About three and a half years ago, we posted an article “Some Thoughts on How We can Experience and Learn from the Past Virtually” in which we raised the question “Whether we can make use of the information from the past, not only to relive past events, but also to make use of that information to learn from it.” The idea is whether we can make use of Artificial Intelligence” (AI) to learn from the information in the past and then creatively built on that to lead to new scientific discoveries. The current article provides some specific thoughts on that proposal.
Instead of debating for hours on creativity and whether computers can have the creativity to come up with outstanding discoveries or inventions, below we discuss some specific examples from physics which with the help of leading questions to AI computers could lead to significant scientific discoveries.
Some Specific Thoughts on Making Use of AI in Physics: The idea is to choose a good topic, compile the knowledge we currently know about this topic, including, e.g., who are the major thought leaders on that topic, what are the major thoughts of lead investigators on this topic, formulate some key questions to ask about that topic, and gather relevant data related to that topic. Then provide this information as input to AI-capable computers. Then utilize artificial intelligence to help us to investigate that topic. As part of asking the computer to do AI work, we could also ask the computer to carry out a virtual discussion or brainstorm with a group of scientific researchers on a difficult problem that they might all have researched on previously. (If we do that, then we may need to provide the computer with information about the background of such lead investigators, not only their technical background, but also information on the type of person they are, on their personality and their methods of thinking, etc.)
The idea is definitely not new. As a matter of fact, many people probably have thought about this idea, and many people probably have also dismissed the idea because they argue that a major discovery will require great intelligence and creativity that are beyond the reach of our current computers. This leads to the question of creativity, and whether computers can have that kind of creativity to come up with outstanding discoveries or inventions.
Examples of Possible Leads as Input to AI-Enabled Computers: We probably can discuss for hours on the definition of creativity and wouldn’t be able to come to agreement on its definition and whether computers can exhibit that. However, let’s not talk in abstract, and actually look at some of the discoveries in the last 50-100 years that were considered to be important discoveries. In particular, consider the field of high energy physics, or elementary particle physics. In my opinion, some of those discoveries could have come from computers with suitable questions or inputs from a knowledgeable researcher or a team of knowledgeable researchers, then with the help of AI-capable computer(s), some leads suggested to the computer could enable the computer and/or researcher(s) to make a new discovery. Here are a few examples:
- For the asymptotic freedom theory (leading to Quantum Chromodynamics or QCD, the current theory of strong interactions of quarks and gluons) of Yang-Mills gauge theory from the work of Politzer, and Gross and Wilczek in 1972-1973 that resulted in their 2004 Nobel Prize in Physics, it turned out that two-three years earlier Anthony Zee investigated several theories for this asymptotic freedom property. Unfortunately for him, one of the few theories that he didn’t investigate was Yang-Mills gauge theory. If he did, he probably would have discovered it. So if someone in 1970-1971 had fed this information to an AI-enabled computer and asked the computer the question what other theories they could have investigated for this property, the computer might have suggested Yang-Mills gauge theory for investigation and then the researcher would have discovered it.
- Even parity violation of Lee and Yang for their 1956 work with respect to weak interactions. If someone had fed the information to a smart computer that there were strong experimental data to support conservation of parity in strong and electromagnetic interactions, and had asked a smart computer to search for evidence of conservation of parity in weak interactions, the computer would have answered that there was not much evidence, and they could have proposed non-conservation of parity in weak interactions before Lee and Yang, which was what Lee and Yang did.
- Even on the question of the expansion of the universe originally discovered by Hubble in the 1920s (Hubble didn’t get the Nobel Prize in Physics because at that time astronomy was not considered part of physics) and the more recent accelerated expansion discovery of the universe by Perlmutter/Schmidt/Riess (Nobel Prize in Physics in 2011), a computer with the right inputs and the right questions could have discovered or led to discover that.
- Three-degree cosmic background radiation that got Penzias and Wilson of Bell Labs for their work in the mid 1960s leading to the 1978 Nobel Prize in Physics could have been discovered by a smart computer with the right questions and inputs, instead of the accidental discovery of Penzias and Wilson (at first, they didn’t even know what they discovered), even though at that time a group at Princeton was looking for that kind of astrological evidence), but they didn’t have smart computer with AI in the early-mid 1960s. This means that more groups might have looked into this area of research around the time of discovery of Penzias and Wilson.
I think if we work on it, we could come up with many other previous new ideas or discoveries not only in physics, but also in other fields, that could have been made or led researchers to by computers with AI, as long as appropriate questions and relevant data are input to the AI-enabled computers. Of course, this may be an iterative process, meaning there could be going back and forth with the computers before a meaningful new idea or discovery, or before leading to a new idea or discovery, will emerge.
This is not taking away any credit for the people who achieved these past achievements.
I also do agree that certain discoveries such as from Newton’s gravitation theory to Einstein’s General Theory of Relativity may require such a great leap of creativity that it is unlikely that currently anticipated AI-enabled computers could have come up with that discovery or invention, or saying it in another way, it is unlikely that the persons feeding the computers could have come up with the right questions and feed the computers with the appropriate inputs.
Other Important Topics in Physics: In the physics field, other examples of possible topics for AI-enabled computers to attack include:
- What is dark energy?
- What is dark matter?
- Why is there so much asymmetry between matter and antimatter?
- Deep paradoxes of Quantum Physics, e.g, collapse of the wavefunction and quantum entanglement
These are all important and well-known problems. The key is to figure out the next deeper level of questions and appropriate data to feed and interact with the computers. So it is not that we just assign a problem to the computers, like assigning it as a Ph.D. thesis topic to one or more graduate students, but we have to work closely with the computers, like often with the Ph.D.-seeking graduate students, and through what could be a long, difficult, and creative collaborative process before some meaningful results, or before some new leads, can come out. As far as the above list of topics, I think that we need to probe to one or two levels deeper to come up with suitable questions to ask the AI-enabled computers.
Closing Thoughts: Even though the idea discussed here is very simple, I think probing AI-enabled computers with field-specific knowledge-based questions and then working closely with AI-enabled computers has lots of potential, across all fields (whether it is science, engineering, biology, medical, economical, social, political, etc.). I am sure that more and more people will work in this area and make progress along this line of reasoning, especially as AI becomes more sophisticated and more creative.