The academic Asunción Gómez Pérez. Ximena and Sergio
Asunción Gómez-Pérez (Azuaga, Badajoz, 56 years old) is an eminence in the field of artificial intelligence (AI).
She is among the 2% of the most cited academic authorities in the world, according to the index of Stanford University, where she studied, in addition to training at the Polytechnic of Madrid, of which she is today vice-rector for Research, Innovation and Doctorate.
What this computer scientist deals with is the ontology of AI, that is, translating and defining in machines the common language that they must use to reason and identify what they are referring to.
For this reason, since last May 21 she has occupied the chair q of the Royal Spanish Academy (RAE), to accompany the institution in the challenges that its field faces regarding the language.
Her profile is so unusual within the institution that her response to her entrance speech—Artificial Intelligence and Spanish Language—was responded to by Santiago Muñoz Machado, her director.
The challenges of a language career in the technology space are enormous.
Companies in this field develop their own programs and corpora with sources that are not entirely rigorous and the RAE has decided to get involved so that Google, Amazon or Microsoft, for example, use the language well in this area.
Won't your entry into the RAE be a ruse so that when you meet on Thursdays they will take the opportunity to ask you about their problems with cell phones?
No no no…
Sure?
Nothing nothing.
They have a department in the house that takes care of that.
They are not abusing, I tell you no.
Did you follow your speech well?
It was very technical.
I think so, many are interested in corpus issues, intellectual property rights and that artificial intelligence is respectful of previous creations.
When did you first see a computer?
In the third year of BUP, he must have been 15 or 16 years old.
What did they tell you was inside?
A friend of mine named Manuel Monterrubio taught it to me. He had already made his first steps with Basic and the Spectrum that existed then, but I came to computer science when at COU I decided to do a degree that had more mathematics and physics instead of Pharmacy.
My family comes from that field.
My father was a doctor.
And he moved to Madrid…
Yes, my cousin was doing Exact Sciences in the Computer Science branch and she convinced me.
She told me that apparently this was going to have a great future as a preamble to Computer Science.
I entered the Polytechnic in 1985.
That then was an unknown language.
Yes, in those years you could only study Computer Science in Madrid, the Basque Country and Catalonia.
I came here because it was seven hours from Badajoz, closer.
At that time, we didn't know very well what computing was.
We did internships on computers that were in a computing center and we had to reserve the hours.
I discovered what computing was as I spent six years studying.
When did you realize that what you had chosen was going to be the thing that would change the world to what we have today?
I think I realized it after finishing the race.
In third year I perceived the future that digitalization would have, but what hooked me was artificial intelligence.
Already in third year we were creating expert systems to detect faults in trains.
Something practical to apply in real life.
Did you come from the analog world and understood that we would end up heading to the digital world?
Yes, exactly.
In the late eighties and early nineties we were in elementary processes that would become fundamental for the entire world.
And artificial intelligence, then, at what stage was it?
AI tries to provide machines with cognitive capabilities.
We were trying to see how certain types of problems that man solves could be tackled by computer programs.
Expert systems allowed problems and solutions to be conceptualized that were then implemented in a computational language.
The Internet then was not available to everyone.
Computers did not have the computing capacity they have today.
We worked with toy problems, limited by the amount of data they could handle.
Not like now, when you ask any question on the internet and it automatically answers you.
As you said in your speech, the most powerful
IBM computer
of the time had cost about 25 million dollars and had less capacity than a cell phone that costs 200 euros today.
Does that data quantify the leap we have made in a few decades?
When these computers emerge, they do so to facilitate numerical calculation.
That's why they invent them.
But there comes a time when they try to see if they can go beyond mere calculation and that is when artificial intelligence is born, to see if machines can emulate the reasoning of human beings.
AI was announced in 1956, at the Dartmouth Conference, where universities and companies began to work, but the big leap was made in the 1990s.
Alan Turing, with his Enigma machine to decipher espionage messages, is a precedent in World War II.
But I like to mention Leonardo Torres Quevedo, who designed two machines to play chess.
Someone to claim a lot, indeed.
He was a unique inventor in his time, he not only did that, but also many other things, like the Niagara ferry...
The AI remained very united in the development of its chess research, why?
Because it is a complex game.
It delimits a space, the board, with a finite number of pieces and rules that can be executed.
You can develop a strategy, so it allows you to explore possible solutions up to a certain point.
That is until, in 1996, the Deep Blue machine defeated Kasparov.
The branch in which you have specialized is AI ontology, what does it consist of?
We could divide the field into two branches.
One of them is symbolic, it uses words, ontology.
Another is subsymbolic, learn.
The first reasons and explains, the second, as I say, learns.
Is this second one the one that scares us the most?
It's the one that scares many people, yes.
But well, I don't know if it should be said like that.
Man, it produces so much fear that even those who have taken it to its greatest limits today, like
Sam Altman,
ask that it be regulated.
Yes, but he did it after his company, OpenAI, launched ChatGPT, that is, after putting the product on the market.
The problem is not AI, but how it is used.
That is, are we the problem?
That is.
The technology is neutral.
The problem comes from the use we give it.
It has always been like this.
The problem is us.
Since the ontology uses words, that explains why you have entered the RAE.
So that?
The dictionary includes definitions of words so that we are able to understand behaviors or objects.
Ontology helps the computer identify the properties of an object and differentiate it from others.
Concrete, therefore.
Yes, and it also indicates that there are certain relationships between objects, that the table and the chair usually accompany each other, for example.
That is, do they create dictionaries for computers in their field?
That's right, yes, but it goes further.
The systems take words, transform them into numbers, and then generate words again, which may have lost the meaning of the first source.
How did they convince her to enter the RAE?
The RAE has decided to introduce all AI technologies in its work processes with language.
The institution is concerned about the fact that large technology companies do not generate, let's say, their own dialects of Spanish, they do not invent their own grammar and definitions because that is what the RAE has been for for 300 years.
The Academy is concerned with ensuring the proper use of Spanish on social networks and the media in the different geographical areas that unite more than 500 million people in the language.
To this they add concern about the fact that, in the technological field and the Internet, the dominant language is English.
It is also an important problem.
The technology is created mainly in English, in that we are behind.
For this reason, large technology companies, when they translate, must opt for using the basis of their programs in Spanish, but in correct Spanish.
That requires high quality materials and quite an investment.
That is what the LEIA project is based on, basically, where progress has been made with technology companies.
Which is it?
Google, Amazon, Microsoft and Telefónica, to date.
And Meta?
Not today.
Don't they consider it a priority?
For me, this is more influenced by Google, for example.
When someone was looking for a word in Spanish not long ago, the definition that appeared in first place was that of the Oxford Dictionary, today that of the RAE appears, which is very good.
Another objective of the LEIA project is also to build an open platform in the cloud with all the works and works of the RAE, with resources, dictionaries, grammar, corpus.
In this way we will build new applications that serve the Academy and Spanish speakers in a shared innovation system.
We are starting to work on that.
And do they have sufficient funds?
Until very recently, the RAE has been an institution in which you practically had to ask the State for funds and aid when it deals with something that concerns us all: language.
This costs money.
The Government has provided some funds, but a significant investment is needed that goes beyond the funds from the European Recovery Plan that we have planned.
In total, for the RAE and other institutions, 340 million, as announced by Nadia Calviño at the last Spanish Congress in Cádiz.
Enough?
It's never enough.
It has to be approached as a sustainable investment over time.
We have that very clear.
Since when have AI language models existed in Spanish?
The first bet was a system called María, for the corpora of the National Library and the Barcelona Supercomputing Center.
Then there are others: one called Rigoberta, BETO in 2019 and BERT in 2018, which is developed in Chile.
But now, none of them reach the 30 million parameters that those developed by technology companies have.
Those are the GPT of Spanish.
If those you mentioned first emerge from the academic world and institutions, is there an open struggle between them and those who develop the companies?
Well, those from the private initiative have more computing, more corpus.
The key is in the process.
They take large amounts of texts and words and respond within those parameters.
When generating language models, it is very important to have corpora that are rigorously formed.
The biases will be smaller, the language richer and the geographical variety greater.
The language will be much more reliable and you will not have, say, hallucinations when mixing unrelated content.
If not, in the answers, can a Dada poem appear?
Exactly.
Is the best in this area the famous ChatGPT?
It is the one with the largest market share right now.
But all companies are developing theirs.
Which one do you think will prevail?
Predictions in technology are often wrong.
The one that has the greatest computational infrastructure resources, the one that is trained with the best texts and has good engineers for the algorithms will prevail.
They're all in it.
Do you think that in recent years ethical awareness has improved in the world of digital and technological power?
I think so, that a clear awareness is beginning to emerge, for example, regarding aspects that have to do with sustainability or ecology.
A great step has been taken in regulation with the two laws approved in the European Union, the Digital Act and the AI Act.
This must be regulated responsibly.
Several countries, organizations and companies are aware of this.
Now we are no longer starting from scratch in this sense.
In addition, we have data protection laws in various fields or in the public sector and the health system.
We already have rules.
Have we learned enough from the jungle that was the emergence of the Internet at the end of the last century?
Technology is always ahead, but its responsible use also depends on companies to put safe products on sale that generate trust in citizens and respect fundamental rights.
Is it important, on the other hand, not to increase panic?
Yes, technology helps progress.
It is true that others use it to generate division, confrontation, hoaxes, and do a lot of damage.
The bad thing is that they are usually elements that want to achieve power.
You have to find balance.
I insist, technology companies must think about the utopian and dystopian aspects that their products can generate in the market when they release them.
In the doctoral theses that we direct here we always try to see the positive side and analyze the negative.
At school we have a subject with ethical and legal scope, we force counterweights to be exercised, to create awareness among students of how they can contribute to the common good as citizens.
Many products have gone from the laboratory to society with enormous impact.
There has not been an intermediate process that avoids certain effects.
Then they are the ones who get scared by the use that certain people give them, I repeat.
The monsters created when there is no turning back.
Going back to Altman, he was the one who put the product on the market and he could have thought twice.
He went through with Microsoft and a program, TAY, in 2017, which they removed after two days when they saw that it produced inappropriate behavior, such as racist comments and so on.
Then they joined forces with OpenAI, but anyway...
It had been 90 years since the director of the RAE had responded to an entrance speech.
It was an honor.
Does that indicate to what extent his profile was so rare within the institution and that it could only be explained by a need for strategy with a view to the future?
The director had already begun to send those messages within the institution.
They were worried and concerned about the effect of artificial intelligence in the field of Spanish.
To the point of putting the good health of the language at risk?
In communication between people with digital media, we have messaging applications that recommend a word to us.
They do it through what they learn from us through more frequent use.
And they are very puritans, it is difficult for them to recognize, for example, a relevant and well-introduced comment on WhatsApp.
Well, in part, yes.
But they are in the dictionary!
Yes, but the serious thing is that the words they use produce impoverishment or confusion when expressing them.
If you have a good dictionary it will be enriched with synonyms and no spelling mistakes.
Santiago Muñoz Machado, in his speech, also highlighted that you are among the 2% of the most cited academic authorities in the world according to the Stanford University index.
Do I have to treat her as eminent?
No man no!
There are many researchers, these things make me very ashamed.
Did you blush when the director mentioned it?
I don't know, maybe my blush would show...
Subscribe to continue reading
Read without limits
Keep reading
I am already a subscriber
_