11/03/22: Eric Jonas

[GabeNicholson]

Comment below with a well-developed question or comment about the reading for this week's workshop!

If you would really like to ask your question in person, please place two exclamation points before your question to signal that you really want to ask it.

Please post your question by Tuesday 11:59 PM, We will also ask you all to upvote questions that you think were particularly good. There may be prizes for top question askers.

[lbitsiko]

Dear Prof. Jonas,

Many thanks in advance for presenting your work to our workshop. To be honest my knowledge of neuroscience is extremely limited. Nonetheless, I found your discussion of limitations in neuroscience research utterly fascinating.

Inspired by your paper, I would be more concerned about your views on where the field is directed research-wise. Would you suggest that neuroscience is experiencing a severe crisis? To paraphrase Gramsci in the context of your discussion, would you think that this crisis is based on the fact that the "old world is dying, and the new world struggles to be born" hence living a time of neuroscience-model "mosters"? Or would it be more accurate to say that the discipline is under a change of paradigm in the Kuhnian sense as suggested in the PubPeer discussion? Maybe simplifying the question, would you view optimistically or pessimistically where neuroscience research is going?

In addition, I would be grateful for the following clarification: Is your study of a microprocessor directed by the assumption that the brain is a computational machine?

Many thanks and looking forward to your talk this Thursday.

Kind regards, Loizos

[shaangao]

Thank you for sharing your work with us, Prof Jonas. The paper argues that current methods in neuroscience are insufficient for really uncovering how the brain works by applying these methods to microprocessors and showing, analogously, that these methods are insufficient for us to understand how microprocessors work. It seems to me that the neuroscience research covered in this paper is mainly devoted to Marr’s implementation level of analysis, while the paper advocates for higher levels of analysis, i.e., algorithmic and computational levels. It is, undoubtedly, vitally important to go beyond how neurons connect and sum, and understand how the brain works algorithmically and computationally. And neuroscience research does study those aspects as well, oftentimes through looking into the patterns on the implementation level using the methods referred to in this paper; for example, using ISC to study how we comprehend schematic events. While there is research addressing questions within each level of analysis, a more difficult question, however, is how to connect the three levels of analysis, such as how the universal "algorithm" of using event schema emerges from networks of neurons, or even the good old philosophical question of where qualia comes from. Many of these challenges seem to be unique to neuroscience compared to computer science, and I would be curious to hear what you think computer science can offer in terms of helping neuroscience solving these difficult questions Thank you.

[Yuxin-Ji]

Dear Prof. Jonas,

Thank you for sharing your work with us. While I have limited experience with neuroscience, I found the issue you addressed in the paper of great relevance to many other computational-related fields. Particularly, you suggest in the paper that the naive data analytics and algorithmic approach could not provide a satisfying interpretation of the underlying mechanisms, which made me think of the BlackBox problem computer scientists encounter for machine/deep learning tasks. In natural language processing, for instance, though the machines could sound almost like a human under certain contexts, they do not truly understand the language they produced. I wonder would you say your concern for the computational approaches in neuroscience could be resolved when there is an advance in algorithms or if there is need for adjustments in neuroscience methods even with suitable algorithms?

Thank you in advance.

[y8script]

Dear Prof. Jonas,

Thanks for sharing your work with us! As we usually take the analogy from the brain to microprocessors, it is really cool to see how these neuroscience methods actually work on electrical systems. I have some questions about the extent to which we can transform the validity judgment from microprocessors back to the brain:

1. Given the failure of many neuroscience analytic methods to reveal the essential mechanisms of microprocessors when showing similar results to the actual brain, how do we account for the seeming success of these methods in certain neural systems, for example, part of the visual system? As neuroscientists do find some converging understanding of the brain with several different methods and in different scales, does it mean that these methods are more suitable for neural systems than for electrical systems to imply real mechanisms? Or does it mean that our current understanding of the brain may still be trivial and fundamentally different from the real mechanisms?

2. When picturing the optimal methods and approaches for computational neuroscience, do you think they should necessarily capture the commonalities of neural and electrical systems, and thus work for both systems? If a method doesn't pass the sanity check of the microprocessor, is it possible that it doesn't capture the commonalities of the two systems but captures the specialties of the neural systems? If so, is it reasonable to place doubt on a method even if it doesn't pass the microprocessor sanity check?

[Dededon]

Dear Professor Jonas, Thank you for your sharing! So here are my questions: As a person from computer science background, I wonder how the neuroscientists view the design structure of the current computer system, for instance, the von Neumann structure? Are there biological references of the von Neumann structure, or caching that are fundamental to contemporary computer architecture? Can neuroscience provides alternatives to improve the computer architecture? Thank you for your notice!

[GuangjieXu]

Nice to meet you, Pro.Jonas. Do you think it will cause some ethic problem or not?

[GuangjieXu]

Nice to meet you, Pro.Jonas. Do you think it will cause some ethic problem or not?

[zihua-uc]

Dear Prof. Jonas, I think using man-made complex systems such as the microprocessor is a really great idea to validate current analysis methods in neuroscience!

However, I am concerned with overfitting. Using too many man-made systems to validate neuroscience methods could eventually select for methods that can explain man-made systems well, but the human brain comes from years of evolution in nature, which is beyond man's understanding for now.

Do you have any suggestions to balance the trade-off between validation of known systems and learning the unknown?

[Hongkai040]

Hi Professor Jonas, Thank you for sharing your work! I think the idea you proposed in the paper is really cool! (And perhaps it is one of the coolest I've ever heard of this year) I always think of the relationship between biological systems(our minds) and computer systems. Usually, we are thinking of borrowing something from the neuroscience domain. So, it's fascinating to see a paper discussing the reversed version.

My question is about the divergence between a brain and a microprocessor. Actually, I don't much about either of them, so my question may be stupid. Based on my shallow experience with large deep learning models, digital circuits, and signal processing, I think those silicon-based processors and algorithms are designed in a simply complex way: they could have billions of units, and parameters, but they're usually organized systematically. They're incomparable in computing flops but immature in things like fluid intelligence or cognitive ability. However, on the other hand, I feel like(maybe I am wrong), our brains have some randomness and chaos inside. They're not efficient(but energy efficient!), but are somewhat a generalized model that can quickly learn and do various tasks. So, I wonder what are the caveats could be in the design of testing neuroscience tools using microprocessors?

[borlasekn]

Prof. Jonas, Thank you for sharing your work with us. The implications of your discussion on neuroscientists being limited to the approach they are taking was interesting to me. This made me think about how easy it is for us to get stuck in loops of their own methods and theories that then bias our own research and can limit our view of different issues. Often, taking new views for problems requires deconstructing one's basis of research and theory. What are your thoughts for how we can encourage researchers to rethink issues from new perspectives?

[iefis]

Hi Prof Jonas, Many thanks for sharing your work with us! We have been discussing complex systems and complexity in our perspectives class. The complexity we have touched upon is largely viewed from the lens of sociological inquiry, which emphasizes the interaction between actors situated in networks. I am not very familar with neuroscience but it seems to me that the complexity of the brain is induced more from its innate hierarchical structure, which we have not been able to thoroughly understand. I am wondering how would you define/characterize the complexity of the brain and what are some challenges that impede us from understanding its hierarchy of imformation processing?
Thank you in advance.

[erweinstein]

Hi Professor Jonas!

I would like to note that (like several of my fellow students) my knowledge of neuroscience is limited, in my case to what was part of a Intro to Psychology course and plus I've learned secondhand from friends who are MD (psychiatry) or PhD (psychology) grad students/residents/post-docs, as well as some "pop-sci" books/articles.

Since methods for causal inference (and adjacent concepts) are a very important part of my own work, I particularly appreciated the section on lesion methods and your point about how what we learn about them is often superficial or spurious. As you say, there is no such thing as a "Donkey Kong transistor" vs. a "Space Invaders transistor".

First, are you and Professor Kording intending to make a broader point here about the danger of breaking something/lesioning something as a tool for causal understanding (not just one that applies to neuroscience/brain research)? I think you are correct that this process is often followed (and expected to yield insight) without sufficient "sanity checking"....

Second, speaking of popular science, this sounds like exactly the problem that genomics researchers often complain about, (to paraphrase Kevin Mitchell, among many others): Journalists and popular writers love stories about "Scientists discover gene for XYZ”. Mitchell's complaint (from 2011) sounds very similar to your (and Lazebnik's) reductio ad absurdum based on microprocessors and radios: "...'genes for autism'? That phrase really makes no sense – the function of these genes is certainly not to cause autism, nor is it to prevent autism. The real link between these genes and autism is extremely indirect." Would you agree with that parallel? (For all I know, the existence that longstanding issue in genomics or other biology sub-fields may have been an input into your work; I'm genuinely curious, not trying to argue that you may have accidentally stolen their meme. :)

[yjhuang99]

Hi Prof. Jonas, it is really our great pleasure to have you join our workshop. While I am totally unfamiliar with neuroscience, I still find your paper interesting that researchers tend to reach a conclusion about certain data with a quite limited number of analytical methods, and somehow convince themselves that the results are indeed revealing the truth. On the other hand, what would then be your suggestion to produce a meaningful understanding of neural systems (or, any subject of interest)? Thank you!

[AlexBWilliamson]

Dear Dr. Jonas

Thank you for sharing your research with us! I just have one question. In your research you talked about a few of the changes to your experiment that could have helped our hypothetical neuroscientists to understand the microprocessor. Specifically, you mentioned trying to better control for other influences while running a test, data analysis structures that allow for hierarchical structures. I don't know much about neuroscience, so I was wondering how likely it is that these specific examples of changes would help with understanding the brain? Or were these answers just a continuation of your example of the microprocessor?

[sdbaier]

Professor Jonas,

Not just neuroscience, but much of social scientific research is stifled by the difficulty in evaluating weather a conclusion is correct and extends beyond the sample data. While I am by no means extensively steeped in the neuroscience literature, how do you think about the expansion of your findings to adjacent disciplines?

For example, you mention it being “[…] extremely difficult or technically impossible to produce behaviors that only require a single aspect of the brain.” in the section on Lesion a single transistor at a time, and that “[e]ven if brain areas are grouped by function, examining the individual units within may not allow for conclusive insight into the nature of computation.” in the section on Analyzing tuning properties of individual transistors. If we abstract away from brains as modular, functionally-partitioned, and complex systems to similar systems, would the conclusion that current analytical approaches are inherently limited in producing meaningful insights, independent of the amount of data, (at least tentatively) still hold? Arguably (and this is coming from someone without much familiarity of computational neuroscience), the core argument of the article is of relevance for different social systems with similar characteristics – stratified societies, organizational populations, or functional organizations.

[edelahayeUChicago]

Professor,

Thanks for your insightful and lucid paper, it had me on the edge of my seat!! One question that I had was the extent to which this is dependent on the type of processor? I am no expert on computer architecture but I imagine that the structure of the data that is possible is dependent on this and hence we need to be sure it is as close to a brain as possible! I don't know what brand of processor my brain would be for instance! Similarly, how well would this microprocessor used extrapolate to brains that work differently from the neurotypical brain, might the perfection of a computer processor inadequately capture the differences here?

[ddlee19]

Hello Prof Jonas,

Thanks for your paper! What are its contributions to neuroscience?

[yunshu3112]

Hi Professor Jonas,

Thank you for sharing your research with us. Just like many of my fellow students, my prior exposure to neuroscience is limited. I gained a lot of insights from your paper.

In our computational social science perspectives class today, we also discussed the study of complexity. I wonder how you define complexity and complex system in the study of neuroscience? Is their a formal definition of complexity in computer science, and how does this correlate with the complex system of human brain? What is the role of simulation in studying neuroscience?

Thank you very much!

[bermanm]

Thanks for a very interesting and thought provoking paper and for pointing us to the very interesting discussion points. I have two questions. The first is whether we really should believe that a neuroscientist could or should understand a microprocessor. On the surface, I would think yes, but then digging deeper I'm not sure. If we flipped the question and asked could an Electrical Engineer understand a brain, I would probably expect the answer to be no. In that sense fields of study seem to tailor the tools to their system, and it may not transfer to other systems. At the same time, your point about needing more theory really resonates with me. I would doubt that that is a problem specific to neuroscience. Second, some of these other questions posed here have shown that neuroscientists have done a lot to understand the brain. Hubel and Wiesel, Sperry, Kandel and others all have made Nobel prize winning discoveries towards understanding the brain using their neuroscience tools. Not that Nobel prizes are the measuring stick for scientific progress, but a lot has been learned about the brain over the years from neuroscience. So how do those discoveries in neuroscience jive with this critique of neuroscience. Again, thanks for a very thought provoking topic!

[fiofiofiona]

Professor Jonas, thank you for sharing your research in our workshop, it is a very interesting idea of examining neuroscience techniques through understanding microprocessor. However, I am wondering if the analogy between human brain and microprocessor was based on an assumption that they have the same or highly similar hierarchical structure, and how was this assumption proved by current neuroscience evidence. With that being said, whether human brain and microprocessor should be treated and studied using the same approaches is what I would like to learn more about. Moreover, I am curious about your thought on understanding human brain using the approaches to analyze microprocessor -- would you expect a similar result that the latter could hardly explain the former, or microprocessor could be a critical building block for understanding the key characteristics of brain structure?

[yhchou0904]

Hi Professor Jonas, thank you for sharing your work with us! At least for me, neuroscience is a topic that is hard for me actually to understand the biological mechanism and the information process. The paper also indicates this difficulty with analyzing the microprocessor, which gives me an understanding of the possible misunderstanding or lack of interpretability of some seemingly naive theories or methods. From my perspective, it might help if we standardize and rule more within an experiment. Or are there other intuitive guidelines for us to reconsider and adequately utilize these methods?

[sushanz]

Hello Professor Jonas, thank you for taking your time! With my limited knowledge in neuroscience, I have several questions in your arguments. If microprocessors could among those artificial information processing systems that are both complex and that we understand at all levels, how did you prove that it is different with the traditional measurements? Would you mind explaining in details why implicit conclusions fall short of producing a meaningful understanding? Thank you!

[yuzhouw313]

Hello Professor Jonas, thank you for sharing your work on the relationship and cooperation between neuroscience and microprocessor. It is enlightening to get to know both concepts, and I am looking forward meeting you on Thursday to hear a more comprehensive view on such topic.

By reading your method of converting the binary transistor state of the processor into spike trains, I found it difficult to envision this artefactual process. Can you elaborate on this process more with some examples? Thank you so much!

[awaidyasin]

Thanks for sharing your work. I just have one comment on the reading.

While I have a very limited background in neuroscience, I see great parallels between a neuroscientist's attempt to understand a microprocessor and an economist's attempt to understand the macroeconomy. Both parties are trying to comprehend a system that is intricately complex, has a large number of interrelated agents, and generates vast amounts of data. However, still, there is no one across-the-board accepted theory of how things operate.

The one difference that I do see is that the former attempts to understand the system with all its complexity, which is exponentially difficult and hence ambitious. Economists, on the other hand, rely on simplifications that allow them to explain the major characteristics of the data (the rest isn't of primary importance).

[Peihan12]

Hello Professor, thanks for sharing your work with us. It truly gives us a lot of new insights and ideas in neuroscience. In your paper, you offer multiple ways and development in some areas to better understand the processor. Would you mind giving us more specific examples for elaborating on?

[YijingZhang-98]

Hi Prof. Eric, thanks for sharing. I was always impressed by the computer science and neuroscience. I was wondering, from the operation research perspective, is there any space where neuroscience could help solve / speed up optimization problems? Thank you.

[ChongyuFang]

Hi Professor Jonas, it is very exciting to see your work! Though I have little prior exposure to neuroscience and microprocessor, my question would be: in the Granger-causality part, did you conduct any stationarity analysis regarding the signal series?

[YLHan97]

Hi Professor Jonas, Thanks for sharing your research with us. I have a question as follows: In the article “Could a Neuroscientist Understand a Microprocessor?”, you have mentioned that microprocessors are among those artificial information processing systems that are both complex and that we understand at all levels, from the overall logical flow, via logical gates, to the dynamics of transistors. Since I’m really interested in the machine learning applied in real world, but not so much familiar with neuroscience area, would you please provide more real world examples in your relevant area?

[LynetteDang]

Hi Prof Jonas, thank you for sharing your work with us. After reading the paper, I am wondering what you envision to be the future of neuroscience: would we be able to understand the neural system using a new analytical approach? Do you feel like human will ever be at a point to be able to understand the complexity of microprocessor?

[cgyhumble0612]

Hi professor, thanks for your sharing. It's really an unfamiliar but interesting field for me. I would like to ask what's some empirical benefit for learning the relationship of neuroscience and microprocessor. How can we apply the founding of this paper in real world?

[tangn121]

Hello Professor Jonas, thanks for sharing your work! As you have mentioned, it may be an important intermediate step for neuroscience to develop methods that allow understanding a processor. But our brain is not like a processor, it’s much more complex, could you please elaborate more on how to deal with the conflict and uncertainty when applying these techniques to understand our brains?

[BaotongZh]

Hi Prof Jonas, thank you for bring us such a great work. I am curious to what extent that such a micro processor could replace human's brain. Thanks a lot.

[jinyz1220]

Hi Professor Jonas, I'm looking forward to meeting you this Thursday! My question is: in general, what should behavioral scientists be cautious about when making implications from neuroscience studies? How can we bridge the gap between the computational level and the implementational level? In other words, how can we promote analysis of the algorithmic level according to Marr's theory?

[yujing-syj]

Hi Prof Jonas, thanks so much for sharing this interesting topic with us. To me, neuroscience is a brand-new and difficult field. Do you have any suggestions about the introductory books and literatures for the layman? Also, what are the key knowledges that we should have when we get in touch with neuroscience study?

[jiehanL]

Hi Prof.Jonas, Thank you for presenting your work! To address the process of "understanding," you used many methods in the paper (connectomics, lesion studies, tuning properties, etc.). However, how should we define "understand" other than a descriptive understanding of the workings of the microprocessor?

[xin2006]

Hi Prof Jonas, thank you for sharing the work with us! I am interested in the analysis of behavior. In the paper, you mentioned that how our brain generates behavior seems to rely heavily on the environment and the route is various and complete. And you also illustrated how the behavior relied on the collection of neural data. The relationship between the brain, neural and behavior reminds me of the AI application, which imitates human behavior. So I am curious how the neural data and mechanism helps to explain that AI could act like humans to some degree, but not substitute us? Could you please elaborate that using some examples?

[WonjeYun]

Dear Professor Jonas, Thank you for sharing your wonderful work. The idea of using the processor, the so-called 'brain' of a computer, as an approximate model for the 'real' brain was very interesting. Although the article was logical in itself, I could not bother thinking about how close the microprocessor is to the actual brain. As stated in the article, the real brain and a microprocessor differs in many aspects and in complexity. Moreover, in my short understanding, the way the microprocessor works and how the brain works differ from its basic structure. What is the further reasoning behind using the microprocessor in comparing it to the real brain other than the microprocessor being complex but also understandable? How close were the microprocessor and the brain so that you were able to develop the research questions?

[jiayan-li]

Hi Prof Jonas, looking forward to your presentation! The question I have is: why do you think computation within our brain and microprocessors share so many similarities you mention in the article? Obviously human brains come first, but I wonder if the design of computer processors borrowed how human brain works, or is it a coincidence of some sort.

[linhui1020]

Hi Prof Jonas, Thanks for sharing your work! New to this field, learned a lot! Would you explain more about how they are similar with each other? and What are some areas still unknown?

[hazelchc]

Hi Prof Jonas, thank you for sharing your interesting work with us! I'm interested to know what are the use cases of microprocessors in reality? I'm also curious to know if there are any ethical issues/ concern involved? Looking forward to your presentation!

[beilrz]

Hello professor, what is your opinion on deep neural network, which suppose to simulate how neurons in brain works? do you believe mimicking biological brain is a good path forward for computer algorithm design?

[XTang685]

Hi professor, thanks for your work on this subject. Could you please talk a bit about how or whether this processor would take place of human brains? Looking forward to your presentation on Thursday!

[zhiyun0707]

Hi Prof. Jonas, thanks for sharing with us your research! As I have not exposed to neuroscience a lot and very new to the field, I find that it's interesting that you argued in your research that despite the amount of data, the analytic approaches may fall short to produce meaningful understanding of neural systems. My question is that could you please provide some empirical applications for the neuroscientists understanding of microprocessor?

[Anmin-Yang]

!! Dear Prof. Jonas,

Thank you for sharing your interesting work. As I understand it, your work seems to suggest that we need a stricter null hypothesis when employing the methods you mentioned when interpreting brain data.

My first question is related to your methodology. The task of the microprocessor is to run video games, which is similar to the naturalistic stimuli in neuroscience research. However, the dominant research method is still a lab-controlled experiment where only the hypothesized function of interest is tested (e.g., the representation of bar orientation). Do you think the unsatisfactory results from some methods may be because of the research design?

My next question is about the logic of neuroscience research. In your paper, you seem to define understanding as a building process from the element functions implemented by circuit elements that eventually give rise to computation outputs. This reminds me of a related opinion from Samuel Gershman's commentary paper Just looking: The innocent eye in neuroscience that argues the pure data-driven approach in neuroscience would not work if one does not have a theory of what the neurons should be computing. He either would not design an experiment but use naturalistic stimuli or could not interpret the however big data he curated. I wonder would you agree for neuroscientists the first step of research should be to hypothesize a basic function and then design experiments and analysis, or pure data-driven could also give rise to understanding?

My last question concerns using deep neural networks in neuroscience research. People find units with various brain-like turining properties in those deep nets. Do you think deep nets could be a plausible model to perform in-silico experiments as some may advocate? And more importantly, how strong would the conclusion be drawn?

Looking forward to your talk on Thursday.

Best regards, Anmin

[taizeyu]

Dear Prof. Jonas,

What is the application of this technology. And what is the effect of this improvement? Thanks

[secorey]

Dear Dr. Jonas,

Thanks for coming to present your paper. As you note, there are many parallels between the structure of the brain and the structure of microprocessors, to the point that some would argue that, if we had a complex enough computer, we could essentially mimic the brain. Furthermore, it seems that many people who are developing AI models are trying to develop a machine that is as "human-like" as possible. However, I wonder if this is a worthwhile endeavor, as the brain's purpose is fundamentally different than that of a computer, and humans are incredibly fallible. Do you see the brain as a sort of prototype for what computers can be in the future, or do you think that this idea should be abandoned?

[bowen-w-zheng]

Hi Dr. Jonas,

Thank you for presenting your work and highlight the current short-coming of neuroscience method. I wanted to point out that not all neuroscientists take a data-driven approach and most current PIs approach modeling the brain from a theoretical physics perspectives. What do you think the merits and limitations of such perspectives and what the road forward should be?

[yiang-li]

Hi Prof Jonas, thanks for presenting your work! My question was: could you please elaborate on the mechanisms behind the similarities between the human brain and the microprocessors you mentioned in the article? How do you think the data-driven interactions might have some ethical issues?

[koichionogi]

Dear Prof Jonas, thank you so much for your research work! I have a question regarding the next quote from your paper "the problem is not that neuroscientists could not understand a microprocessor, the problem is that they would not understand it given the approaches they are currently taking" What kind of research do you think it would have an ideal combination of theory work and data analysis work that can help researchers understand the microprocessor? What is needed?

[mdvadillo]

Dear Professor Jonas, thank you for your presentation! My question is the following: in your paper you mention your approach revealed the structure of the data but not the hierarchy of information processing going on in the microprocessor. Do you have any suggestions or work in progress focused on how to tease out the hierarchy?