Hey everyone and welcome to today's episode of Developer Tea.
In today's episode.
It's been quite a while since we had a guest, and today we are joined by Brian McCann.
Brian is the CTO at You.com.
We have an excellent conversation.
It's quite a wide-ranging discussion.
We dive into some philosophy.
We dive into some conversation about AI.
Brian has been on the ground floor, you know, working in very early research on AI.
And I hope you enjoy this discussion.
It's a little bit of a different you know a different kind of approach, not as advice heavy in today's episode, a little bit more kind of theory and just a discussion between two people who are thinking a lot about these topics on a.
Hopefully this is insightful for you and helpful for you to think about.
You know, perhaps kind of crystallize your own philosophies on these subjects.
Thank you so much for listening.
Let's get straight into our interview with Brian.
Brian, welcome to the show.
Hi, Jonathan.
Thank you.
Thanks for having me.
Absolutely.
It's been quite a while since I've had a guest on the show, quite honestly.
And we've been doing a lot of the kind of like the solo episodes in the past couple of years, mostly because my career has kind of ramped up in terms of what is demanded of my time.
So I'm excited to have you as my first guest back because you kind of well, I'll let you kind of get this information out.
For somebody who has no idea who you are or what you've done in the past.
What are the first couple of things you tell them about the work that you do?
Great question.
I'm honored to be one of the first guests back.
I've listened to some of the more recent episodes too.
Very interesting stuff.
Very interesting to see how you're thinking about some of these things and some of the concepts you're introducing.
I really appreciate it.
And maybe with that it's a good segue to.
I usually open with telling people that I was a philosophy major.
And I think on the outside, it's obvious that I'm doing CTO stuff at an AI startup.
It's a little less obvious, but if you looked online, you could easily find out that I did AI research for several years at one of the industry labs before starting this company, ucom.
But before all of that, I was really interested in philosophy.
And I actually still see everything that I'm doing as an extension of some of those original questions.
So I was really interested in meaning, still interested in meaning.
And in some sense I only got into research to study those questions just happened to be with computational tools.
Making neural networks was the best way to do it.
Super interesting.
Okay, so for somebody who heard that and doesn't, I guess, intuitively connect the dots between going from philosophy to calculus for writing a neural net or something, can you help me understand?
There was a moment at some point in that journey as a philosophy major and even before that, the spark to get you into philosophy.
But then another spark, it sounds like, to get you interested in this applied version of philosophy to AI.
Can you tell me about that moment?
Yeah, I guess we can go back many years now.
I was talking with a friend about this recently.
When was the real first moment?
And the first thing that could actually come to mind was when I was in a religious class in fifth or sixth grade, something like this, and there were questions that were being asked that I couldn't quite make sense of and I didn't have good answers for, and it stood out against all of the other subjects that were more about learning and doing applying your knowledge.
But I could not get answers to some of those questions.
And I think that's when my brain first started really thinking about what's going on for real and why and how can we explain it?
When I went to college, I still had a sense of this.
It had transformed over the years through high school to become very interested in these ideas of meaning broadly speaking.
We're having, I hope, what will be a very meaningful conversation.
It may be meaningful to a listener to hear our thoughts.
Most of those conversations around meaning get, I would say, reduced or pushed into conversations and questions about language, at least in the academic philosophy world.
Because we can hear language.
We can kind of see language when we write it.
We can study that.
It's a little bit more tangible and perceptible than the more general questions of what is meaning.
And so the philosophy of language cuts into syntax and semantics and all of these other things.
I studied that for several years. while I was studying computer science on the side for fun.
Much in the same way I was when I was in fifth or sixth grade, studying a lot of subjects, but then I had these big, meaty questions in my mind that were unsolvable until my third year in undergrad, when I felt like I had exhausted the philosophy of language approach.
I felt like I'd hit a dead end.
And there was only so much armchair philosophy one could do.
And I got this idea to maybe use the computational tools that I was learning how to use on the side for fun, to study language as well and try to make machines that could make meaning, thinking that if I could do that, I must learn something about what meaning is along the way.
And those questions captured your interest right in my understanding, because they were kind of fundamentally unanswerable or because at the time you believed that they would be the most difficult questions, or perhaps like a longer journey to answering them.
Did you have the belief that maybe there would be an answer that you could arrive at one day?
At the time, I definitely thought that making this switch and approaching it in these different ways would at least give me insights about it.
I don't know about final answers to the world's deepest questions, but I definitely thought that I would have better answers, or could get to better answers, when I started.
And that changed in my first couple of years of doing research.
I think I changed to think that it was less about finding those answers, that no matter how many neural networks I built and no matter how much data I gave them and how much better they got, when you look inside the machine you can't pinpoint any meaning module or you know point to something and say this is, this is where the meaning is.
So my expectations around that changed actually quite a bit.
Yeah.
Yeah.
It's interesting because you know one of the one of the goals I have.
I have three things that we try to you know, elucidate or or help people find on this show.
And one of them is, is purpose.
It's quite a loaded term.
It's, it's so individual.
And do you feel that you know this?
This was you tracking to a purpose, or would you say that it was more tracking towards an interest or or following some kind of intuitive leading or Which resonates, I guess, better for you?
Well, I think it was both at different times.
I think when I first started doing it, it was... it was mostly out of purpose.
And then when I started doing the research further, then it became more interest.
And when I looked in the machines and didn't see that I would ever find an answer to the purpose questions.
Perhaps it became more about interest.
But For me to keep doing it.
I then kind of re-derived another purpose from it in trying to make more unified approaches to AI.
And I developed other purposes, I guess, within the field.
And new questions came up after the old questions I had answers to, even if they were indirect answers.
I think I...
I learned to disentangle what we might perceive as meaningful perhaps in this case, like coming from machines and from what is actually meaning.
And I at least made the categorization that meaning, in the sense that I was looking for it, is just an innately human thing.
It is that thing that is more tied to purpose.
It is more tied to reality.
Narratives that we use to uh, guide ourselves but also shape our sense of reality and shape who we are.
Um, and even if a sequence of symbols is generated by a machine um, it's really us perceiving that in some sense and it registering in our minds and having that resonance kind of strike throughout our linguistic community.
That gives it some sense of meaning.
And that meaning is just close to the other forms of meaning that we experience, but it's not the same.
So all three are actually different.
It's like it's making connections with some other thing that we've either experienced or that we've thought about.
I talk to my children about this often, and we talk about various rules and cultural norms.
So we live in Tennessee, and there are rules, of course, in every school system.
We have a particular set of rules in their school system.
One of the rules has something to do with, I think they can't color their hair what the Tennessee, our area, calls an unnatural color which I take a little bit of.
You know personal frustration with.
But you know, and so I was talking to my children about this and trying to give some kind of best interpretation.
And my what I've, what I've tried to teach them, is that different things mean different things to people.
You know to sorry.
Different things resonate differently with different people.
And so you know, for you this might just mean that your hair is red, but Or I guess red is what they might consider a natural color.
But anyway, there's some kind of signifying thing.
This is symbolic or it triggers some kind of other process for this person.
And we show respect to the communities that we are a part of by understanding what things mean to other people.
It's a deeply human ability to have empathy.
And to understand that this means something to that person, that it doesn't necessarily mean to me but me discarding it would be an affront to them in the same way that I wouldn't.
I wouldn't want to wear something that's offensive, or offensive say something that's offensive to my neighbor, unless I'm.
You know I'm being offensive on purpose, right.
Like, and generally that's not a value that we want to teach our children.
So I guess all that to say, there's no inherent meeting.
It's just you know the scientific level.
It's light waves that are, you know a particular frequency.
How does that change whether my child can learn in school?
Yeah. that's not really the point, right?
If we were to boil everything down to its pure scientific description, then all of those symbols you're talking about become the same thing.
It's just light wave transmission or whatever.
In that case, it's just symbols in a particular form.
But we, I guess, imbue that experience with meaning.
Is that kind of where your research led you?
I think that's a reasonable way to phrase it, yeah.
I think we kind of yeah, we light it up, you know, and make it actually meaningful, at least in the sense that you know, I constrain my sense of what meaning is now.
It's interesting.
You mentioned, you know, connection and you mentioned maybe, this reduction.
Let's take view down to symbols or light waves, and I think this is something that I was thinking a lot about in parallel with that because well, there's one example.
I've been kind of playing with a classic philosophical example right,
Like, do we have free will or not?
Is everything deterministic?
And on the one hand, we can go into that question.
You can have lots of debates.
There are kind of endless arguments on either side about which one is actually true or not true.
But you know, even if I believe on some level that everything is determined and mechanistic, it's really
Almost, I think it's impossible for me to actually live that way.
And so you know, I still wake up and I do things and I experience my reality as if I have some agency in the world.
And Maybe one could transcend that.
But I think there's something similar with these questions around symbols versus meaning.
You can't completely dissolve away these constructs of meaning which are, I think, fundamentally based on this actual framework of connection.
Connection again, kind of more broadly speaking, like maybe you and i are connecting through this conversation, maybe we're connecting to uh, your audience.
Um, maybe you know, two things are connected with a phone charger.
There are different forms of connection, but connection is definitely a real thing and it it has a, has some sort of relationship to different kinds of meaning as well.
You see this in the kind of, in the details of the math of neural networks that they try to measure distance between even different symbols, different words.
But in some cases, it's not so much the distance that matters.
It just matters that things are connected.
And that might be a more kind of fruitful path in thinking about I don't know, what is reality?
Is it the light waves or is it more our experience of it?
I think these are some of the things that I've more or less started to throw up my hands and think of them as very interesting philosophical questions, that I do hint at some further scientific inquiry as well.
But I actually see all the recent stuff in AI as evidence that we're just tapping into it.
It's like we just built the first machines that can harness the power of electricity.
Those electromagnetic fields were always out there.
We just didn't have a way to know that they were and we didn't have a way to use them and harness them.
I think there's some sort of framework of meaning or connection out there, resonance in the biological neural networks and humans and animals and all of these things.
That's real.
But we are somehow creating tools that can measure it, harness it, tap into it, draw from it.
And you've seen what happened with electricity.
So I do expect big changes based on that in the world.
Yeah, and fairly rapid ones.
It seems that...
And this is not my area of study.
So I'm going to say words that probably mean very different things to you than they might mean to me.
So it's kind of a naive take.
I love that.
As we go through here.
But it is in some way and I think this is probably felt by the researchers working with this one of the first bits of research that helps us do the research itself using the thing that we just discovered.
And similar to how tool building or like in the industrial revolution, You have the standardization of bolts and various sizes and things like that.
And because of that, you could now turn around and build new assembly lines.
You could build new tools in order to continue building things more effectively.
And so it's a bit like a flywheel effect.
Do you think that that is having a major impact on how this research is even carried out in the first place?
Absolutely yeah, i think in many ways.
Uh, that's the next big goal for folks talking about.
Uh, you know, i think the new term is to talk more about like super intelligence and artificial general intelligence and and things like this, um ai, that is certainly already helping us build things.
Um, i see it in our company at least.
It seems like within the last year, companies like cursor, uh tools that they provide, and cloud code and and others, have definitely changed development and they're already helping us build The nature of the building blocks themselves.
I think that's a big goal for folks right now as well to try to get not just tools for us to build better bridges, but for the AI to build better versions of itself more and more automatically and get that flywheel going where?
It accelerates on its own at the rates that it can handle that humans can't really handle.
And that would be the path towards some sort of superintelligence, even beyond artificial general intelligence, which is arguable whether we have achieved it or not.
But superintelligence being something that is obviously better at so many things than all of humanity ever could be.
Right.
Right.
Yeah.
Because our classic belief has been that humans are kind of the peak of intelligence or like if you were to rewind to you know 1960s or something, or even just look at like the Jetsons.
Right.
The the kind of vision that we, that that culturally we held, was that machines might one day be like humans in this kind of anthropomorphic vision that we had of artificial intelligence.
I imagine that most people listening to this now find that ridiculous, and certainly researchers, I believe, would.
Machines are very quickly surpassed humans in special intelligence, right?
Just hold a calculator in your hand and you can recognize quickly that special intelligence is something that machines are incredibly good at.
And so then it stands to reason, I think, that machines may become good at many things, much better than humans were.
Like you're saying, superintelligence seems like the aspirational goal to strive towards.
Or this is kind of setting aside questions of moral clarity, I suppose, or ethical clarity.
But assuming that progress was kind of directly aligned with what we care about uh, as as a species, then you know the the next step or the goal would be would be this progressive move towards, towards super intelligence, setting aside that anthropomorphic vision entirely.
Would you agree with that?
Yeah.
I, I like the idea of setting the anthropomorphic vision aside for, for now.
Um, A lot of the themes of my research, as I was making large language models and things as well, was trying to frame it as a tool, as a creative, a new creative tool for people.
You know, not replacing, not anthropomorphizing.
It's a tool that allows you to maybe go from zero to one much faster, but then you can always go, you know, to some place new after that.
It doesn't need to be an entirely complete substitute or replacement um, but i do think you know.
But it's The anthropomorphization.
It plays a role too.
It generates a lot of excitement.
So, just practically speaking, it's this odd fascination we have with I don't know if you want to call it playing God or what.
We want to make something in our image.
And it does drive a lot of interest.
It drives a lot of capital in that direction.
It is endlessly fascinating.
But it's also very easy for us to adapt very quickly, I think, to new forms of it, until we see the gaps and something doesn't quite feel right and you're kind of instantly impressed.
But then the goalpost moves.
Whereas with the other things, the maybe more specialist types of intelligence, it's just obvious that These things can ingest and understand in some definition or the word understanding, maybe fundamentally different type of understanding than us sequences of amino acids proteins biology chemistry physics, and probably make connections, real connections that are meaningful, that we aren't going to make.
Even if you have the best, most intuitive way genius, sit there and try to design molecules.
Machines are probably going to be better at that.
And that's not really something we even need them to be human-like for.
I think that's actually a better use of most of our investment, except for the fact that I acknowledge garnering that investment sometimes requires playing to the anthropomorphic narrative.
I don't personally need robots to be better in my life than the humans.
What I so consider the human things.
I also distinguish that from the fact that I think they could be, you know like yeah, i've got a best friend, i love him.
Uh, i don't think it's impossible for machines to be better as a best friend than me one day.
You know, if they're embodied robots and like they can probably be more understanding, more compassion, or they can at least seem that way sure, right?
So i don't think uh, we should.
I don't think we've found the thing that makes us unique or special yet.
And in that search the only way to find out if there is anything is to kind of keep going and see what isn't possible to replicate and replace.
Um, But for now, I don't think we've found it.
We've had a lot of suggestions, intelligence being one of them.
But I don't know.
I'm not sure.
Certainly pushing on the edge of that, right?
Yeah.
Yeah.
So I think I want to get around to questions about skeptics here in just a few minutes.
I'm curious.
I see two different kinds of skeptics.
But before I ask those questions, I want to kind of take a step back and...
You know, this is probably it sounds like this is something you you think about talk about often.
But what do you wish in these kinds of discussions, interviews, etc?
What do you wish interviewers would ask you more about?
What is this is usually something I hold till towards the end of the interview.
But what's what's a topic or an area that you wish you could talk more about?
Yeah, great question.
I think the things that I'm thinking about, I suppose outside the normal topics, I mean, these aren't my jam.
I like thinking about these things all day.
I think a lot about how we might We have unresolved questions seemingly in physics and how we understand the universe there.
I love thinking and talking about organizations and people.
Part of why I left research was because I wanted to build companies and teams.
If I couldn't find the meaning itself in the machines after making them, that meaning was kind of reserved for humans.
Then, like I like working with humans and making things meaningful,
So talking about, like, organizations and how they're changing, thinking a lot about that.
And art as well, you know, and the role that art still plays for us um, even as ai may or may not be like invading part of that world, i think it's something really important to me.
Um, and they all play into uh, what i kind of call like a search for meaning um, whereas maybe you know maybe uh, Maybe that's what we're really good at.
We're very good at continuing that search.
And in some sense, that looks like moving the goalposts.
But in some sense it just looks like reconceptualizing ourselves and finding a new narrative, finding a new story to keep going and to motivate ourselves.
I heard this described as spiral learning.
Are you familiar with this term?
Where you come back around to something and you refine it each time you touch it.
And I think that's probably like a more.
It's kind of the same things you're saying with moving the goalposts, but
I think as we uncover, you know, learning or defining meaning is an emergent thing.
It's not necessarily, or I guess you know.
One hypothesis would be that meaning is not just a fixed location that we are searching, as if it was like a you know on a two dimensional map.
Once you get to these coordinates, you found it.
I think it, it continues to emerge through a variety of, of manner.
So, and, and is also not necessarily just deterministic for the, For all of all of people, right?
Like every person, I think you could imagine that it's going to be different for each person and also different for that one person over time.
So that search becomes, I guess, becomes the unearthing rather than the searching.
You're finding things that are meaningful.
But I do like the idea that you mentioned earlier about the fact that we haven't quite found something that is truly fundamentally what makes us a different thing.
That intelligence, we thought that was it.
We thought that certain types of relational you know, being able to have empathy for others is another thing that seems like it could be it.
But it's still easily challenged.
We can observe that in nature.
Why do you think we even need that?
Is that like a survival thing that we just inherently feel like we need to be differentiated and have moat from other things?
The only thing that makes sense to me is that we have some intuition.
We believe it's intuition.
Maybe we're just describing away a reason that we feel different.
But that feeling different might just be like you're saying.
You know, if we didn't feel different, perhaps we wouldn't be as motivated to continue surviving.
If we just became part of you know the larger ecosystem or something, then maybe, maybe the fact that we are the most dominant species is was dependent on us feeling special.
Right.
And so that's how we got to where we are.
And, now it's become a bigger search, but in fact, it was just a mechanism of survival.
Yeah.
No, I, I think that's probably, that's my, that's my sense as well.
Um, Once again, I'd like to thank Brian McKinn for joining me on the show.
This interview and none of the interviews that we've done in the past was a sponsored interview.
But I did want to mention, in case anyone is looking for a role, that Ucom is indeed hiring right now.
They didn't ask me to say that, but it's worth going and checking out their job board.
Thanks so much for listening to today's episode.
I hope you enjoyed this discussion with Brian McCann.
This is part one of two parts.
So we normally split our interviews like this in half because we want to make sure that we continue to deliver on releasing episodes that are certainly under an hour.
We try to target much shorter, but these interviews tend to run a little bit longer.
So you will hear the second part of our interview with Brian McCann in the next episode of Developer Tea.
Thanks so much for listening.
And until next time, enjoy your tea.
