GEOFF BENNETT: This week has been a week in which concerns over the rapidly expanding use of artificial intelligence have reverberated in Washington and around the world.
Vice President Kamala Harris met with executives from leading AI companies yesterday.
Development, Microsoft, Google, OpenAI, and Anthropic.
The vice president discussed some of the growing risks and told companies that they had a “moral obligation” to develop AI.
safely.
The conference comes just days after one of the leading voices in AI, Dr. Geoffrey Hinton, announced he was leaving Google over concerns about the future of AI.
And that could lead to it not being checked in the end.
I plan to hear some of these concerns with Dr. Geoffrey Hinton, who is joining us from London.
Thank you for using our service.
And what can you express freely about artificial intelligence that you couldn’t express freely when you worked at Google?
GEOFFREY HINTON, ARTIFICIAL INTELLIGENCE PIONEER: It wasn’t that I couldn’t express myself freely when I was hired by Google.
That is, working for a company inevitably leads to self-censorship.
You tend to think about the impact it will have on the company.
I want to be able to talk about the risks of superintelligent AI that I now recognize
Don’t worry about the impact on Google.
Jeff Bennett: What are those risks as you see them?
Jeffrey Hinton: There are quite a few different risks.
With so much fake news at risk, we no longer know what is true.
By making people click on things that make them resent, you run the risk of fostering polarization.
You run the risk of putting people out of work.
In other words, making things more productive and significantly more productive should help everyone.
But there are concerns that it might only help the rich.
And there are risks that I want to talk about.
Many others have spoken of other risks, such as the risk of stigma and discrimination.
I would like to talk about another risk. That is the risk of superintelligent AI.
take control away from people.
GEOFF BENNETT: So how does human or biological intelligence compare to machine intelligence?
Jeffrey Hinton: That’s a very good question.
And I have a rather long answer.
Biological intelligence has evolved to consume very little power, so we only use 30 watts.
There are also a huge number of connections, such as 100 trillion connections between neurons.
And learning consists of changing the strength of those connections.
The digital intelligence we’ve been creating uses as many megawatts of power during training.
We have far fewer connections, just a trillion, but we can learn far more than we know. This suggests a better learning algorithm than what the brain has.
GEOFF BENNETT: AI smarter than humans
the system?
What – what are you worried about?
Jeffrey Hinton: So the question is, what motivates them?
Because they can easily manipulate us if they want to.
Imagine you and a 2 year old.
You can ask it, do you want peas or cauliflower?
Well, 2-year-olds don’t realize that they don’t really have to have either.
For example, we know that just by manipulating people, you can break into a building in Washington without going there yourself.
But imagine something better at manipulating people than any current politician.
GEOFF BENNETT: So why is AI
do you want it?
Doesn’t it require some sense?
Jeffrey Hinton: Let’s not confuse it with sensory issues.
I have a lot to say about sensations.
But I don’t want to confuse the issue with that.
Let me give you an example of why I want to do that.
So let’s say you have an AI
To do something, you give it a goal.
It also provides the ability to create subgoals.
For example, if you want to go to the airport, create a subgoal to get a taxi or something to get to the airport.
You will soon find out that there are subgoals. If you can achieve that, it will be easier for you to achieve all the other goals people give you.
And the sub-goals that make it easier are gaining more control and gaining more power.
The more power you have, the easier it is to get things done.
So there’s the coordination worry of giving a perfectly reasonable goal and deciding what I’m going to get in order to achieve it.
It’s much smarter than us, and it’s been trained from everything people have ever done, so it’s read every novel ever, read Machiavellian, and knows a lot about how to manipulate people. increase. Manipulating us to give us more power, we may have no idea what’s going on.
GEOFF BENNETT: When you were at the forefront of this technology decades ago, what did you think it could do?
What applications did you have in mind at the time?
GEOFFREY HINTON: There are many very good applications. So it would be a big mistake to stop developing such applications.
It will be of great help in medicine.
For example, would you go to a family doctor who has seen thousands of patients or a family doctor who has seen hundreds of millions of patients, including those with the same rare disease?
This way you can make a better doctor.
Eric Topol has been talking about it recently.
We can improve the nanotechnology of solar panels.
You can predict floods.
We can predict earthquakes.
You can do great things with this.
GEOFF BENNETT: So is the problem the technology or the people behind it?
Jeffrey Hinton: It’s a combination.
Clearly, many of the organizations developing this technology are in the defense sector.
Also, the defense department doesn’t necessarily have to be built in or be nice to people as the first rule.
Some defense departments want to include and kill certain types of people.
So you can’t expect them all to have good intentions for everyone.
GEOFF BENNETT: There is the question of what to do with it.
This technology is advancing faster than governments and societies can keep up with the capabilities of this technology.
That is, they leap every few months.
What it takes to draft legislation, pass legislation, and develop international treaties takes years.
Jeffrey Hinton: Yes.
That is, I openly tried to encourage more resources and more creative scientists to enter the field.
Taking over by machines is a threat to everyone, so I think this is an area where we can actually do international cooperation.
It is a threat to China, to the Americans, and to the Europeans, just as a global nuclear war did.
And for a global nuclear war, people actually worked together to reduce the chances of that happening.
GEOFF BENNETT: AI HAS OTHER EXPERTS
I would say that the concerns you are raising, this dystopian future, distract from the very real and immediate risks posed by artificial intelligence. Some of them are the misinformation, fraud and discrimination you mentioned.
How do you respond to that criticism?
GEOFFREY HINTON: Yes, I don’t want to be distracted by them.
Those are very important concerns and I think we should address them as well.
I just want to add this other existential threat that it takes over.
One of the reasons I want to do it is because I think it’s an area where we can get international cooperation.
Jeff Bennett: Is there a way back?
If you say that the age of AI is coming
is smarter than us, but will anyone come back from it?
Jeffrey Hinton: I don’t know.
We are entering a period of great uncertainty, dealing with the kinds of things we have never dealt with before.
It was as if aliens had landed, but they spoke good English, so I didn’t really take it.
GEOFF BENNETT: So how should we think differently about artificial intelligence?
GEOFFREY HINTON: We have to realize that things will soon be smarter than we are.
And they will be great.
They can very easily do all sorts of things that we would find very difficult.
So these things have huge positive potential.
But, of course, there is also the possibility of a big minus.
And I think we need to put more or less equal resources into AI development
Understand how to make it much more powerful and keep it under control, and how to minimize its bad side effects.
Jeff Bennett: Dr. Geoffrey Hinton, thank you for your time and your input.
Jeffrey Hinton: Thank you for inviting me.
