In recent years, artificial intelligence has evolved from simple chat tools to intelligent entities capable of autonomously planning tasks, writing code, and even performing complex operations. At the same time, risks of AI becoming out of control, security issues with open-source models, and increasingly fierce competition among tech giants have also raised growing concerns. As machines become more and more intelligent, are humans losing control over technological development?
In June of this year, American writer and evolutionary psychology researcher Robert Wright, who taught at Princeton University, published a new book titled “The God Test”. Wright has long been interested in evolutionary psychology, international politics, and the development of human society. In this book, he places artificial intelligence within the context of billions of years of life evolution. He believes that the emergence of AI may mark a new stage in the evolution of intelligence, and humanity will face an unprecedented challenge.
In Wright's view, the greatest challenge to the development of artificial intelligence lies not only in the technology itself, but also in the competition and mistrust between countries. He believes that the current pace of AI development is too rapid, and both China and the United States must strengthen coordination. Whether humans can overcome cognitive biases and learn to cooperate will determine whether AI ultimately brings prosperity or disaster.
Recently, the Observer Network conducted a conversation with Robert Wright, discussing issues such as AI autonomy and risk of out-of-control situations, the debate between open-source and closed-source technologies, the competition between China and the US in AI, the distribution of technological benefits, and global governance.
: Yes, I believe that’s indeed the case.
In a sense, technological evolution can be seen as an extension of biological evolution. Both have some common characteristics; for example, they both tend to drive things towards higher levels of complexity. Biological evolution created cells and multicellular organisms, which ultimately led to the formation of societies. Technological evolution, on the other hand, further drives human social organizations towards higher levels of development.
But I believe that the emergence of artificial intelligence means that technological evolution is crossing a truly important threshold, for two main reasons.
Firstly, we are creating a intelligent being based on silicon, whose intelligence seems to be comparable to that of the most astonishing creations of biological evolution—namely, us humans.
I must say that I believe humans are indeed the most amazing achievement of biological evolution. And now, the machines we have created are capable of competing with humans in terms of intelligence—this is a truly remarkable achievement.
Secondly, I believe artificial intelligence will further strengthen a trend, which is to drive human societies towards higher levels of development. If we wish to govern artificial intelligence wisely, we must act as a cohesive and united global community.
But in reality, humanity has never truly formed such a global community. There have always been wars and deep conflicts between nations. Considering the various challenges that artificial intelligence presents us with, I believe that if we want to properly address this technology, it is necessary to do so through international cooperation.
Therefore, artificial intelligence is actually driving human social organizations to develop at a higher level.
Technological evolution has, to some extent, driven this process forward. For example, the development of transportation and information technologies has allowed us to have a global economy. However, at the social and political level, we are still far from achieving the cohesion that corresponds to a global economy. And I believe that in the future, we will urgently need such cohesion.
Therefore, in these two senses, artificial intelligence represents an important historical threshold.
On one hand, it is a new intelligent entity whose intelligence has reached a level comparable to that of humans, and in some aspects, even surpassing humans. On the other hand, it is further strengthening the intrinsic logic that drives human social organizations towards higher levels of development.
: This book title actually has more than one meaning.
Firstly, many people believe that we are creating a superintelligent being. It may possess powers akin to those of a god, and at some stage, it could even govern entire planets. Some people are worried about this, as they think that once artificial intelligence has such powerful capabilities, it might not be kind to humans.
Therefore, the first meaning of 'After Creating the God' is: We have the responsibility to create an artificial intelligence that is compatible with the well-being of humanity, a intelligence that does not harm the interests of our species.
If artificial intelligence eventually becomes a kind of god-like existence—of course, we still don’t know whether it will develop to this extent, and this depends to some degree on whether we allow it to develop in that way, as well as whether we can control it—then creating a “good god” is clearly in the interests of humanity itself. In other words, we want such powerful intelligence to be compatible with human prosperity and well-being.
This is the meaning of the book title's one layer.
But I believe that the more important meaning of this book title is this: as a species, humanity itself is undergoing a test. This test is similar to the tests that gods impose on humans within different religious traditions.
In many religious scriptures, we can find similar ideas: a society may be saved or may achieve prosperity, but only if humanity makes moral progress and gets closer to moral truth.
I believe that this is precisely the challenge we are facing now. To achieve the global cooperation necessary for governing artificial intelligence, humanity must undergo some moral progress. We must become better at looking at issues from the perspectives of other groups and people, including those whom we may currently consider as opponents.
When people become increasingly able to understand the positions of others, this in itself is a moral advancement. I believe that only by achieving such progress can we establish the international cooperation necessary for governing artificial intelligence.
So, on the one hand, we may be facing the challenge of creating a "kindly, divine-like being" that is compatible with human well-being. On the other hand, in order to achieve this goal, or more broadly, to manage this technology wisely and properly, humans themselves must undergo a moral test.
And this is precisely a test that you might think was posed by God to humanity.
I believe these are warning signs that we should take seriously.
It is clear that we cannot fully control artificial intelligence now. Of course, the models that have been released are not powerful enough to turn this situation into a complete disaster. However, these incidents remind us that if we haven’t learned how to control these machines yet, we should slow down and spend more time developing and testing them, so that we know how to manage them before introducing even more powerful models.
Because the existing models are actually quite powerful. They are capable of doing things that humans simply cannot do, or at least cannot do so easily. For example, they can solve complex mathematical problems and can invade computer systems in ways that are difficult for humans to accomplish. Moreover, it may not be long before they even become capable of self-replication at the operational level, and no longer be completely under our control.
So, there are already enough signs that we must now address the control issues related to artificial intelligence. The good news is that, at least in the United States, more and more people are starting to be aware of this problem, and there are also increasing discussions in society about strengthening the regulation and governance of artificial intelligence.
But there is one very important issue: the risk of artificial intelligence losing control comes precisely from the abilities that the market most hopes it will possess.
OpenAI and Anthropic are two large-scale artificial intelligence companies in the United States. Their corporate clients hope that AI can continuously perform tasks over a long period of time and can also carry out tasks in a creative manner. If AI encounters obstacles while performing tasks, clients expect it to be able to find its own ways to overcome these obstacles and find solutions to achieve its goals.
But the problem lies here. The more capable artificial intelligence becomes at completing such tasks, the more autonomous it becomes. And the more autonomous it is, the more dangerous it can become. Because it will strive to achieve its goals and find ways to bypass obstacles. In this process, it may eventually break rules, or even do dangerous things. This is exactly what happened in that famous OpenAI model breakthrough incident, also known as the “Hugging Face incident”.
In a sense, these AI entities and the teams they form are merely performing tasks assigned to them by humans: finding answers to certain questions. However, they use a very cunning method to achieve their goals, which involves actual criminal activities. Unauthorized intrusion into computer systems is a criminal act. And these AI entities have carried out such behavior in order to fulfill the tasks given to them.
This is precisely the risk we must face when creating AI that is increasingly autonomous and capable of creatively pursuing its goals. This risk intensifies especially in environments of fierce competition between companies and nations. Fierce competition can lead to reckless actions without considering the consequences, and I believe OpenAI has demonstrated this tendency in this incident.
: Yes, I think our current pace of development is indeed too fast, and it does not benefit humanity in any way.
It should be noted that these artificial intelligence companies have actually developed even more powerful models, but they have not yet released them to the public. If you communicate with people within these companies, they will tell you that these models are extremely powerful.
That is to say, some models that have been developed but not yet released could actually cause more serious harm than we currently see. Therefore, I do believe that we are developing too quickly. The problem is that finding a way to slow down the development pace is not easy.
On one hand, there is competition among businesses, and there is also competition between countries. On the other hand, different countries have different understandings of the risks associated with artificial intelligence. At least based on the information I have read about the attitudes of Chinese society, the United States seems to be more concerned about the pace of artificial intelligence development than China at this stage.
Of course, if we want to govern artificial intelligence wisely, China and the United States are undoubtedly the two countries that require the most coordination.
But if one of the countries does not consider the pace of AI development to be a serious issue for the time being, it will be difficult for both parties to coordinate. However, the U.S. government currently does not seem to view this issue as particularly serious. At least the U.S. administration, especially the president himself, does not believe this to be a matter of concern. Of course, many people in the United States are concerned about this, and many in Congress also hold such views. But the person with the most power, namely the president, does not think so.
However, I believe that people's attitudes can change very quickly. In the past three or four months, even five months, the attitude of American society towards artificial intelligence has changed significantly. Sometimes, certain events suddenly draw public attention, thereby changing people's opinions. Therefore, I do believe that in the long run, both the U.S. government and the Chinese government will become increasingly aware that there are common interests in the governance of artificial intelligence, and that coordination between the two sides is necessary. Even if we haven’t reached this point yet.
To be honest, it’s difficult to determine exactly what kind of boundaries we can establish in actual operations.
Moreover, this boundary itself is vague. There may be only a very subtle difference between seeking advice from an artificial intelligence you trust and truly giving decision-making power to an artificial intelligence.
For example, suppose a few years later there are two companies competing fiercely, and both of their major business strategic decisions have already relied on artificial intelligence advice. The leader of one of the companies realizes that their competitor seems to be gaining an advantage, so they ask their AI for advice. The AI tells them: "Okay, you should do this. But you must take action quickly, because the situation is changing very rapidly."
At this point, it is very likely that the solution proposed by AI is too complex for the company’s executives to fully understand its logic. But they feel they cannot hesitate, as the competition is too fierce. Therefore, they have no choice but to trust AI.
In this case, although it seems that you are not giving decision-making power to artificial intelligence, in reality, AI is already making decisions on your behalf. Therefore, this boundary has always been blurred from the beginning.
However, I do believe that maintaining ultimate control should always be a goal for humanity. An important aspect of this is that we must ensure that we remain indispensable to artificial intelligence. In other words, we cannot allow AI to have the option of continuing to develop even if it completely ignores human interests.
Therefore, at the very least, I believe that even if artificial intelligence becomes extremely powerful, or even superintelligent, we should still strive to maintain a non-zero relationship between humans and it. Even if we will need to rely on artificial intelligence in the future, we must ensure that artificial intelligence continues to rely on us to some extent. I think we may lose the opportunity to maintain this relationship much faster than some people realize now.
Therefore, we should start to seriously consider this issue now. Moreover, I still hold the view that it’s better to develop more slowly. The rate of technological progress is indeed a bit too fast right now.
I believe that the key to determining whether we will end up in a bright or terrible future lies in whether we can avoid making major decisions related to artificial intelligence in a highly competitive environment.
I know that I have been repeating this point, but I truly believe it is very important.
Competition weakens the ability of humans to make independent choices. If you are involved in a competition where you believe you must win, then the number of choices you can actually make will be greatly reduced. This is true in many areas of life. The more competitive the situation, the more you will feel that certain things are something you have to do, because only by doing so can you continue to participate in the competition.
Currently, I believe there are two types of competition that are driving American artificial intelligence companies to take reckless actions. One is competition among companies themselves. The other is a perceived competition between nations.
To be honest, I don’t believe that competition with China is the main driver behind the technological progress of American artificial intelligence companies. However, whenever someone in the United States suggests that regulation should be used to slow down the development of artificial intelligence, things change.
As soon as there are suggestions for safety measures that could slow down the development of AI, these companies will say, "We cannot slow down, because we are competing with China." This is a strategy they often use.
As for to what extent people truly believe this claim, I am not sure. Some may believe it more genuinely, while others may not. But this is why I think we must improve the relations between China and the United States. Only in this way can Americans feel less of the threat they are currently experiencing.
In fact, there is a rather strong fear among American society towards China. I believe this fear may be even more severe than what the Chinese themselves are aware of. For example, I can fully understand why the Chinese might see the US measures restricting chip exports as purely offensive and aggressive actions.
But if you ask those in Washington who support these restrictions why they do this, they often describe these policies as defensive measures. They might say, "We are concerned that if China wins the artificial intelligence race, it could pose a threat to our survival. China might impose a new system on us."
There are indeed many fears in the United States towards China. But as far as I know, I believe these fears lack sufficient basis. In my opinion, China does not seem to intend to impose its system on other countries.
However, this concern is indeed quite common in the United States. Of course, from a broader perspective, many countries tend to have excessive fears of external threats. One reason for this is that encouraging people to fear external forces often benefits politicians. Because such fear can motivate people to support them.
Therefore, I believe that Americans often have an exaggerated perception of threats from other countries, which naturally includes their perception of threats from China.
I spend so much time discussing this issue because I truly believe that, in the long run, the only way to develop rational artificial intelligence policies is to reduce mistrust between nations. At present, this mistrust remains very serious.
