I get a fair number of emails from interesting people with interesting ideas who would like to spend time having conversations with me. I enjoy having conversations like these. Each of these opportunities would be genuinely stimulating and enjoyable, but I do have quite a few other commitments.
In responding to a recent request, I was able to express some ideas that might be of more general interest.
Dear Prof. Kuipers!
I hope this email finds you well.
My name is ___, and I'm an independent thinker working on characterizing certain systems as "4D informational objects" -- structures that integrate experiences over time (processed entropy) to determine their present functioning and future actions.
The core idea is to reframe biological agents like humans (or other animals, plants, fungi, etc) and even some AI systems, not just in physical/material terms but as temporally-extended informational/entropic entities. One example that might not be immediately noticeable as a kind of 4D object would be an AI system such as a corporation, which I've read some of your papers on in the past.
I don't have a formal background in this area, but I'm very passionate about developing this conceptual model out of a personal interest. That being said, as someone thinking about these topics from more of an outsider's perspective, I could greatly benefit from the insights of an eminent expert like yourself.
I would be incredibly grateful for the opportunity to outline my ideas to you and receive your candid feedback and guidance. Specifically, I'm seeking your critique on:
If you're open to it, I would love to interview you more broadly about your own background and perspectives that shaped your pioneering work in this field.
If you're willing, I would highly value your time for either/both the feedback discussion and interview. Is this something you'd be interested in?
Thanks for your time!
Hi ___,
I'm pleased to hear that you are interested in these things, and that you are working to create your own conceptualization of them.
Lack of a formal background can be a problem, not for me, but for you. It's not because of lack of some credential to get you in the door. It's because there are some ideas that are deep, subtle, and complex, and may require careful study and thought over weeks to semesters to years to a lifetime to grasp adequately.
One example, where I DO know (part of) the relevant basic knowledge, is control theory. If you want to drive a robot along a hallway, you need to steer it to avoid crashing into the walls or obstacles or pedestrians. Control theory is based on the mathematical theory of differential equations and dynamical systems. When you have a basic understanding of that theory, a lot of the relevant ideas are reasonably obvious and intuitive. But without that theory, you can't really tell why things work or don't work, and you can't design solutions to problems.
A second example where I do NOT know the basic knowledge is quantum theory. I have read a lot of the popular discussions about quantum reality, and several of the technical textbook presentations. I have an unconventional idea that makes sense to me with all the examples I have ever heard of. But I haven't mastered the relevant mathematics enough to present my idea to experts, because I don't know their language for that discussion. I sit with my idea, pursuing it when opportunity arises, hoping that someday spending the time to learn that math will be a priority. But I am forced to accept the fact that I can't really contribute to that research community (even if my idea is as great as I think it is!) without knowing that language. So it goes.
[Note: There are plenty of areas, including in AI, where an intelligent non-specialist with good intuitions can make a significant contribution. Sometimes terminology and jargon are unnecessary barriers to entry, so clear and precise thinking can cut through the nonsense. But sometimes there really are non-obvious concepts that are essential to being able to make a useful contribution. The hard part is telling the difference.]
You are asking for an extended conversation with me. And I would certainly enjoy the conversation with you. The problem is the finiteness of time, and the other things on my plate.
Likewise, I would like to have conversations with a lot of the great thinkers out there. Fortunately, I can do that, because many of them have written accessible books and papers that I can read. Reading their work IS a conversation with them. I need to read it AS a conversation. I share an interest with them. They write at length, describing their thoughts on that topic, generally with great care. Since I am genuinely interested in the questions they are addressing, their explanations on those topics are gripping.
They raise interesting points; I have questions I would like them to answer. They may well not be accessible to me, due to being far away, or too busy, or long dead, or some other reason. But more than likely, if I keep reading, they will actually answer my question. Maybe not in what I'm reading today, but maybe in another paper that they cite. For some of them, I can send an email. An email with a specific set of well-informed questions shows that I have read their work, and is much more likely to get an answer. (As you see from the example of your very good email to me, which prompted me to spend the time to answer it.) So, you can carry on stimulating conversations, even with long-gone experts.
On my website, you might well enjoy my 2008 paper on consciousness, titled "Drinking from the Firehose of Experience". Also consider reading my 2000 paper, "The Spatial Semantic Hierarchy". Both can be found on my website, under "What's New?" or “Highlights”. There’s lots more you might enjoy.
For more formal background in AI, get and read the leading textbook on AI: Russell & Norvig, Artificial Intelligence: A Modern Approach. The fourth edition is the most recent, but the earlier editions are probably just as good (perhaps better) for your purposes, and can very likely be purchased inexpensively at a used book store. Jump around to topics you are interested in, the way you would have a conversation, going back and forth as needed to understand it. Include reading about reinforcement learning, which certainly has a chapter or two in Russell & Norvig, but the leading book on this is Sutton & Barto, Reinforcement Learning.
I like your focus on “structures that integrate experiences over time . . . to determine their present functioning and future actions”. However, I'm not very enthusiastic about the term "4D informational objects". Obviously, there is an important role for temporal change (1D), but the remaining 3D is too simple and constraining to characterize the state of one's knowledge of the world at a given time (see my paper, "The Spatial Semantic Hierarchy"). So, when I see the "4D" in that term, I imagine that you think of the world as 3D, and add 1D for time to express change. To see the issue, imagine thinking about whether your bathtub will overflow when you've left the faucet on. You have a 3D system, the bathtub, plus 1D for time as it fills up and maybe overflows. But then, this is a system you think about from time to time over the years, and you get progressively better at thinking about it, so we have a separate 1D temporal dimension representing your changing knowledge representation. It's at least 5D then! Though actually, the bathtub doesn't need to be modeled as a 3D object: you really only need 1D to model how full it is, so the overall model is now 3D = 1D (the bathtub) + 1D (dynamic time of the filling process) + 1D (time of your learning process). We build all sorts of simplified models for particular purposes, switching among them as needed. Does this make sense?
Please carry on having conversations with me, partly through your reading of my writings on these topics. Read and think. Write interesting stuff. Think carefully about how your writing provides value for your Reader. Once you have something interesting, please send me more email. I’d like to carry on the conversation.
Cheers,
Ben Kuipers