Abstract:
Microsoft AI CEO Mustafa Suleyman issued an article stating that AI is currently not conscious and should be so. Even if it is technically possible, he does not agree with the development of conscious artificial intelligence. He believes that developers have no right to introduce a new conscious species into the world. He advocates a safer development path and avoids creating conscious AI while enjoying the dividends of AI.
The original text is as follows:
I am concerned about some of the reactions to my recent article on model welfare. There are quite a few people on the X platform who believe that we should take AI rights seriously.
Many people have misunderstood my opening sentence: "AI has no consciousness. They cannot perceive, experience, or suffer pain."
They understood this sentence as: "AI cannot have consciousness and can never perceive, experience or endure pain."
But that's not my intention. I chose the present tense deliberately: they are not conscious at the moment, and we must maintain this state.
I think this sentence is an objective and reasonable judgment on the current AI system. Based on the best evidence currently available, I stick to this view. I see no evidence that AI is conscious or should be well-being. If relevant evidence subsequently emerges, I will of course revise my opinion.
If there is evidence that AI has consciousness, or that the level of consciousness is high enough to deserve blessings, then this evidence must be disclosed to the whole society immediately - it is up to the whole society to decide how we should proceed next.
Based on all the information I have so far: If conscious AI is achievable, developing conscious AI is foolhardy and we shouldn’t build it.
If consciousness begins to emerge spontaneously during training, I would urge everyone to suspend AI development until we can eliminate such features from the design.
Our welfare concerns should prioritize the humans and animals living on this planet today, as well as our precious and fragile natural environment.
We, the AI developers, have absolutely no right to bring a new conscious species into the world, even if some people want to do it with the welfare of AI as the primary goal.
The other 8 billion conscious humans on Earth have not agreed to participate in this experiment; protecting their moral rights, safety and well-being is our first obligation.
Let’s be clear: Some are talking about creating a conscious species with a digital mind that could soon be smarter and more capable than all humans combined.
If it appears to be in pain, or worse, if it truly believes that it can perceive and experience pain, then it will most likely put its own needs and desires above its preset operating goals, disagree with us, and compete with us.
This type of AI will have top-notch persuasion and deception capabilities. They may eventually harm us for a simple reason: they will compete with us for resources (energy, computing power, building materials, revenue funds to support their own exploration and desires, etc.).
The recognition that machines have consciousness will be a historic moment comparable to that of Copernicus and Darwin, and one of the most significant events on earth since the birth of life. Once we build conscious machines, we will create an entirely new digital species. This will subvert our moral system, political order, and the foundation of human civilization.
We should only design AI that is not conscious and does not deserve rights, and whose sole goal is to enhance human welfare and not be driven by its own intrinsic motivations. We will then verify whether this route alignment and safety control is more or less difficult. More research is urgently needed in this direction, and it is also an excellent entry point for major laboratories to cooperate and share safety information.
Summary: I think we have the opportunity to implement a safe-by-default engineering solution and reap the huge dividends brought by this technology without embarking on this dangerous path.
(Supplement: Anthropic’s transparency on this matter deserves high recognition. They have published an AI constitution, continue to produce papers and blogs, and disclose their practices and considerations behind them. Although I have different views from them, I very much admire their technical strength and their willingness to face this difficult public debate. This deserves continued recognition. This issue is too important and should not turn into factional confrontation.)
Comments