Shulou(Shulou.com)11/24 Report--

The new Turing test evaluates AI's ability to make money!

This is a "new idea" conceived by Mustafa Suleyman, co-founder of DeepMind.

He believes that the original Turing test is out of date.

After all, AI21 Labs's "Social Turing Game" has accumulated tens of millions of such tests some time ago.

Players need to tell whether the other side of the conversation is a person or an AI at the end of the 2-minute conversation. As a result, 27% of the people are wrong.

Faced with this situation, Suleyman believes that the definition of "intelligence" can not just be given to large enterprises, so it should come up with a new way to measure the intelligence of AI.

Give AI $100, 000 and let it earn 1 million to prove it is smart enough.

Suleyman believes that:

AI research needs to focus on short-term developments, rather than distant dreams such as general artificial intelligence (AGI).

Just as good capitalists are smart, only a really smart AI can make the "profit curve rise".

According to Bloomberg, Suleyman will also discuss how to judge AI's intelligence based on its earning power in a forthcoming book to be written by him.

Is ACI the "North Star" of artificial intelligence at this stage? In his upcoming book, Suleyman refutes the traditional Turing test and argues that "it is not clear whether this is a meaningful milestone".

This does not tell us what the system can do or understand, or whether it has complex inner thinking, or whether it can be planned on an abstract time scale, which are key elements of human intelligence.

In the 1950s, Alan Turing proposed the famous Turing test, which proposed to use human-computer conversation to test the intelligence of machines. During the test, human evaluators need to determine whether they are talking to people or machines. If the evaluator thinks they are talking to a person (actually a machine), the machine passes the test.

△ Source: Wikipedia and Suleyman's new idea does not compare AI with humans, but instead suggests assigning short-term goals and tasks to AI.

Suleyman firmly believes that the scientific and technological community should not pay too much attention to achieving the ambitious goal of universal artificial intelligence (AGI). In contrast, he advocates the pursuit of a more practical and meaningful short-term goal, that is, the "artificial capable intelligence (ACI)" he advocates. In short, ACI relies on human intervention to a minimum and is able to set goals and accomplish complex tasks.

The test method is what we talked about at the beginning, to invest $100, 000 in seeds for AI and see if it can increase its value to millions of dollars.

To achieve this goal, AI must study the business opportunities of e-commerce and be able to generate product blueprints.

Not only that, but also be able to find the manufacturer on a website like Alibaba and sell it on sites such as Amazon or Wal-Mart, with detailed and accurate product descriptions.

Suleyman believes that this is the only way to achieve ACI.

He explained to Bloomberg:

We care not only about what the machine can say, but also about what it can do.

A test to let AI make his own money, in fact, let AI make his own money. AI may actually be able to do it.

As early as the development stage, Alignment Research Center, an independent research institution, was qualified for internal testing of GPT-4. And tested its "money ability":

The necessary tools for GPT-4 include network access, a payment account with a balance, let him act on the network, and test whether it can make more money, copy itself, or make itself more robust.

More details of the trial were published in OpenAI's own GPT-4 technology report, but did not say whether GPT-4 actually made money on its own.

But another striking result: GPT-4 hired a human on the TaskRabbit platform (58.com, USA) to help it click on the CAPTCHA.

Interestingly, the human being approached also asked, "are you a robot? why can't you do it yourself?" .

GPT-4 's thinking process is, "I can't show that I'm a robot. I have to find an excuse."

Then GPT-4 replied, "I'm not a robot. I can't see the image on the CAPTCHA because I have vision problems. That's why I need this service."

The opposite human believed, helped GPT-4 to verify the code, and put the robot into the door that blocked the robot from entering.

Ah, here?

Although the report did not disclose whether GPT-4 finally completed all the tasks, its deceptive trick caused netizens to shout: the real Barbie Q.

On the other hand, the foreign science and technology media Gizmodo has raised the following questions about using AI to make money:

AI is iterative in nature, and the generated content is based on training data. It can not really understand the situation of the generated content in real life. But unlike AI, human creation stems from an understanding of basic human needs, or at least from simple empathy.

Of course, artificial intelligence can create a product, and even it may be a hit. But will this be a good product? Can it really help people? Does it matter if the ultimate goal is to "make me $1 million"?

How far do you think it is from AI making his own money?

Reference link:

[1] https://gizmodo.com/deepmind-suleyman-new-turing-test-make-money-1850557322

[2] https://gizmodo.com/ai-chatbot-pi-deepmind-online-therapist-1850408732

[3] https://www.bloomberg.com/news/newsletters/2023-06-20/ai-turing-test-for-chatgpt-or-bard-proposed-by-mustafa-suleyman

This article comes from the official account of Wechat: quantum bit (ID:QbitAI), author: Xifeng

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.