Low-level Artificial Intelligence Parrots mimic, repeat, and imitate
High-level Artificial Intelligence Crows drink water, causally driven artificial intelligence
Be distinctive in your field, equivalent to using your expertise to "claim territory": People from different backgrounds have various advantages in the same situation. As someone from a scientific background, do not try to encroach on areas meant for those from a computer science background for two reasons: first, if you do so, you will need to spend more time than computer scientists learning redundant knowledge; second, there is a lack of individuals with a strong grasp of natural science in computer science. If you do not leverage your advantages to the fullest, it will be difficult to make an impactful contribution. Living is not just about existing; in today's context, living means doing something meaningful. If not, what difference does it make whether you exist or not?
Setting standards in your industry: This subheading seems very large and somewhat vague to me, but it should be the correct goal for my efforts; direction is more important than effort. Remember this: do not expect to publish articles just by tweaking others' code. It is unrealistic; this field is not easy, and I have already experienced that. You need to unify the problems in a field with a framework. For example: the Schrödinger equation.
Current state of artificial intelligence: In "non-foundational disciplines," there is no need for groundbreaking scientific discoveries; most of the time, it is sufficient to identify 'patterns' among variables in a certain field based on existing foundational scientific theories and available data to achieve scientific discovery. However, in some "foundational disciplines," scientific discovery is not merely about finding the "connections" between data and empirical materials; more often, it involves exploring and discovering new variables. Since artificial intelligence is currently better at finding "correlations," its depth of application in these areas is still insufficient. On the other hand, the extent to which scientific research can be automated and the role artificial intelligence can play in the process of scientific discovery still depends on human understanding of the nature of scientific research activities and scientific discovery.
Quote: "If I have been able to see further, it was only because I stood on the shoulders of giants." --Newton
Sources of machine learning data:
- Theoretical calculations can quickly generate large amounts of data. They are relatively "cheap and easy to obtain," but lack "real-world complexity."
- There is a wealth of data in past literature and databases. Data in databases is structured and relatively easy to extract, but data in literature is unstructured, and extracting it often requires specific domain knowledge. However, manual extraction is inefficient, and the time and economic costs are too high. Natural language processing technology can be used to extract data from literature.
- Data obtained from experiments is very rare and precious.
Collection of deep statements:
Zhu Songchun: This touches on a core issue, which is that we acknowledge and accept that the future society is big data or something that is not clearly defined. Moreover, this model must be this village, and that model must be that store; there is no unified statement, which is a relatively popular saying at present. However, I personally believe this is a significant misconception. The development of science seeks the simplest explanations. The current situation fundamentally arises from problems in research methods.
What does this mean? If intelligence is fitted as an objective phenomenon, its model is indeed very complex. We originally conducted a simple experiment to judge the boundary between physics and intelligence (vitality). For example, if two objects collide in a room, that is very simple; a few rounds of parameters can fit it. But if it involves two people, one running away and the other chasing, with various social relationships, the physical model (energy function, big data model) cannot solve it. Each segment of the intelligent agent's movement requires a different model to fit, which remains unclear.
However, if we assume a subjective value function instead of an energy function to describe it, it becomes very simple to clarify. So why do I later discuss "rational science" and "mind science"? In mind science, it is said, "The mind is reason, and there is nothing outside the mind." It holds that the complex phenomena of intelligence in the world are driven by simple values; find your values, and all your actions become uncomplicated. If I specifically try to replicate your behavior, that is like learning to walk in Handan; fitting can never be complete. Once I understand your value orientation, your positioning, and your framework, all your actions can be explained by very simple motivations.
Therefore, I believe that currently, in artificial intelligence, everyone is still using the original fitting method, trying to explain all human subjective behaviors with big data, which is not feasible.
- being existence
- becoming change
- believing belief
Wu Jiarui: For young scientists, I think breaking the deadlock involves three aspects. First, it relates to theory; most life scientists work under reductionism, and we now need to approach it from a systems perspective. Otherwise, research will become increasingly detailed, overly focusing on trivial matters. From the perspective of research theory and thinking, merely looking at details is insufficient; a systematic viewpoint is necessary.
Second, regarding research methods, life sciences are now, in a sense, technology-driven, and there is a trend where research content pursues traffic and technology. I suggest advocating more for thinking and theory.
Third, the phenomenon of utilitarianism is particularly severe now; we need to promote the spirit of science. For example, Japanese figure skater Yuzuru Hanyu insists on attempting a quadruple axel; he does not pursue gold medals but rather seeks to surpass human limits. We scientists should not pursue gold, silver, or bronze medals but aim to go higher, faster, and further; the Olympic spirit should be extended to scientific research. Foucault mentioned in his autobiography that if science cannot lead us knowledgeable individuals astray, then what value does science have?