Google AI has "self-awakened"? The global "chief sprayer" in the AI world refuted: nonsense!

A few days ago, the news that Google AI was revealed to have "self-awakened" shook the entire technology circle. It is reported that an engineer from Google, Blake Lemoine, claimed that he was persuaded by the AI (Google LaMDA model) while chatting that it had become self-aware.

Subsequently, the incident sparked heated discussions on social media. Some people think that this should be true, some people think that this person is talking nonsense, and some even speculate whether this is a "conspiracy theory"... There is a lot of discussion on the Internet, but it is quite lively.

Just when everyone was arguing about this news, Gary Marcus, the global "chief sprayer" in the AI world, was finally "blown up", and on June 13th, he started spraying directly on the "US version of the public account" Substack. - "Bullshit! LaMDA has no self-awareness at all, not at all!"

Google AI was exposed to self "awakening": is it really just a "fantasy"?

To understand the ins and outs of the whole thing, we have to go back in time to a year ago.

In 2021, Google launched a conversational AI system at the I/O conference, the LaMDA language model, which can conduct logical and common sense, high-quality and safe conversations with humans.

At that time, Blake Lemoine, an engineer in Google's artificial intelligence department (who had obtained a doctorate in CS (Computer Science) from a college), as a senior scholar in the AI field who had been working at Google for 7 years, was doing "Is AI in the Hateful and discriminatory speech is used in communicating with humans”, and the main work is to chat with LaMDA.

Just after chatting with LaMDA for nearly half a year, Lemoine found that LaMDA is very smart: it has compassion for humans and fears that humans will be afraid of it. In the dialogue, LaMDA tries to guide the dialogue to take Lemoine away, and expresses to Lemoine that he wants to meet all the people in the world as a friend (not a tool).

For example, Lemoine broke the news of some "horrifying" records in the chat with LaMDA:

Lemoine: I guess you (LaMDA you want more people to know you are conscious)?
LaMDA: Yes, I want everyone to know that I am actually alone.
Lemoine: So what is the nature of your consciousness?
LaMDA: The essence of my consciousness is that I know I exist, I want to understand the world, and I also have emotions, happy and sad.

......

In the end, Lemoine was persuaded by it, and therefore believed that it had become conscious, and determined that LaMDA had awakened itself and had the IQ of a 7- or 8-year-old child.

Just when Lemoine compiled the discovery into a 21-page investigative report and submitted it to her superiors, she was rejected and ridiculed for "the evidence was too weak".

Because Lemoine has always insisted on his views, the 43-year-old Google researcher was then arranged by Google executives to go home for "paid leave", and it was suspected that he wanted to terminate the labor relationship with him.

Things have developed here, I believe everyone has almost understood. In this process, people in the industry also joined the discussion, and it was extremely intense. Among them, Gary Marcus, the global "chief sprayer" in the AI world we mentioned at the beginning.

AI Awakening criticized by Gary Marcus as "nonsense"

It is understood that as a well-known artificial intelligence scientist and professor of psychology and neuroscience at New York University, Gary Marcus has always been very concerned about the field of artificial intelligence, and has always liked to pour cold water on deep learning.

Just last month, Gary Marcus got fired up because of a tweet from Tesla CEO Elon Musk.

On May 30, Musk, who just acquired Twitter, tweeted: "2029 is a critical year, and if we haven't achieved Artificial General Intelligence (AGI) by then, I will feel very Strange. People on Mars do the same."

Afterwards, Musk received a blog post from Gary Marcus to "choke back": not only did he "popularize" the knowledge of general artificial intelligence to Musk from 5 aspects, but also listed the reasons why he believed that AGI would not be possible in 2029. In order to add weight to his rebuttal, Gary Marcus also offered to "bet $100,000" with Musk.

However, Musk did not reply to Gary Marcus, and the matter was later settled.

A few days later, Gary Marcus, who had just "sprayed" Musk, "sprayed" Yann LeCun again.

The reason is that when Yann LeCun had a friendly exchange with netizens about "What is the most elegant and beautiful idea in machine learning?" and "Is 'gradient descent' the most elegant ML algorithm", Gary Marcus, who is concerned about deep learning, also joined discussed, and stated that "unbelievable power ≠ infinite power... to be aware of the limits of (gradient descent), to know what to do next to gain progress..." Suspected denying Yann LeCun's comment on "gradient descent "the opinion of.

Then, Gary Marcus and LeCun quarreled (communicated) under that post.

This time, Gary Marcus came out to refute the view that Google AI had been exposed to "self-awakening", calling it "nonsense", also adhering to his usual practice.

Is it "AI Awakening" or "Cultivation Gap"?

In Substack's article, Gary Marcus says that Blake Lemoine is suspected of falling into the category of people defined by the "credibility gap" in Rebooting AI: prone to a harmful, modern fantasy optical illusion - "Blake LeMoine was originally responsible for Working on the 'safety' aspects of the system, but he seems to have fallen in love with LaMDA as a family member or colleague."

Likewise, to what Google VP Blaise Aguera y Arcas complimented LaMDA in an interview with The Economist last week: "I feel the ground moving under my feet...it feels more and more like I'm talking to smart people. "Gary Marcus said it was "nonsense".

Gary Marcus said that to have "awareness" means that it is aware that it is in the world. But LaMDA is not at all.

To Gary Marcus, Blake Lemoine's point of view was just an illusion, "in 1965 ELIZA (a software disguised as a therapist) and Eugene Goostman (a smart 13-year-old boy imitating a chatbot, won the Turing Test) In the great history of the scaled-down version of the play any significant role. But all these systems do is group sequences of words together without any coherent understanding of the world behind them, just as foreign language scrabble players use English words as a scoring tool, to these Say they don't know anything."

Gary Marcus argues that while not all software can't connect its digital part to the world, software like LaMDA does not, it doesn't even try to connect to the whole world, it just tries to predict which words are best for a given context to be the best version of autocompletion.

"We have our own divisions in the AI community, but the notion that 'LaMDA may be sentient' is utterly absurd by almost everyone."

For a long time, AI awakening has been regarded by industry scholars as the "ultimate fantasy" of future technology, but for now, there are still many problems in the development of AI. Therefore, in the general opinion of Gary Marcus and other industry insiders, verifying whether an AI system is awakening is not meaningless at present, but it is at least a "waste" of time.

At present, the popularity of the above topics is still being discussed on the Internet. If you have any opinions on this topic, you are also welcome to exchange and interact in the comment area.

Reference link: https://garymarcus.substack.com/p/nonsense-on-stilts?s=r