On this episode of the AI For All Podcast, Peter Voss, CEO & Chief Scientist at Aigo.ai, joins Ryan Chacon and Neil Sahota to discuss AI chatbots, ChatGPT, and artificial general intelligence. They talk about how chatbots with a brain compare to average chatbots, why more people aren't working on AGI, the benefits of AGI, the business need for AI chatbots, using ChatGPT for business, GPT and memory, trusting AI chatbots, modeling the mind, and the future of AI chatbots.

About Peter Voss

Peter Voss is the CEO and Chief Scientist at Aigo.ai. For the past 15 years, Voss and his team at Aigo have been perfecting an industry disruptive, highly intelligent, and hyper-personalized chatbot with a brain for large enterprise customers.

Interested in connecting with Peter? Reach out on LinkedIn!

About Aigo.ai

Aigo.ai presents the most intelligent AI chatbot that acts as an enterprise digital assistant, enhancing customer experience.

Key Questions and Topics from This Episode:

(00:55) Introduction to Peter Voss and Aigo.ai
(01:14) Chatbots with a brain versus average chatbots
(03:11) What is Artificial General Intelligence (AGI)?
(05:25) Why aren't more people working on AGI?
(08:06) Benefits of AGI for business
(10:28) What was the business need for AI chatbots?
(12:21) Applications of AI chatbots
(15:15) ChatGPT and use in business
(19:12) GPT and memory
(22:38) Benefits of AI chatbots
(26:13) Trust issues with AI chatbots
(28:38) Automation vs innovation
(31:03) Modeling the mind in a chatbot
(34:52) The future of AI chatbots
(37:30) Learn more about Aigo.ai

Transcript

- [Ryan] Welcome everyone to another episode of the AI For All Podcast. I'm Ryan Chacon with my co-host Neil Sahota. Neil, let me pass it over to you to do a quick introduction.

- [Neil] Hey everybody. I'm the United Nations AI Advisor and co-founder of their AI for Good initiative.

- [Ryan] Also with me today is our producer, Nikolai.

- [Nikolai] Hello.

- [Ryan] Today's conversation, a very good one. We're excited about it. We're talking about chatbots. We've all probably interacted with chatbots whenever calling customer support. And thanks to artificial intelligence, these chatbots are rapidly changing and even at times able to pass the Turing Test.

So to discuss this today, we have Peter Voss, the CEO and Chief Scientist at Aigo.ai, a company that has been focused on perfecting an industry disruptive, highly intelligent, and hyper personalized chatbot for enterprises, so a chatbot with a brain. Peter, thanks for being on the podcast.

- [Peter] Yes. Thank you for having me.

- [Ryan] So let me kick this off by asking you a question.

What does it mean to say that a chatbot has a brain versus maybe what we're used to experiencing when interacting with them?

- [Peter] Quite literally our chatbot has a brain or a cognitive engine that can deeply understand, reason, has context, short term memory, and so on, which none of the other chatbots have. The chatbot industry is still very much stuck in the model of IVRs from 30 years ago where you basically somehow identify what the customer wants, from a menu, press three for sales, and four for support.

Now, of course, today, that's a little bit more sophisticated where you have deep learning categorizers that can be quite efficient where you can say, blah, blah, blah weather, and then it'll pick the weather action. Or you can say, I hate Uber, don't ever give me Uber again, and it'll probably still pick the Uber action. And then the action itself is really just a flowchart program where- where do you want to go, how many people are going, and do you want Uber X? And that's really still the technology that's used by every other chatbot. Now, with some levels of sophistication and specialized code that's written for- to try and do things and integration to APIs and our approach has really been- I started off 20 years ago coining the term artificial general intelligence together with two other people, and that's really to drill down on what does it mean, what does it require to have intelligence?

And it requires understanding, using context, and reasoning. And this is really what our chatbot or our conversational AI does, is it has a cognitive engine.

- [Neil] Peter, you used the phrase artificial general intelligence. Just some of our viewers may or may not know what that is. What is that and why is that different than what most things are today, artificial narrow intelligence.

- [Peter] When the term artificial intelligence, AI, was coined some 60 odd, 60 plus years ago, the idea was to build a thinking machine. A machine that can think, learn, and reason the way humans do. And they thought they could crack this in a few years. That was 1950 something. Obviously it turned out to be a lot, lot harder.

So what happened over the decades, AI morphed into narrow AI where you basically pick one particular problem and solve that. But there's a huge problem with that. So take- let's take for example, the great breakthrough, I think in the eighties of Deep Blue, IBM's Deep Blue becoming the world chess champion.

Very, very impressive breakthrough. But it was really the intelligence of the engineers to say how can we use a computer, how can we use the power of a computer to write special algorithms that can basically play a good game of chess. So the intelligence wasn't really in the AI, but it was external.

So A, it's narrow AI solving one problem at a time or one set of problems at a time. Deep Blue, world chess champion, couldn't even play checkers. And then it's the external intelligence. So that's why in 2002, I got together with some other people who felt- we felt that the time was ripe to go back to the original dream of AI.

To build thinking machines, and- but that's why we felt we needed a different label for that. We wrote a book on the topic, and we came up, the three of us, came up with the term artificial general intelligence to basically go back to what AI meant in the beginning. And there's a small subset of AI research, a tiny subset that really are seriously working on AGI.

- [Neil] I'm curious, Peter, because I don't know that many people like seriously working on AGI. Do you have any insight as to why that is?

- [Peter] Well, it is hard. And you really have to have a long term view of it. And I call it the narrow AI trap. You know that even people who started off thinking they wanted to build AGI, they want to build a general thinking machine. When it comes to getting funding, when it comes to trying to commercialize things, it's just a thousand times easier to write some code, to use the external intelligence basically, or to train some model, which of course is the more common way of doing AI these days is to train deep learning models. So you can get specific results to solve a particular problem much, much quicker by doing narrow AI.

And this is basically what happens. People run out of inspiration or run out of money or whatever that try to pursue AGI or just get distracted. DeepMind, I think, is a good example of that. Their mission is to solve intelligence. Their mission is to crack AGI.

But, they got trapped by the success of deep learning, the narrow successes of deep learning. And they were spectacular, Go, AlphaGo beating Go world champions, which prior to that, AI researchers typically thought Go is really, really hard. If you crack Go, you've cracked general intelligence.

It turned out not to be the case. Again, it's good at playing Go, but again, it can't play checkers. So they got sucked into, where can we get success? Where can we get recognition? Where can we get papers published? Where can we get more funding? Attract people, smart people, and it's in having quick wins or relatively quick wins. And the quick wins in the last 10, 12 years have all been big data, big compute. If you can throw more data at it, more computing power at it, you can achieve better results. And so it's almost an accident of history that deep learning, machine learning, and now GPT has been so incredibly successful in solving particular problems that it sucked the air out- the oxygen out of the air for real AGI work.

- [Neil] I get the need for quick wins and, sorry, and rapid ROI, I'll say. What's the benefit for a business if they were to pursue AGI?

- [Peter] Well, I mean, to really crack AGI, we have what we would call a proto AGI engine. Our engine is still a long way from human level intelligence, but it does have all of the core components that intelligence requires. The ability to deeply understand, to reason about what you're hearing, to use context, use short-term memory, and to be able to remember things, to have long-term memory as well.

And that is a spectacular advantage. We- one of our customers is 1-800-FLOWERS using it for hyper personalized concierge service for their customers where it can remember what you- what gifts you bought for whom. 1-800-FLOWERS is a group of about 12 companies, they have Harry and David and chocolate and so on.

So having a chatbot with a brain, it can- you can really use it as a concierge type personal assistant for that. In the short term, even with proto AGI, what we have, there are significant difference- improvements of containment and customer experience that you can have. Now, in the long term, of course, AGI changes everything, and I think there's a lot of noise about that right now, good and bad in terms of ChatGPT 5, 6, 7 with the way people think about it, but that's really talking about AGI, human level intelligence. Imagine having, training one AGI to be a cancer researcher and then you make a million copies of that.

You now have a million cancer researchers chipping away at the problem. Just one example and whether it's pollution or energy or poverty or whatever problem you want to address, AGI will be able to help us solve these problems much, much quicker. That's once we get to truly human level learning and understanding.

- [Ryan] Peter, let me ask you, so when we first started this conversation, you mentioned how chatbots have evolved, where- what they were able to do, what you all are focused on allowing them to do now or creating them to do now. When you looked at the market prior to test this- trying to solve this problem, what was the problem that a lot of these chatbots were not able to do that companies really had a need for. You talked about what you're doing with 1-800-FLOWERS, but just like generally speaking, what is it that companies were yearning for and needed, and this is obviously working to solve that problem from a business case standpoint.

- [Peter] When I started commercializing this in 2008 in a company called Smart Action, we were focused on IVR, on call automation. And whenever I- somebody asks me what I do, and I explain to them what I do, they say, oh yeah, I hate these things. I always press zero to get to an operator.

That kind of answers your question. What is the business need? What does business need? What do customers, consumers want when they're interacting with a machine? They wanted to do whatever they're trying to achieve, and IVRs are notoriously bad. I'm sure we all still experience that frequently with bank and insurance companies and whatnot, airlines, how awful most of these IVRs are because they don't have a brain.

So being able to have something that truly can provide the service, there's tremendous need for that. And that's just call center automation. There are other areas I can go into as well, but call center automation is obviously a big one.

- [Ryan] What are some of the other applications of this? Because if you think about- when we think about chatbots as like an enterprise digital assistant, when- the first thing people think about when you mention chatbots is what we've talked about, customer support, help on a website or on the phone.

But where else does this kind of venture into? What are the other applications that you all have seen or potentially will see, and we will see in the future as this becomes more widely adopted, more sophisticated, and more usable.

- [Peter] So, the call center applications are essentially owned, and the agenda is of the company that runs the call center trying to provide service for their customers. But then there's this whole area of the individual, as a personal assistant to the individual. Now, Siri and Alexa, especially Siri and Google Assistant are giving us a taste of that. When Siri first appeared on the scene, people were blown away by the promise of what it could do and the promise. Now, unfortunately, they seem to just be treading water. They don't seem to have done much with Siri at all. But the personal assistant and whether it's for a student at a university to help them cope with student life, all the different things.

It's just starting off there. You don't know your way around, where do you get food, where do you get meals, and how do you schedule your classes, and a million and one things, so a student- a personalized student assistant that's personalized to the individual student would be one example.

An assistant for a salesperson. Salespeople hate to update Salesforce. So what if you have an assistant, you can just talk to it and tell me about my next appointment. What product were they interested in? Do they have kids or what are his hobbies? Or whatever. You're done with your sales call.

You can then just talk to Aigo or to your assistant and say remind me next Tuesday to follow up, send the customer brochure X, and let my boss know what's going on. So that would be personalized productivity aid or diabetes management, helping people manage diabetes. So whole range of things.

One other category I just wanna mention quickly is what we call a copilot. A lot of software out there is horribly underutilized because it's too complex. Pretty much any software that you have now from dashboards, performance dashboard, whether it's Excel or whatever, Salesforce, if you had a copilot that you could talk to, interact, and tell it just to do stuff or to find stuff for you, that would allow people to use software much more effectively. Like Clippy but working properly.

- [Neil] I'm gonna try and draw a, I don't know, metaphor, I don't know what's the right word here, but people are familiar with chatbots. They're like automated IVRs, like you're saying, Peter. We're at the stage now where we have the sort of almost assistant, maybe concierge, like you used the great college example.

I know some businesses that are actually using that. They train like ChatGPT on their business. So central business partners, investors can do due diligence. This idea of a kind of a copilot, almost like it's a knowledge worker, and like you're talking about, a thousand cancer researchers, what's the next step or is there currently a next step for businesses to try and engage down that path?

- [Peter] ChatGPT is very interesting. It really- it's really a game changer because I think it just overnight recalibrated or reset our expectations of what current technology is capable of. The- how fluid the conversation is and the variety of sophisticated conversation that it can produce.

It's really phenomenal and shocking. The problem is to actually use it commercially, you hit a wall pretty quickly, and the reason for- a couple of core reasons that are inherent in the technology and the one is that it makes up stuff, confabulates it. You can't rely on what it says because it was trained with garbage, so that garbage will show up unexpectedly, and there really isn't a remedy for that, because you would need another layer that is actually smarter than ChatGPT to be able to correct it, or- so they use tricks to try and put guardrails around it, but they're not super effective.

So the- that you can't rely on it. So for any corporate- serious corporate application, you really can't use ChatGPT. The other inherent problem with this, that it really cannot learn new knowledge interactively. It can within a buffer and so on, but that knowledge isn't really deeply integrated.

It's not permanent because it's a black box, there isn't really a way of it learning interactively. The third inherent limitation is that it doesn't have metacognition, which is the key to human intelligence that we can think about thinking, and ChatGPT doesn't have access to its thought process because they aren't thought processes.

They're statistical calculations basically. So the bottom line is ChatGPT is excellent for helping you on stuff that it happens to have in memory where there's a human in the loop to validate, is this useful, is it not useful, for generating ideas, for reformatting text in a different way.

There's so many different uses for it, but ultimately I think it's generally recognized that it's a dead end, or some people call it an off ramp as far as AGI is concerned.

- [Neil] It's not like it's a step in the evolution of AGI. It's like a different branch entirely?

- [Peter] It's created tremendous amount of excitement, a tremendous amount of money flowing into AI again. So definitely an AI summer as far as that goes. So it shows the promise of where we want to go and being able to go there, but I see that more as a sort of psychological stepping stone than as a technological stepping stone because I think it's actually going in- it's in the wrong direction just trying to do things with purely statistical means.

- [Nikolai] So I have a technical- this might be a technical question, but I'm curious what the- why there's such a limitation on memory. So what is- why can't GPT-4 remember more than it does? What are the technical challenges with memory in something like a chatbot or GPT?

- [Peter] Yeah so, GPT, just the name itself already gives a clue. It's generative. So it generates stuff from its model. Pre-trained is the key here. It's pre-trained. It's never inherently been designed to try and learn on the fly interactively. It's pre-trained.

It's- in the most basic form, it's a read-only model. Now, of course, it's more than that because they've now added input buffers to it, and these buffers are getting bigger and bigger. But the problem with the buffers is that they need to be processed as input every time you use the system. It's not knowledge that is then integrated into the network.

So if you give it a new fact, if you say, my sister's cat, Toby, is at the vet or something, every time you now ask it a question, it has to process that again to have it in its bias or in its- in order to predict what the next word is, quite frankly. It's not integrated into the core model.

There's just simply no mechanism to integrate it into the core model because that would require retraining the system on the fly. That's- it's a black box, so the new knowledge that comes in can't just be inserted in the right place because you wouldn't know where the right place is.

- [Nikolai] So then the Scaling Hypothesis obviously wouldn't fix something like that because you're just scaling the same architecture, right?

- [Peter] Correct. Yeah, and I think Sam Altman, for example, has already said that. He said we are pretty much done with scaling. That- couple of reasons. It's becoming so incredibly expensive, hundreds of millions to train these models. But then also if you want to experiment with different models, you can't do that.

But also there isn't enough data, useful data out there anymore. They're already- and it's diminishing returns. So you give it 10 times the amount of data, it costs you more than- costs you a hundred times more to build it, but you're getting maybe 5% extra benefit. So I think it's diminishing returns.

So it- that's why I think people talk about it being an off-ramp in terms of the trajectory, or I would call it a dead end as far as AGI is concerned.

- [Neil] Just to clarify for everybody, that doesn't mean that like generative AI doesn't have value. It's just not a stepping stone towards AGI, general intelligence.

- [Peter] Correct. It's tremendous value. Tremendous value. The technology is phenomenal in so many ways. It helps programmers. I use it every day to find stuff. It's just, it has to be a human in the loop.

- [Ryan] Let me ask you a question, we were talking about the applications of certain businesses, especially for the consumer side or the customer side of things. When you look at it the other way, and I was talking about the productivity side for individuals, like salespeople and such, but when a company is looking to bring in a solution like this into their business, what are the benefits that they see on the business side of this?

I'm assuming there is an element of it helps with training. I'm assuming there is a monetary or at least bottom line benefit of being able to do things more efficiently, do things better maybe without requiring as many people to do this. But just out of curiosity, what have you seen from people that have adopted?

- [Peter] Yeah, let me talk about chatbots in the call center first, because that's what I'm most familiar with. Call center operators have had a really torrid time with COVID, that was like the first hit. And since then it's become really hard to find people.

I think the average length is now something like six weeks. There are a lot of people that will just say, okay, I need the money. Let me go and work in a call center. And then- some of them you can work from home. But very quickly, they'll find this is not for me, I don't like to do this. So they're having a really hard time finding people, keeping people, and training them to do a good job. So the quality is a problem. Keeping people is a problem. So that- there's tremendous pressure on call centers and you- we hear that all the time. It's getting worse and worse.

So alleviating that pressure on the one hand. But it used to be that you talk about IVR and chatbots as containment. Can we prevent the call from going to a human because it's expensive. So it was second best. But with the right technology, you actually turn that on its head where the chatbot can provide a much better service than the human.

For a number of reasons. A) Because the humans may not be that good. We've all had experiences where we talk to a human, but we're still not having a lot of joy getting done what we want done. But there's no wait time. It's available 24/7. It remembers your previous call. It remembers you. If you called two hours ago because your router went down and the first thing it said, well, try rebooting it, okay. You tried rebooting it. You call again two hours later. It's not gonna say try rebooting your router. It will say did that help? So which a human would not be able to do.

No wait time, 24/7, and potentially on a good, well-designed system, the experience will be much, much better. You'll get your results. It has access to all of the information instantaneously. Where's my order? What product is available, and or how do I fix something?

How do I reboot, what button do I press to reboot, and so on. Tremendous savings. Typically, people- the cost of automating it with a chatbot would be something like a quarter to a third of what a human costs. Tremendous savings. So good for the bottom line, and good for CSAT for customer experience scores, that you actually improved them.

- [Ryan] Let me ask you also, when it comes to the trust element of it for people that are interacting with these chatbots, obviously you mentioned earlier, and I think we've all done this, is if we interact with either on the phone or with a chatbot that we know is not very sophisticated, the first- all we wanna do is get to a person because we feel like we can connect with them, we can explain our problem more clearly.

It's gonna be more easily understood, and we're gonna get to a solution faster. How have you seen or what have you seen happen as far as what's required in order to develop that level of trust with the consumer for them to really appreciate and find value in that experience with a chatbot as opposed to just trying to get to a human as quickly as possible?

- [Peter] Very good question and something that we fight because automation has such a bad reputation for good reason because a lot of them really are horrible. So we often have long debates with our customers, the enterprises that wants to implement it, where you have legal and their QA people, their marketing people, tell us you have to say, you have to start off by saying your business is very important to us.

Please listen carefully because the options may have changed. Your call may be recorded for quality assurance and blah, blah, blah, blah, blah. Or and here are the menu options or whatever. I can do this for you, I can do that for you. Don't, don't, don't, don't, just whatever the minimum is that you have to say legally, get the customer to get to- to move forward in the conversation. If you can engage them, and they feel they're moving forward. You basically craft the conversation in a way. You don't ask stupid questions. You acknowledge that you're moving forward, and you're giving them results as quickly as possible. It's basically the sort of human engineering, human factor engineering that goes into it to earn that trust very quickly.

On the other hand, we also very much encourage our customers that if you say operator, it says would you like to talk to an operator? And yes, you transfer, and you transfer whatever knowledge we have to the operator to make it a better experience.

- [Neil] There's an interesting shift going on with like chatbots, like particularly in customer service, where I'm starting to see more and more enterprises coming in saying rather than train the chatbot on the average, we have enough data, we can look at who our best performers are, and ones that connect the best with the customers and actually use that as our training basis.

Can we train the chatbot to be like our best CSRs, for example, and they're seeing some interesting resonance from that. But this goes into something I always talk about is that when we talk about like technology, especially AI, we always focus on automation, Peter, rather than innovation, and I'm not saying there's anything wrong with automation, it's just we're not tapping into the full value and capability of the toolset.

Do you see some of that, like kind of everyone thinks about automation rather than innovation when it comes to chatbots or digital assistants.

- [Peter] You hit onto an important point there. A lot of the success of these implementations is to do a good implementation. To really understand do you know what all the business rules are? Do you have access to all the APIs? Which has nothing to do with the underlying technology, to do a really good job.

Now in terms of just doing what you're best CSRs do. The problem I have with that, you definitely want to understand and learn from them, but if you just feed this into a statistical model, then you're losing the hyper-personalization because you're then getting back to the average.

You're gonna talk to everybody in the average way. You're not going to adjust the conversation to the particular customer. So, I think one needs to balance that off. With the technology that we have, we are literally getting 90% self-service, that people find that it's working for them.

They're getting the answers. They can change their orders or place an order or do whatever they want to them. And obviously I'm not saying that every application, level three tech support, no. We're not gonna have that kind of containment.

- [Nikolai] Something I'm curious about is you talked about- I think we went over briefly about like how your chatbot is modeling the brain more than your average chatbot. Could you go into more detail about the specific things that you're doing and what's being modeled and how it's being modeled?

- [Peter] Yes. There's an important distinction between modeling the brain and modeling the mind. So we are not trying to reverse engineer the brain. We really want to understand mental process. In fact, a third of our employees, we call them AI psychologists. They're not engineers. They basically have a cognitive psychology background or linguistics or both?

So it's really understanding thought processes and understanding learning and things of that nature. Now, on the technical side, our approach is called the cognitive architecture. DARPA calls us the third wave of AI. Just very quickly, first wave was what's now called good old fashioned AI.

Those were basically expert systems, logic based systems. The second wave is the tsunami we are riding right now, which is basically statistical systems, neural networks, statistical deep learning, machine learning, GPT, all fall into the second wave. And the third wave are cognitive systems. Systems modeled on the mind, on mental processes. How do we understand things. Now, specifically how we implement our cognitive architecture is there's a backbone of a knowledge graph. So all the knowledge and skills that the system has is encoded in this scrutable knowledge graph. So you can actually pinpoint exactly what knowledge it has.

So if it doesn't give the right answer, you can pinpoint it. It's not a black box. And we had to design, one of the key innovations is that we had to design a knowledge graph that is fast enough for this purpose. And our knowledge graph is literally a hundred times faster than any commercially available graph database.

And that's a key to it because we've spoken to large research centers that have tried to use knowledge graphs within a conversational AI and they can't get it to work for that reason. If you need, yeah, and then all of the components that we have, the passing, the understanding of the sentence is deeply integrated into this knowledge growth.

So you have access to short-term memory, long-term memory. So if you're talking about Bob, then it would know all the Bobs, but also contextually, which Bob might you be talking about, and it could be your dog, so having all of the components, the reasoning, same with our reasoning engine or the reasoning components of our system, are deeply integrated into systems so that during passing we can do the reasoning and say, does this interpretation make more sense than some other interpretation?

So it's having all of these cognitive components that are needed like short-term, long-term memory, reasoning, context, goal directed behavior, and so on. Having those all deeply integrated with each other and with the underlying knowledge that the system has and can accumulate.

- [Ryan] What is- we talked about a lot of really interesting elements of this and the applications of it, the benefits and so forth. Where do you see all this going? This is obviously a long-term kind of thing that we're working on, we're working towards. Where- what is the next evolution of these enterprise digital assistants, these chatbots?

Like where does it go from here from where we've already been?

- [Peter] Well, first of all, very few companies are anywhere near state of the art. So there's a huge opportunity for existing enterprises to really get the right kind of technology. There's a huge gap. So there's tremendous development. Now, we are continually increasing the IQ and capability of our system.

As I say, there are just tremendous opportunities. Most chatbots and IVRs that are actually out there implemented are awful, and basically, hey, let them- get them to be less awful, and ideally give them a brain. But looking forward, with the excitement we have with ChatGPT and the money flowing into AI, we are actually right now looking to substantially increase our R&D or development, it's not really so much research because we've done most of the research we need to do. It's really more developing our system and implementing smarter versions of it. We are looking to raise a substantial amount of money to substantially increase our development cycle.

And people are now really seriously talking about let's get to human level AI. So that we can utilize them, whether it's for cancer research or to have a level four tech support person or whatever it requires. The world needs so much intelligence to help us solve our problems, and there aren't enough humans to really work with all the things we want to do and need to do.

- [Ryan] Absolutely. Neil, any last words?

- [Neil] I think it's just been fascinating. I think this whole journey from chatbot to assistant to copilot is incredible. Different value for businesses across the way, but it's really interesting to see Peter, based looking on some of the work that you're doing and for other people, how we're ultimately trying to get to that copilot and that we're literally scratching the surface of the iceberg of the value we can actually create for people and for business.

- [Ryan] Absolutely. Peter, thank you so much for taking the time. For audience out there who wants to potentially touch base or follow up or learn more about what you all have going on, what's the best way they can do that?

- [Peter] Yeah, aigo.ai, a i g o dot ai is our website. And there I have a link to a lot of articles, which are also on medium.com under my name, Peter Voss, about everything from free will and ethics and AGI and third wave of AI and whatnot is there. And you can also reach me on Twitter and- or just [email protected].

- [Ryan] Sounds good. Well Peter, thank you so much again for taking the time. Really appreciate it. Excited to get this out to our audience.

- [Peter] Great. Yeah. Thank you. Good questions. I enjoyed this.