DREAM / The Prime Minister resigns and wakes up under the Parisian bridge!

 Kliko këtu për të tjera
Voxpress:

BATTLE/ The four clans that command the Artificial Intelligence

2023-11-29 19:15:54Kosova&Bota SHKRUAR NGA REDAKSIA VOX
BATTLE/ The four clans that command the Artificial Intelligence
The four clans that command over Artificial Intelligence

Techno-apocalyptics, utopians, those concerned with ethics and security: the four groups dominating the debate over AI development. Whose ideas have ignited the battle for OpenAI leadership.

The saga that has had as its protagonist the company OpenAI (which created ChatGPT) in recent weeks, has brought to light a conflict that has been going on for some time, and that has involved the Artificial Intelligence sector.

In fact, at the origin of Sam Altman's firing (then reinstatement) lay major differences between OpenAI's founder and the board of directors (whose members were subsequently replaced, almost entirely) over the proper way to further develop this innovation. So opposite visions, which sometimes take on ideological nuances.

On the one hand, the previous board of directors – led by chief scientific officer Ilya Sutskever – considered artificial intelligence an inherently dangerous technology that needed to be handled with care. On the other hand, Sam Altman, convinced - as he wrote in a post a few months ago - that strong artificial intelligence (that is, able to achieve human capabilities from any perspective) "can elevate humanity".

What happened inside OpenAI follows the conflict that has been going on for some time in Silicon Valley. This clash has at its opposite poles the so-called doomers (the techno-apocalyptic, who are convinced that a strong AI poses "existential dangers" for human beings) and the techno-utopians, who are sure that this technology will lead us to a better future, and will allow human beings to reach levels of well-being never seen before.

By exploiting their economic, political or academic influence, these groups are leaving in the background those - like most computer scientists - who are trying to focus on more concrete and realistic issues.

Techno-apocalyptics

In 2014, Elon Musk declared that the development of Artificial Intelligence was the technological equivalent of "summoning the Devil". At the time, generative AI (capable of producing text itself like ChatGPT, images like Midjourney or even sounds and videos), was still a long way off. The most widespread applications of "deep learning" were related to image recognition or personalized suggestions. However, Musk, certainly influenced by the writings of the controversial philosopher Nick Bostrom (author of the essay Super-intelligence), warned at that time against what he considered to be the great dangers of this technology.

Since then, Musk has become a leader of techno-apocalyptics and the theory of "existential danger", according to which the development of Artificial Intelligence will inevitably lead to a super-intelligence, which in turn can escape the control of human beings and rebel against them.

This vision prompted him to found OpenAI, which aimed to develop Artificial Intelligence in a "safe and responsible" way. Over the years, the theory of "existential risk" presented by AI has gained more and more support among the most radical members of many think tanks, including the "Future of life" institute, which a few months ago published ​​an open letter calling for the ban on the development of these tools. The theory of "existential risk" was at the center of the recent summit on AI, organized by British Prime Minister Rishi Sunak.

AI Safety: The guarantors of the safe development of Artificial Intelligence

It is the group that, above all other considerations, wants the development of Artificial Intelligence to happen safely. He researches at every moment the most promising ways to prevent AI from escaping our control and rebelling against human beings, even without realizing it.

Imagine the following and now well-known scenario (which we again owe to Bostrom): a particularly advanced Artificial Intelligence takes command to maximize the production of paper clips. Taking its given objective literally, this AI consumes all of our planet's resources to produce as many paperclips as possible, inadvertently causing the extinction of human beings as well.

As Melanie Mitchell, professor of Complexity at the University of Santa Fe, points out, "we want machines to do what we want them to do, but not necessarily what we tell them to do." Is it possible to achieve this objective, and ensure that Artificial Intelligence is able to contextualize and balance our commands, interpreting them as we humans do?

According to this thesis, conveying our human values ​​to machines would allow them to interpret commands correctly, not limited to "maximizing the objective function" (that is, to complete the task given to them, in the most efficient way possible), but independently understanding what we humans really want, and what the limits and constraints are that must be respected (for example, avoiding destroying the planet to produce an excessive number of paper clips).

However, the problem of how to effectively transmit our values ​​to these machines remains open and far from solved. Currently, several experiments are underway, which have not yet led to relevant results. Among the followers of this school of thought is the former scientific director of OpenAI Ilya Sutskever, the founder of DeepMind, Demis Hassabis, those of Anthropic Dario, etc. This movement has produced some important scientific work, but often approaches the more extreme positions of the techno-apocalyptic.

The techno-utopians

The leader of the most optimistic current among experts regarding the development of Artificial Intelligence is undoubtedly the very powerful investor Marc Andreessen, founder of the company A16Z, and author of a recently published essay entitled "Artificial Intelligence Will Save the World" .

Of course, even in this group, attention is paid to the possible dangers that come from Artificial Intelligence. And there is no doubt that they also agree with the prediction that a very powerful Artificial Intelligence can be born in a few years. What sets them apart from others is the belief that if developed in the right way, a technology will lead the world to a new socio-economic utopia.

This is a vision described at length in a recent essay entitled "AI 2041" by computer scientist and investor Kai-Fu Lee, where through a science fiction story, he describes how Artificial Intelligence will create an "economy of abundance" , which will free us from any kind of economic difficulties, will allow us to extend our life span immeasurably, will free us from the yoke of work, but not only.

For this reason, utopians are against any regulation that could slow down the development of AI. They aim to get to an all-powerful AI as soon as possible. In addition to Andresseen and Sam Altman, figures such as the executive head of Microsoft Satya Nadella or Google Sundar Pichai, are often included in the group of utopians. But are we really sure that their vision is a true vision of the future, and that they are not just interested in maximizing the return on their investment in this sector?

Artificial Intelligence Ethicists

The groups we have mentioned so far are concerned about the future, about the emergence of a super-intelligence, how to avoid it, how to better manage it or make the most of its potential. All these groups also share a very ideological and science fiction interpretation.

Meanwhile, the fourth group undoubtedly has more concrete, realistic and immediate concerns.

They are the experts in the ethics of Artificial Intelligence, who instead of dealing with apocalypse or utopia, try to draw attention to the actual dangers that this technology presents.

Why do ChatGPT or Midjourney tend to produce stereotypical and discriminatory images of society, and what are the dangers of a technology that tends to perpetuate the status quo, rather than help us overcome it?

What are the dangers of a purely statistical tool, used in very sensitive areas such as surveillance, health care, job selection, etc.? To this group – mostly from universities – we owe the discovery of how (and why) the Amazon algorithm used for personnel selection systematically discriminated against women, how facial recognition systems used by US police are deeply flawed, more often when it comes to identifying women or people belonging to ethnic minorities, etc.

The best-known exponent is researcher Timnit Gebru, author of a 2020 study on the risks of discrimination inherent in systems like ChatGPT (the study that led to her firing from Google, where she headed the ethics committee). So is Joy Buolamwini (founder of the Algorithmic justice league), Kate Crawford (author of the essay "Neither artificial nor intelligent"), and University of Washington linguist Emily Bender

Coming primarily from the academic world, and inevitably endowed with much lower economic resources and media power than Musk, Altman et al., Timnit Gebru and other researchers are managing to at least partially counter the futuristic visions of other groups. . Thanks to them, the European AI Act focuses above all on concrete risks. His American counterpart does the same.

One can only hope that their influence will continue to grow. And the media also have their responsibility for this./ WIRED Italy


Në përputhje me standarded JTI

Video

SHIKO KANALIN NE YOUTUBE
×