The power of ChatGPT

The potential of the new AI tool demands a study on its undesirable effects.

Artificial intelligence has invaded the public conversation justifiably: the potential that the ChatGPT tool (a kind of robot from the OpenAI company capable of conversing with the user) offers is formidable and millions and millions of people have already discovered its responsiveness. Both the US and China and different European states, including Spain, have activated mechanisms to study possible unwanted effects.

The suspicion of violating the legislation on data protection has been the cause to open an investigation of the Spanish Agency for Data Protection, as it has also announced that the body that coordinates the European agencies will do (while Italy has blocked access to ChatGPT). They argue that generative AI could be feeding on both private data and the billing data of users and companies, including the conversations that each citizen has with the system to train the algorithms.

Do you answer the same question about Joseph Muscat as a PN or PL voter? In fact, there are indications that it invents answers with traces of plausibility but unfounded.

These are legitimate concerns about their impact on multiple humanistic fields – from teaching to journalism or recreational knowledge. The tool is nourished by what it has at its disposal: the entire internet, although it should exclude (according to European regulations) databases of limited access or sources protected by copyright. The discursive consistency of the artificially generated text may be sufficient to convey as truthful what is only plausible or merely coherent. But in countless human matters there is no reference truth, there is no truth about drug use or about sexuality.

The ability to fine-tune answers that are persuasive to the user, according to their own algorithm-trained response expectation, is part of the problem because there is no single answer to most of the questions we humans ask ourselves.

Commissioner Thierry Breton's recommendation perhaps falls short: not only should one be warned that the response is produced by artificial intelligence, but nothing and nobody can guarantee the correctness of the answer offered. It will be necessary to analyze its operation with reliable information and independently monitor the risks to make a prodigious invention profitable, without avoiding the harmful effects that its manipulation may have.

Spain has recently set up in Bilbao an Authority of Artificial Intelligence to grapple with the proper, legal ethical use of AI.