Noemí G. Gómez |
Madrid (EFE).- The words “intelligence” and “artificial” were first linked in the 1950s, and although great progress has been made since then, the real boom has only now occurred. 2023 was the year of “everyday” artificial intelligence, which still has a number of problems: protection of minors, privacy, misinformation or prejudice.
Although it has been used for years in voice assistants, in monitoring social networks, in navigation systems or in customer services, the real revolution of this technology has been brought about by generative artificial intelligence (AI) models, especially ChatGPT, capable of “conversing” and creating texts, photos, videos or music from thousands of existing data.
“Despite the intense presence of artificial intelligence in our lives – the question should be asked almost where it is not present – it is true that ChatGPT has brought artificial intelligence into the headlines of the media and public discourse,” Nuria tells EFE. Oliver, co. founder and scientific director of the ELLIS Alicante Foundation, focused on researching ethical, responsible artificial intelligence for social good.
Probably one of the main misconceptions until it came along was that it was made up of robots, the Massachusetts Institute of Technology (MIT) AI doctor continues. Now that tools like ChatGPT have become increasingly democratized to access and use, the company has realized that they are actually software, not necessarily bots.
The American company OpenAI was the first to take a piece of the cake of this glory when it made its ChatGPT chatbot available to citizens on November 30, 2022. A few months later and in 2023, it launched its -controversial – improved version (GPT-4) and competitors like Google entered the scene with its Bard model.
Based on machine learning, these and other artificial intelligence systems have begun to creep into everyday classroom life, medical diagnoses, financial markets, scientific and space research, or weather forecasting.
But the dizzying pace at which they have spread, especially generative artificial intelligence, has raised concerns about data and privacy, bias or artificial intelligence as a new way to violate the rights of women and girls – as in the case of fake acts of minors in Almendralejo, Badajoz.
Rethink AI
It is this speed that prompted more than 33,700 experts from various fields around the world – including Spanish – to come together in an open letter last March, asking the laboratories to suspend training of the most powerful systems for at least six months. than GPT-4.
Behind this requirement, also signed by well-known businessmen in the technology sector, was the need to foresee the possible consequences of its use and to regulate it.
However, this letter was responded to by other scientific and technological sectors, who saw it as a distraction to avoid addressing the real problems of artificial intelligence, the development of which, on the other hand, is unstoppable and necessary to solve major challenges such as climate change.
Still, OpenAI, which now has a significant investment from Microsoft, has agreed to halt development of ChatGPT-5 for now, although last month it announced the release of “GPT-4 Turbo,” its most powerful trained chatbot with information by April 2023.
Despite these inconsistencies, most experts agreed on the need to regulate (not ban). The European Union has just agreed on the world’s first artificial intelligence law.
Still provisional – it will have to be ratified by the European Parliament and the Council of the EU – the standard allows or prohibits the use of artificial intelligence depending on the risks it entails.
For example, it bans all biometric categorization systems based on political, religious or philosophical beliefs, or racial and sexual orientation, and vetoes those that can detect emotions in workplaces or schools.
The evolution of the law, slower than advances in this field of computing, has had to adapt to developments, namely generative artificial intelligence systems, which is another of its strengths.
In Europe, they will have to meet transparency criteria such as specifying whether a text, song or photo is generated by artificial intelligence and guaranteeing that the data used to train the systems respects copyright.
And the danger and concern with this type of artificial intelligence is that it strives for a believable result that isn’t necessarily true. Examples abound, such as hyper-realistic photos of Pope Francis with white feathers or Donald Trump resisting arrest.
But also much more serious actions, because they are small, like the case of fake nude pictures of Almendralejo girls.
It was this matter that led Spain’s Data Protection Agency to launch an ex officio investigation last September. This was not the first time she had acted on AI issues.
It also launched preliminary investigative steps against OpenAI for possible non-compliance with Spanish and European data protection regulations.
With the explosion of ChatGPT and other similar tools, users are uploading a lot of information, not only their own but also that of others, and there is uncertainty about what data companies collect, what they do with it if it is passed on to third parties. or if it is an international transfer.
AI to survive as a species
“AI has enormous potential to help us solve the great challenges of the 21st century (…). In fact, we know that we need AI to survive as a species,” emphasizes Nuria Oliver for EFE, but at the same time it is not without limitations and presents ethical challenges that must be addressed to ensure a truly positive impact.
Challenges beyond privacy or cybersecurity include “deepfakes” that are used to spread fake news and revenge porn, algorithmic biases or stereotypes, lack of diversity in the teams that invent AI systems – only 12% of researchers in this field are women, according to UNESCO – or the carbon footprint it creates.
2023 was not just the year of ChatGPT – who was chosen by Nature magazine as one of the scientific (non-human) characters of the year – Bard, Gemini (also from Google) or Bing (Microsoft). ESMFold, an artificial intelligence system implemented by Meta, was able to predict the structure of more than 740 million proteins.
As published in The Lancet Oncology, AI screening was shown to detect 20% more breast cancers compared to a radiologist’s routine double reading of mammograms, and machine learning was used to create a single-cell atlas of whole human lungs.
Google unveiled GraphCast, an AI capable of predicting dozens of weather variables for 10 days across the planet in less than a minute, research published in Science included Spanish participation.
And a few days after starting (or continuing) with nougat, Campofrío presented its traditional Christmas ad, this time written with AI.