatments and improving people’s quality of life are some of the potential uses that the design of algorithms could have if they were a little more humble.
