Jürgen Schmidhuber has been striving since he was a teenager to build artificial intelligence more intelligent than himself (Jürgen Schmidhuber)
About eight years ago, the New York Times said that when artificial intelligence matures, Professor Jürgen Schmidhuber may be called “Dad.”
At a time when children have matured and become an integral part of our daily lives, many experts, including Geoffrey Hinton, who is also one of the fathers of artificial intelligence, expressed their fear that their children would dominate humans. In May 2023, he even announced his resignation from Google, Claiming that this would give him the opportunity to speak freely about the danger of his children, he was not ashamed to disavow them, saying that he "regrets working in this field."
In contrast to Hinton, Schmidhuber, the current head of the Artificial Intelligence Initiative at King Abdullah University of Science and Technology in Saudi Arabia (KAUST), fiercely defended his children amid this torrent of “shock talk that spreads frustration” by focusing only on the negative aspects. The man - who also holds the position of director, went on at length. The scientific laboratory of the Swiss Artificial Intelligence Laboratory - in talking about “the positive role that its children play in facilitating human lives, which far outweighs any talk about their negatives.”
In an email interview with Al Jazeera Net, Schmidhuber spoke with great enthusiasm about the advantages of one of them, which is language processing models such as Google Translate and Apple's Siri, whose development began from the neural networks that he worked on in the 1990s. He also spoke about the roles of generative artificial intelligence, Such as "GPT Chat", the development of which was based on his research and applied achievements in the fields of "deep learning" and "unnatural linear transformers".
In words that were not devoid of frankness at times, and a deep philosophical vision at other times, the man - whom Elon Musk once said that he “invented everything” - refuted what many people say about the dominance of artificial intelligence applications over humans, and provided an explanation for the fierce attack to which these applications are subjected. He repeated what he had repeated since he was a teenager, which is that he hopes to “build an artificial intelligence that is smarter than him so that he can retire.” The following is the text of the dialogue.
Scientific Director of the Swiss Laboratory for Artificial Intelligence Jürgen Schmidhuber speaks at the Global Summit on Artificial Intelligence for Good, held at the International Telecommunication Union in Geneva in June 2017 (Jürgen Schmidhuber)
I read the details of many press interviews conducted with you, including an interview published by The Guardian newspaper on May 7, 2023, in which you try to remove the fears that were raised about artificial intelligence, and it seemed to me as if you transferred artificial intelligence from the satanic state that some are trying to portray, to an angelic state. But in reality, I did not find strong evidence for that angelic state in your words. Can you explain to me in more detail: What is the evidence on which you based your opinion?
There is a ton of evidence of “good AI.” The AI we have been developing since the 1980s makes human lives longer, healthier, and easier. For example, in 2012, my team had the first artificial neural network that won a medical imaging competition, a biopsy competition. Breast cancer, this helped revolutionize healthcare, and today everyone uses such methods.
Furthermore, one of our most famous AI systems is called Long Short-Term Memory, and a simple search on Google Scholar brings up countless medical articles that have employed this system in many applications, for example to improve disease detection. Diabetes, arrhythmia diagnosis, cardiovascular disease risk factor prediction, ECG temporal signal analysis and classification, 4D medical image segmentation, radiological abnormality detection, automated sleep stage classification, blood glucose level prediction, and lung cancer detection. , respiratory prediction, real-time tumor tracking, breast cancer detection from pathological anatomical images, protein secondary structure prediction, genomic data modeling, detection, classification and prediction of Covid-19, and many others.
Our AI is also used to improve chemistry, design molecules, classify documents in courts, read lips, map brain signals, predict what happens in nuclear fusion reactors, stock market forecasting, self-driving cars, diagnose industrial faults, predictive maintenance, and grid applications. Smart, Sustainable Energy Management in Microgrids, Anomaly Detection for Industrial Big Data, Traffic Congestion Forecasting, Education, Tutoring, Teaching, Analysis, Water Quality Forecasting, Air Pollution Forecasting, Renewable Energy Resource Forecasting, Fine Dust Concentrations Forecasting, Solar Radiation Forecasting in Under complex weather conditions, fish tracking, forecasting vegetation dynamics, climate change assessment studies on droughts and floods, and many more.
Our artificial intelligence has also succeeded in breaking down the ancient barriers of communication between people and countries. I remember when I went to China 15 years ago, I had to show the taxi driver a picture of the hotel I was staying in so that he would know where I wanted to go, and after a few years he was talking to me on his phone. The smartphone was in Mandarin (a group of Chinese dialects spoken locally in most parts of northern and southwestern China) and I heard the translation, said something and the smartphone translated it back into Mandarin.. The taxi driver probably had no idea this was being played Using techniques developed in my labs in Munich and Switzerland in the 1990s and 2000s, by 2017 Facebook was using our Long Short-Term Memory systems to perform 4 billion translations per day.
ChatGPT and other similar popular models also use the principles of neural networks that I first published in 1991 (for example, non-normal linear transformers). At that time, computing was very expensive, but today it is millions of times cheaper. Our old technologies now facilitate many office tasks, such as summarizing documents, making illustrations, and so on.
So let us repeat: “Our artificial intelligence is nothing to fear. It makes human lives longer, healthier and easier.”
Jürgen Schmidhuber: The artificial intelligence I have developed since the 1980s makes human lives longer and healthier (Jürgen Schmidhuber)
I appreciate your enthusiasm and zeal for your work, which seems clear in your long and detailed answer, but with all these benefits that you mentioned, which of course have greatly facilitated people’s lives, some have spoken about the harms of artificial intelligence, especially with the emergence of “GPT chat”. For example, he spoke Experts have revealed the possibility of using the latest version of it, “GPT4”, in the field of producing chemical weapons. This may be the reason that prompted Jeffrey Hinton to resign from Google in order to speak freely about the harms of artificial intelligence, so why not also talk about its harms, so that light can be shed. On her and warn her?
Of course, all technology can be used for good and evil. For example, about 800,000 years ago, fire was controlled by early humans and used to keep warm at night and for cooking. However, one can also use it to burn other people.
Fire also has an ability similar to artificial intelligence to spread widely, and for this reason, 800 thousand years ago, they established an ethical committee to discuss the pros and cons of fire, and they concluded that the pros outweigh the cons to the point that it was recommended to continue working on its development, and for this reason it has helped in many areas. Life, and still helps today.
So we can say that in general, 95% of all AI development is about helping humans, and there is huge commercial pressure towards such “good AI”, and this is because companies want to sell you their AI devices, so that you do not buy Only artificial intelligence systems that you consider useful.
As AI becomes ten times cheaper every 5 years, everyone will benefit more and more from it, in line with the old slogan of our company “Naissance”, which is: AI for everyone.
Of course, one cannot deny that militaries are conducting research on smart robots based on artificial intelligence as well. For example, a person familiar with the work of militaries told me that my “long-short-term memory” system is also used to guide military aircraft without... pilot.
But there is an old example dating back to 1994 that reveals a positive side of artificial intelligence, whether in life or military work. Just as Ernst Dieckmans in Munich owned the first self-driving car in highway traffic, similar machines could also be used by the military in search of... Autonomous land mines.
Even as you talk about the good and bad of technology, I see you tend towards over-talking the positives.
Because, as I told you, there are hidden intentions other than what is exported and announced regarding the attack on artificial intelligence. We should be more afraid of the half-century-old technology represented by hydrogen bomb missiles, as a single hydrogen bomb can have greater destructive power than all weapons. World War II combined, many have forgotten that despite the dramatic nuclear disarmament since the 1980s, there are still enough H-bomb missiles to wipe out civilization in a few hours without any AI, so AI does not present a new kind of existential threat.
A photo of “Jürgen Schmidhuber” taken on an occasion in 2019 and treated in the style of the famous painter Van Gogh using artificial intelligence (Jürgen Schmidhuber)
Why, then, did Geoffrey Hinton resign from Google? He explained in his reasons for his resignation that he feared the spread of artificial intelligence for humanity, and he even expressed his strong remorse for working in this field?
In short, he is just seeking publicity. He is well trained in psychological matters, and certainly knows that doomsday scenarios and talk of dangers receive more attention than positive predictions, so he is now repeating arguments against artificial intelligence that others put forward decades ago, and I have criticized him on previous occasions. Because of that, I also criticized him for republishing artificial intelligence methods and ideas without giving credit to their owners.
Hinton was not the only one who spoke about the harms of artificial intelligence. There were many experts who signed a petition prepared by the “Future of Life” Institute in which they demanded a truce during which artificial intelligence systems would stop developing for a period of 6 months until the rules governing them were established, and Elon Musk was one of them. The signatories, do I understand from the context of your answer to my previous questions that there are commercial purposes behind this petition?
I did not sign this petition, none of the major powers in the field of artificial intelligence were interested in it, and some of those who signed it did not stop their work in the field of artificial intelligence at all, because if Company A, for example, stopped developing, then Company A would not do so. b) That, even if it announced that it would also stop, and the latter would gain an advantage over it.
I understand that you do not agree with what the European Union announced last December regarding reaching a “political agreement” on artificial intelligence legislation aimed at encouraging European innovation in these advanced technologies, while limiting their potential misuse?
Last year, I tweeted about the ridiculous hype about regulating AI. One cannot regulate AI research, just as one cannot regulate mathematics, yet one can regulate its applications in finance, automobiles, and healthcare, because such fields already have regulatory frameworks. Constantly adaptable anyway.
I also signed a petition for the Broad AI Open Network, a non-profit organization that makes machine learning resources available to the general public, saying: Don't stifle the open source movement.
I believe that some of what was stated in it has been heard by European decision-makers, and that current artificial intelligence regulations are not as restrictive, as some feared.
According to what I read and your answers, chatbots such as “GBT Chat” were developed based on some of your research. This leads me to ask you about the academic fear of misusing these tools in scientific research, and as a university professor, what is your advice to researchers regarding positive exploitation? For these tools, how can you, as a university professor, detect misuse of this tool by a researcher?
No one has the means to detect all potential misuses of AI around the world, yet the same technologies that can be used to implement “bad AI” can also be used by “good AI” to fight “bad AI.” .
I had an experience in which I asked questions on GBT Chat related to the ongoing war between Israel and the Gaza Strip, and I was surprised that the answers were biased toward one party over the other. What is your interpretation of this bias, and how can it be addressed?
Like humans, AI will be biased towards the data it was trained on.
Jürgen Schmidhuber (right) with Jensen Huang, co-founder, president and CEO of NVIDIA (Jürgen Schmidhuber)
Many of those who knew that I would be conducting a press interview with you asked me about the future of the professions in which they work, and whether they will be affected by artificial intelligence or not. And let me define these professions for you: journalist, doctor, and programmer. What is your assessment of the future of these professions in the age of artificial intelligence, and in general, what are the professions? In danger of extinction?
For a long time people thought that office jobs might require greater intelligence than manual occupations, however it has now been shown that it is much easier to replace certain aspects of office jobs than, say, the job of a carpenter.
There are now artificial intelligence systems that can read a lot of documents and then make great summaries of them, and that's an office job, or you can give some systems a description of the illustration you want for your article and to create good illustrations that might only need minimal fine-tuning, or you tell them like a programmer What your next program should do and they come up with a very good first draft of the code.
All of these office tasks are much easier than real hard tasks in the physical world, and it's interesting that things that people thought required intelligence, like playing chess, programming, or making PowerPoint slides, are much easier for machines than they thought.
But for things like playing soccer, there is no physical robot that can remotely compete with the abilities of a young child with these skills, and it is interesting that AI in the physical world is much more difficult than AI behind a screen in virtual worlds, and it is really interesting in my opinion that We see that jobs like plumbers, electricians, or carpenters are more difficult than playing chess or writing a news story.
In general, my old statement from the 1980s still applies: It is easy to predict which jobs will disappear, but it is difficult to predict which new jobs will be created. 200 years ago most jobs in the Western world were in agriculture, today they are limited to 1 in 2 % Just.
Likewise, many decades ago people expected industrial robots to cause a lot of job losses, yet unemployment rates today are low in countries with a large number of robots per capita (such as Japan, Germany, Korea, and Switzerland). Why?
Because a lot of new jobs have been invented.
For example, 40 years before the World Wide Web was created at CERN, Switzerland, who would have thought that all these people would be making money as vloggers on YouTube?
As we speak with one of the founding fathers of artificial intelligence, we must not miss your comment on the lawsuit filed by Elon Musk against the “Open AI” company, the owner of the “GBT Chat” application. What is your comment on his accusation that the company is deviating from its original mission, and that Should it be renamed from “open AI” to “closed AI”?
In fact, many have argued that OpenAI should switch from open AI to closed AI, but in any case, the “true” open source movement may be lagging behind major AI companies like OpenAI by about Only 6 to 8 months, so all the temporary advantages of the currently leading players in the field of AI tend to erode quickly.
As I mentioned previously, every 10 years computing becomes 100 times cheaper, meaning that artificial intelligence will not only benefit a few large companies, but everyone will benefit more and more from cheap artificial intelligence, in line with the old slogan of our company “Naissance”, which is “Intelligence Artificial for everyone.
Photo from 1963: Jürgen Schmidhuber's father used to beat him at chess, often in less than 90 moves (Jürgen Schmidhuber)
As head of the Artificial Intelligence Initiative at King Abdullah University of Science and Technology (KAUST), what have you accomplished through it so far?
In fact, in 2021 I began running KAUST's ambitious AI initiative, the goal of which is to advance basic AI research on all fronts, in line with my long-standing goal since I was a teenager, which was to "build an AI smarter than I am." So I can retire."
Supported by good university governance, and the university's outstanding interdisciplinary expertise in many different areas of science, our AI initiative supports AI applications in all fields, including healthcare, drug design, chemistry, materials science, speech recognition and natural language processing. , automation, robotics, soft robotics, and other fields.
The university's outstanding visual computing team was among the best groups at leading conferences in the discipline, and I am particularly proud of my wonderful colleagues, including Bernard Ghanem, Peter Richtarek, and Francesco Orabona, and their fantastic teams in the university's AI initiative. For example, Bernard had He and his team submitted 7 research papers at the “Computer Vision and Pattern Recognition” conference in 2022, which is the most influential scientific conference in the world, and 8 research papers at the same conference in 2023. Peter and his team had 12 research papers at the “Norebus” conference in 2022, which is a pioneering conference. On artificial intelligence and neural networks.
In 2016, the university was named the university with the highest impact in the discipline, beating out usual contenders such as the California Institute of Technology, and has since topped the number of citations per faculty in the QS World University Rankings. The university is full of distinguished academics from all over the world. The world enjoys excellent working conditions and a high quality of life.
Finally: What do you hope for by leading this initiative?
I hope that the new artificial intelligence initiative will contribute to creating a new golden age of science, similar to the Islamic Golden Age that began more than a thousand years ago when the Middle East was leading the world in the field of science and technology, especially in the automatic processing of information. Remember that the word “algorithm” Derived from the works of Al-Khwarizmi from around 800 AD.
Source: Al Jazeera