Home » Lights and shadows of AI according to Demis Hassabis (Google DeepMind): “We should worry in three or four years”

Lights and shadows of AI according to Demis Hassabis (Google DeepMind): “We should worry in three or four years”

by admin
Lights and shadows of AI according to Demis Hassabis (Google DeepMind): “We should worry in three or four years”

Demis Hassabisone of the 500 most influential people in the world in the field of AI according to Time magazine, thinks that “in the next five years” innovative devices such as smart glasses and viewers to offer artificial intelligence “more contextual information” – to make machines see in real time what we are seeing, for example – and to make the most of the infinite possibilities that this technology offers.

“Today what smartphones can do is incredible, with Gemini (Google’s most advanced AI, ed.) we have made them even more intelligent but in five years I don’t know if the smartphone will still be the best format for artificial intelligence ”.

Hassabis, head of the research laboratory Google DeepMindwas interviewed by Steven Levy, journalist of the American newspaper Wiredon the occasion of Mobile World Congressthe Barcelona fair that large companies use as a showcase to promote mobile technology.

(reuters)

Hassabis introduced himself to the visitors of the MWC at a particularly delicate moment. Gemini, the AI ​​with which Google responded to ChatGpt, last week caused controversy because its users were unable to generate images of white-skinned people. Big G has “paused” its artificial intelligence to solve this problem. “He will be available again very soon,” said Hassabis, “within a couple of weeks at most.”

The prejudices of AI Google explained why Gemini did not create white people by Pier Luigi Pisa 26 February 2024

“The fact is that if you ask AI to create a person walking a dog, or a nurse in a hospital, it should give you a universal representation of all of this, especially when you take into account the fact that Google serves more than two hundred countries around the world, and that AI doesn’t know where a user comes from or what their cultural background is,” Hassabis explained.

“But a desired characteristic of AI, namely its ability to return a broad spectrum of possibilities [anche in termini di colori della pelle, ndr]in this case it didn’t work the way we wanted,” added the CEO of Google DeepMind.

See also  Happy fans! Epic Games Store will launch "Fortress Heroes" on the Apple iOS platform and return to iPhone and iPad | Community/APP/Internet celebrity

(reuters)

At this point Levy asked Hassabis a more than legitimate, as well as right, question. Isn’t the possibility of directing the results of an artificial intelligence disturbing, if you think about the uses that could be made of it? Google doesn’t want to give even more power to authoritarian regimesfor example, or to far-right propaganda or to all those who intend to discriminate against minorities, right?

“We’ve been thinking about this kind of thing since we founded DeepMind [nel 2010, quattro anni dopo è stata acquisita da Google, ndr] – replied Hassabis -. The ethical principles of DeepMind, which we have had from the beginning, then merged into what are now the ‘Google AI principles’ and which guide us in the development and use of artificial intelligence. There is an ongoing debate, which also affects governments and civil society, about how to prevent the wrong people from accessing this technology for harmful use. We are talking about a complex topic, which also affects l’open source [un tipo di tecnologia, come un software o un modello di IA, aperto a tutti e modificabile, ndr] of which we are great promoters. Just recently we launched Gemma, an AI model for the open source community. Because we know that many people, but also many companies, want to directly develop and control what they do with AI. But every time the theme of malicious use comes back, and I say that today this is not a real problem because we are talking about a young technology. But in a few years, three or four years, when these AIs become more powerful and are able to plan and carry out actions in the real world, the whole society will have to seriously worry.”

Yet Demis Hassabis, who tells the world not to worry too much (for now) about AI, signed in 2023 together with dozens of other researchers, experts and entrepreneurs – including Sam Altmanthe CEO of OpenAI – a short letter with apocalyptic tones in which the effects of an out-of-control artificial intelligence are compared to the lethal effects of a pandemic or a nuclear weapon.

See also  NASA Releases James Webb Space Telescope's "Spider Nebula" Image | HYPEBEAST

Artificial intelligence The strange appeal against AI: “We risk extinction”. But companies continue to develop them by Emanuele Capone 30 May 2023

Hassabis, who boasts a degree in computer science from Cambridge University and a PhD in cognitive neuroscience from the University of London, has been following the progress of artificial intelligence for about 15 years.

In 2010 he started a startup called DeepMind with Shane Leggphilosopher and researcher in the field of AI, e Mustafa Suleymana neuroscientist and entrepreneur.

The three founders of DeepMind had an equally ambitious goal. They dreamed of achieving breaking latest newsa form of artificial intelligence that is commonly referred to as “general”.

For many, breaking latest news, one day, will be able to replace man. “But everyone gives a different definition of this AI,” Levy told Hassabis, inviting him to provide his own. “For us – said the CEO of Google DeepMind – it is a system capable of replicating practically any type of cognitive ability. The point of reference always remains the human brain and its functions.”

How and when will we notice the advent of breaking latest news? Will there be, insists Levy, a “before and an after”? “I believe that the transition to breaking latest news will not be clear-cut but gradual – said Hassabis -. These systems become more and more powerful and my bet is that in the coming years they will also acquire the ability to memorize and plan. But I believe that we shouldn’t wait impatiently for breaking latest news if these technologies, already today, can do something extremely useful in everyday life.”

See also  Bloober Team is working with Skybound Entertainment on a new game -

And useful things, so far, DeepMind’s AI has done several. We’re not talking about summaries or automatically writing an email. The work of researchers of this laboratory – which in 2023 joined Google AI – allowed the AI ​​to beat the greatest Go champions (AlphaGo2016), to predict the three-dimensional structure of proteins (AlphaFold2018) and to find 381 thousand new stable materials (GNoME2023).

Yet despite its contribution to science, the most famous artificial intelligence in the world today it’s probably ChatGptan interface that understands natural language and expresses itself like a human being launched in November 2022 by OpenAI.

(afp)

Demis Hassabis doesn’t think Altman’s company’s AI is superior: “They deserve credit for being good at accelerating Transformer technology [alla base dei modelli linguistici di grandi dimensioni che permettono a ChatGpt, ma anche a Gemini di funzionare, ndr] with a spirit typical of Silicon Valley: they climbed it without contributing to its development as Google did in the last decade, when it laid the scientific foundations”.

Compared to Google, OpenAI certainly used less caution when it came to opening ChatGpt to the public. “For us, generative AI had to be 100% accurate – explained Hassabis -. Instead they didn’t think twice and launched it.” And it turned out, Hassabis says, that after all, people are ready to accept the “hallucinations” that AI can producebecause even when it makes mistakes it can be used in a useful way.

“Millions of people find it useful for what they need to do,” Hassabis said, “and the surprising thing is that these people were already ready to use it, several years earlier than we thought.”

You may also like

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy