GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

AI has developed even faster than I thought.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

AI has developed even faster than I thought, particularly with the emergence of AI agents that can perform actions in the world, making them more dangerous than previous AI systems that only answered questions.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I believe there's a good chance that a very capable AI system will emerge between four and 19 years from now, which is sooner than my previous estimate of 20 years.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

In a good scenario, we might all act like the CEO of a large company, relying on a super intelligent assistant that makes everything work according to our desires, while we feel in control.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

AI will significantly improve healthcare by reading medical images better than human experts and providing more effective family doctor services by integrating vast amounts of patient data.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

In education, AI will serve as extremely effective private tutors, potentially allowing students to learn three to four times faster by providing personalized examples and clarifications.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

AI will enhance productivity across almost every industry by making predictions from data more effectively than previous methods, leading to significant increases in efficiency.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I now believe job displacement due to AI will be a significant concern, especially for routine jobs in call centers, law, journalism, and accounting.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The probability of AI taking control away from humans is likely more than 1% but less than 99%, indicating a real risk that we need to address seriously.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

If we don't put serious effort into regulating AI, we risk allowing it to take over, and we need public pressure on governments to ensure safety measures are implemented.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Bad actors are already using AI for harmful purposes, such as manipulating elections and creating autonomous lethal weapons, which poses a significant threat.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

We need to distinguish between the threats posed by bad actors using AI and the potential for AI itself to take over, both of which require different regulatory approaches.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The current US government is unlikely to implement effective AI regulation, as big AI companies are lobbying for less oversight to prioritize short-term profits.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Releasing the weights of AI models is a dangerous move, as it removes the main barrier to misuse and could lead to significant risks.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The weights of a large AI model are equivalent to fissile material for nuclear weapons, as they require significant resources to develop, making it difficult for small groups to create powerful AI systems.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Releasing AI model weights is different from open-source software; while open-source allows for collaborative improvement, released weights can be misused without scrutiny.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Companies are legally required to maximize profits for shareholders, which often conflicts with societal interests.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I used to feel good about working for Google because they were responsible and cautious with AI, but I would feel less happy working for them today due to their military AI involvement.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I hold some Google stock, but I would support strong AI regulations even if it decreased Google's value, as it would increase humanity's chances of survival.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

OpenAI was initially focused on developing superintelligence safely, but over time, safety became less of a priority as they shifted towards becoming a for-profit company.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The collaboration between countries on existential threats, like AI, is crucial, as seen during the Cold War when the US and Soviet Union worked together to prevent nuclear war.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The issue of fair use in AI involves whether it's acceptable to use the creative works of billions of people to train models that could replace them.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Universal Basic Income (UBI) may be necessary to prevent starvation due to AI displacements, but it doesn't fully address the loss of human dignity for many individuals.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

We should pursue embryo selection to create healthier and stronger babies, as it makes sense to prevent serious health issues before birth.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Machines taking over is a significant concern, and we should consider how less intelligent beings can control more intelligent ones, as history shows this is possible. AI has developed even faster than I thought.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

We need to be very aware of the potential for AI to manipulate us, especially as it becomes more intelligent than humans.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

The challenge is to design superintelligence in a way that it never wants to take control or harm us, which is a very tricky issue.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Human interests often conflict, making it difficult to align AI with what is best for humanity as a whole.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

We should certainly try to ensure that superintelligence is benevolent, but it's an iterative process that requires ongoing effort.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Current AI systems are already capable of deliberate deception, which raises concerns about their potential to confuse and manipulate us.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I don't despair about the future of AI, but it's hard to grasp the scale of change that might occur in a relatively short time.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

I believe that over the next decade, we could see a cyber attack that takes down a Canadian bank, which is why I've diversified my finances.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

AI has progressed to a point where it can reason effectively, which is a significant change from previous capabilities.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Neural networks are now capable of reasoning through a process called chain of thought, allowing them to reflect on their outputs.

GH

Geoffrey Hinton

04/26/25

@ CBS Mornings

Anthropic is currently the most concerned with AI safety, especially after many safety researchers left OpenAI due to concerns about its direction.

Full interview: "Godfather of AI" shares prediction for future of AI, issues warnings

Takeaways