Latest / Machine Dreams / Why AI Companies Are Talking About Consciousness and What It Means for Business
Transcript
- speaker-1: A few years ago, if you asked whether a computer could think or feel, most people would probably have treated it as science fiction, or perhaps even thought that you were crazy, right? But here we are today having serious conversations about whether AI systems might be conscious, whether they can experience emotions, and whether they can understand right or wrong and wrong. and I think what's especially interesting is that we haven't even agreed on what consciousness really means. yet here we are, as I said, already building machines that can speak, that can reason or mimic reason and behave in ways that some of you know these questions increasingly get difficult to ignore. And that's a really big topic today. but where else It's best to learn about this, if not machine dreams, where we have honest conversations with the people building and shaping AI about what works and what doesn't and what comes next. My name is Kolawole Samuel Adebai. I'm a technology analyst. I'm working at the intersection of AI, cybersecurity business and society. And of course, as always, I'm joined by the one and only wonderful selectable. I I think Leo particularly likes that word. Delectable. speaker-2: Hey Sam and hey everyone, I'm Leah Stern, global communications strategist and former venture capitalist. It's not that I don't like that word, it's just I think you can you can give me some different adjectives each episode. Doesn't always have to be the same, no. speaker-1: I have to get into my bag of words and you know, just do some bit of search and speaker-0: Yeah, speaker-2: Do for a living. You're you're a storyteller and a writer, yeah? Yeah. speaker-1: Yeah. speaker-2: Or maybe that's the best you got for me, I don't know. speaker-1: I I I just think that that particularly encapsulates what you represent, you know, but there's there's a lot more. Admirable, immuleable. speaker-0: Ha ha. speaker-2: There we go, there we go. The superlatives bring them out. Well, today we're joined by Andy Tharae. He's founder and principal at the Field CTO. It's an independent technology advisory firm that helps organizations make better decisions about AI, emerging technologies, and enterprise technology strategy. Andy brings more than 25 years of experience in enterprise technology. having held senior roles at companies including IBM, Intel, and Oracle. Over the years he's worked with business and technology leaders to understand how emerging tech can solve real world problems and what their adoption means for organizations and society. Andy, it's a real pleasure to have you. Welcome to Machine Drains. speaker-0: Thanks for having me, I appreciate it. Well if he if he used all those objectives on you, he won't be left with any for me. I'm just kidding. Happy to be on this podcast. I I like I said, I I listen to your podcast all the time. Love it. You guys had a lot of fun recording it. And then when Sam said, Hey, you oughta be on the show, I'm like, Yeah, just say when. Right? speaker-2: Wow, that's awesome. Yeah, it's a labor of love and we do enjoy it. And we're so grateful to have people like you experts sharing your insights. Thanks so much. speaker-1: Great, great. Thank once again, Andy. There's so much that we can talk about and there's a whole lot of issues in the space right now when you think about AI and enterprise technology and and all of that. But I I think a more interesting conversation as as we were saying the the last time is about this idea that AI can start to think, you know, or that AI might be conscious right now. And as I said, there is no general agreement on what consciousness. Even means in you know the philosophical cycles or circles, and then you bring that into AI. So the first place I think that I w want us to begin this conversation is what you think AI consciousness is. Should we even be saying that right now? Is that a thing? Could we get to a point where AI actually becomes conscious? speaker-0: Man, I gotta take a deep breath. Dude, why are you doing this to me? You you told me there's gonna be a very casual conversation about AI and you're starting off with this AI conscious conversation. my God. Yeah, speaker-1: Sorry about that. speaker-0: it's okay, so that's a tough question to answer. So funny enough, matter of fact, I just about dropped a podcast on AI consciousness, and I had two folks on my podcast. that are actually really good in that discussing that. th we we were set out to do a thirty minute podcast. We landed up recording over an hour talking about his AI consciousness. So I won't be doing justice, you know, by talking about that, but I'll try to frame that, right? By the way, those two guys are really good with not only AI, but also with the the Eastern Vedantic philosophy. So they they were talking about we were discussing about what is the difference between intelligence, awareness self-conscious the whole night. But for the sake of this conversation, now look, it has come to a level that anthropic is somehow trying to define their Alamus conscious. You might have seen that latest like a few weeks ago, they had Swami Saratrian, the one of the the well regarded Hindu monks who talks about this, you know, the consciousness awareness from a Hindu philosophy mythology. He was in there visiting with the Mandra NDA and they had a lot of discussions. Anthropic also had a conversation with the with a very conservative rabbi, Mois Nevon, and then also had about fifteen of Catholic and Protestant leaders who were also heavily engaged with the Pope or Vatican, trying to, you know, understand or push or work with to basically define to conscious consciousness or waking and they even ta started talking about existential threat to humanity kind of thing, right? But having said all that, let's go back to the original question of is AI conscious right now? it's hard to answer because then you have to define what is consciousness. Here's here's an example that we discussed in my podcast. If for example, you know, if you have a robot, or even a self driving car for that matter, the self driving car need to be aware of its surroundings. That awareness gets created by sensors. pretty similar to our sensory capabilities, you know, robots or or AI can also have sensors. and that feeds the information in and it'll make a decision based on the sensor it has. So just because car decides to stop immediately when somebody jumps in the road or it sees an object in the front, doesn't mean that it is conscious. It's aware of its surroundings. It has an awareness, but That awareness is about it just its surroundings and then how it has to act. So basically it gets simulated. So the shortest possible answer I can give for this would be machines can act in a simulated awareness, at least as of today. again, AI surprises me every day. So I wouldn't be surprised if models in the future, maybe a year or so from the road, can be fully self-aware. But as of today, my viewpoint, Andy's viewpoint would be yes, it is aware of its surroundings, but it's not exactly self-aware. And then it can simulate reactions. And another point that one of my guests made in my podcast, which was an amazing point, is that if somebody throws a hot water on a dog or a cat, it's going to react to it quickly because it's it's conscious, aware of the pain and because of the sensories. it reacts in a certain way even though nobody has taught them to. And because of the pain and because of the harm it causes to it. But if you throw a hot water on a robot, would the robot behave the same way unless you have trained them what to do when somebody pours a hot water on you, right? So that's that's that intelligence, knowledge and the awareness and the consciousness even can be simulated in a AI, in in a AI model or or robot. But as of today, you need to train for it to do. Otherwise when when the whole unknown unknown happens, it could go nuts. That's where AI failed, by the way. When you do unknown unnounced one AI, it it doesn't know what to do. That still is the case. speaker-2: You know, that's quite interesting, your perspective. And you really bring a unique perspective because you were on the inside at big multinational corporations like IBM, Intel, and Oracle. And now you've gone to the other side, some would say the dark side, you're independent, you're advising probably smaller companies, you know, on their technology. So I'm curious, are there things you saw inside the big machines that surprised you, that really gave you an insight that now you bring to smaller companies and independent clients. What's going on in these places that maybe people don't realize in terms of using AI? speaker-0: Hmm. That's an interesting question. speaker-2: You can talk about it. Perhaps there's NDAs involved. I don't know. speaker-0: yeah, there are NDAs so I can name names, right? on either side of the fence, whether it's a big company or small company. But but one thing I could tell you, there are some big companies that are doing full research on on a pure research basis, right? But most of the larger companies are one because as I say, the big machine takes time to move, rotate, right? So it takes time for it to catch up. than some of the innovative startups. That's why innovative startups can go quickly and then do things so fast, especially with the wide coding nowadays. You can code anything out in a matter of days, or a matter of hours for that matter, right? But with the big machines, one of the things I constantly notice is that they're always about solving a customer problem. That because you know they have been selling to the industry for many, many years, they have solutions in the industry that could do things. They're always the first thing they think of is that, you know, how do I solve a certain problem for a certain customer, right? And that's the mindset they come with, even though they have a research wing. For example, IBM is is a classic case for that. They have a complete research wing. They do a ton of research that I know from my days or even now, that I in constant start touch with a lot of the folks that I did research. they do a lot of research, but then they figure out how to convert that into solve a customer problem. And most of the the startups when they come in Yeah, always about technology differentiation. Majority of them. I I don't want to, you know, typecast. But there are some, you know, that are worried about customer problems and how do we solve that, but mostly people come and say that I gotta shiny new button. And this is where when I'm advising the smaller companies, it's hard to explain to them or digest and and most of them don't even understand when I say that. You gotta worry about how to solve a customer problem. You know, don't don't go and pistol them saying that, you know, you have a shiny shiny new button. You know, don't tell them how that can be used in your environment to solve a certain problem. So it's the same message. It just, you know, you gotta change it in a different way. I don't know if I answered your question or just speaker-2: And just, you know, why did you decide to break out? was it the bureaucracy? Was it the big machine? What was it? speaker-0: Wha what is the question again? W why did it what? speaker-2: Why did you decide to go and do your own thing? That's a personal question. You don't have to answer. speaker-0: Vario various reasons. the predominant one being somewhat of a having a independence opinion and freedom to do what you want to do, right? Creative freedom. again, not necessarily particularly in any companies. Some companies are very open and let you do the things you want to do because, you know, they think you are a true thought leader and they wanna bring your thought leadership. To help their customers solving the problems. Some companies are like, you know, nah, you gotta play by our rules, everything has to go to our comms. We gotta approve everything that you write or speak or put outside, right? So it depends. You know, so again, obviously it is what it is and and they all have their own brands, so they need to maintain certain decorum per se. That's fine, you know. You y you gotta do what you gotta do. If it works out you'll stay, and if it doesn't work out You move on to do your own stuff. That's what I am now. speaker-1: Speaking of independence and creative freedom, I really want to get your opinion, Andy's opinion, on this. and I'm sure that a lot of our listeners and viewers also would have a lot they can take away from this. And I'm going back to that conversation about consciousness. I know it's it's it's very it's it's a slippery slope of salt. But my question is because it seems and I've read a lot of articles around this, a lot of pieces around this that seem to imply that, you know, it looks like Anthropic is running essentially an im a global influence campaign to sort of push the narrative that, you know, AI is now conscious or might be conscious. And they've held, you know, confidential seminars, private meetings with, you know, religious leaders, moral thought leaders and and and philosophers and and the likes. my question is, why is it so important for AI to be conscious? Like, what does that change? Really? If AI let's assume that AI were conscious, I mean speaker-0: That's another tough question, man. Another tough question. By the way, during your question, you made a statement saying that, you know, the consciousness can be a slippery slope. no, it can be. It can only be when if you're drunk. Otherwise it shouldn't be a slippery slope, right? Both consciousness and awareness should be it's straightforward. You should be able to speaker-1: Conversation now can be, you know, different sides of the divide, but I I understand what you meant. Yeah. speaker-0: Yeah. No, the point is so again, to answer your question, that's actually a loaded question. Do you want a controversial take or do you want a just a meh take? What do you want? speaker-1: Give me the controversial one. speaker-0: okay. You're you're you're you're s you guys are set up to put me in trouble, aren't you? With all your light of question. Alright, so my controversial take is this. Again, this is Andy's pure opinion. A and his personal opinion. You know, I mean you can quote me on that, obviously it's my opinion. But this is, you know, just what I feel, right? one of the things that comes across strong, I'm not talking about just Anthropic in particular. All the AI companies all of a sudden in this bandwagon for the last what, maybe maybe just a year or two. Before that that never used to be the case. They start talking about morality, about, you know, teaching LM human values and and and also about, you know, AI refusing to do something unethical in the whole thing, yes. But my view could be controversial, is that there is a specific reason I think all these AI companies are all of a sudden, you know, getting up on their moral compass and then trying to grow consciousness. Because in my view, if a AI model can be proven to act on its own, because when when you have the AI model, it it makes certain decisions. Sometimes they claim that the decisions are made, that's not within its bounds. In other words, it's it's not only obviously machines make a probabilistic decisions, right? If you ask the machine the same question 100 times, it won't get the same answer 100 times. So not being that deterministic, it always makes probabilistic decisions. But the the AI companies are trying to claim that the machines are starting to make decisions on its own. So what I'm thinking is if they deploy that outside of the lab in a situation, if they land up making a certain decision that Becomes a liability issue or causes a major harm, by assigning this consciousness to the machine, they can walk away from it, saying that we didn't do it. We created a model. The model on your lab thought by itself. Therefore, that mo that model is responsible, not as creators. That's my controversial view. So they're controversial view. But that's what I think. Because right now, There is also this question of liability that always comes up when I work with the model users, the enterprise users, when they use it. This is the first question they ask. So I let's say I use the AI to make a certain decision, whether it's a deny a healthcare claim or or, you know, deny a job for somebody, whatever that may be, right? Which involves humans. Right? So when you make a decision, if the decision is controversial. If you can't prove that the humans would have arrived at the same decision, you should have a data points to prove that a machine a human would have made the same decision. Those courts have you know ruled on this multiple times. If a machine makes a decision beyond a reasonable doubt, you have to prove. With the same data points, you have to prove that a human would have made the same exact decision hundred out of a hundred times. Right? But if it makes a count speaker-2: Sorry to interrupt, but if you're saying that machines are making decisions on their own, we're already asking AI systems to make recommendations and decisions that affect people's lives. But can AI genuinely understand right and wrong? And will it always be following some version of the values and instructions we're giving it? And where are these judgments and ideas coming from? Are they shaped by culture, upbringing, experience? Like every everyone has a different my opinion on that. So how how dangerous is this really and what can you believe and not believe and trust? speaker-0: Well, so man, that's a very complicated loaded question. You added like four questions into one. I know. speaker-2: Those are all of them. speaker-0: So so look, you asked about this morality and the AI make a right or wrong decision. But that in itself is a very controversial question, right? How do you know what is right across the universe? There are certain things, right? don't harm the, you know, children, women, kids or or humans in general. It's a universal theme kind of thing. But again, the matter of fact is the truth is that across the board, there can be differences. For example, I remember way back when somebody told me a a patriot in a country could be a terrorist in another country. Right? So how do you decide, you know, who is the hero, right? So the same way that the the the you know When you talk about the morality and the truth and doing the right thing changes. What is the right thing? It changes based on religion, geolocation, culture. Could it, I mean, there are some common things across the board, but also variables. The the variation could be huge or small based on things. So, how do you teach a machine, a neural network, to have a universal truth, saying that that's the only thing, right? But Coming back to your point about Anthropic trying all these things now, I in my view, I don't think Anthropic is talking to religious leaders because they're trying to grow a soul into quad. I think they are doing it because the software engineers and the model builders have realized that you cannot write a deterministic if then code to cover the the infinite gray areas of the human ethics, right? I mean human mind and human ethics is a complicated problem. For example, you all know that the famous MIT trolley problem, right? There's no easy solution to it. It it depends on the situation and and, you know, on what you decide. That there is a multiple variation to that answer. How do you code that in using a normal coding and teach a model that I mean if you already pre taught the model that particular problem, it can solve that by saying that, okay. I follow the builder's views, this is the route I need to take, this is a decision I have. But if it encounters unknown, unknown, how could you make it make a decision saying that this is the right decision? Going back to my original point, how do you know what is wrong or what is right or and depends on whose viewpoint. Is it the model builder's viewpoint? What they think right or wrong? They could have a they could be extremely right or extremely left. So how do you decide that, right? Or based on their their religion or a culture of values. How does model decide between right and wrong? That's where the confusion comes into play, right? But it it is also when I A refuse to do something unethical, just to be clear, it is not it's not consulting or growing a conscious to avoid that. It is just simply calculating and refusing a prompt to answer that, which kind of yields a higher mathematical reward based on its training weights. Right? It's based on a you know oral HF or constitutional AI and what have you. So it's essentially to an extent, it's an advanced calculator doing a statistical ethics, that's what it is. Right. But again, I'm I'm sticking to my controversial take that I had originally. speaker-1: interesting conversation so far, Andy. and I bet that we could spend hours and hours talking about this, but we've we only have a few minutes left. So and my last question, and and I would like you to answer in one minute. It's gonna be really difficult to answer in one minute, but I I I'm hoping that you you can pull it off is you speaker-0: You speaker-1: work with a lot of enterprises. You've been working with them for many, many, many years. The AI revolution or the AI industry is one as you said before that is that moves really, really fast. You know, there's one thing today and tomorrow, and there's another thing. And it seems like there's so much that enterprises have to grapple with, you know, questions around liability, regulations, ethics, you know. how do we make sure that our models aren't hallucinating? How how do we get like the right output from, you know, the way that our employees are using AI models. And so it seems to me that the average enterprise leader today has so much problems to deal with. challenges can be quite exciting, but you know, they can also be very tough. so my question would be, what would you say to enterprise leaders who are facing all of these challenges on multiple fronts? And still looking at and thinking about how do we get the best return on our investments? How do we get the best value from the way we're deploying or building models? What would you say to enterprise leaders in that sort of situation? speaker-0: And you only got a minute, come man. This is a whole day conversation, if not multiple days. So here's a short answer that that, you know, when they ask what is your top three that I give this. one, don't you know, there's a famous saying that, you know, don't get a hammer and then try to find every nail. Don't try to choose a model, choose something, try to solve something else. I w I highly recommend this is where I've seen it's successful. Work backwards. Find what the problem you're trying to solve, specifically a customer problem or or a problem that is not solvable using today's technology, today's processes, what have you. And then figure out a reverse method to go back and say that, okay, what model can I use? what can I do to fix the problem? That's one I always recommend. The second thing that people don't seem to realize is LLM is just a reasoning engine. So it reasons, gives some probabilistic answers. So your harness and the safety and compliance must live outside the model in a deterministic control plane. That AI cannot hallucinate, AI cannot do things in in in the way around and do things. To quote the the famous statement that you know, I I think it was Ronald Reagan who said that trust but verify. When it comes to AI, I would say don't trust, always verify, right? And then and then the last one is that the third point because you asked for three. you gotta stop relying on the model to police itself. asking a probabilistic model to enforce a strict compliance is is is actually architecturally impossible. So don't ask it to do that. Ask use AI for the purpose it is built. It's a cognitive intelligence decision making reasoning engine. That's what it is. Use it for that purpose. And then do the other harness around it. It's not just one or the other. You need to have both of it in place. Otherwise you're just wasting your time. You're sitting on a next liability lawsuit. speaker-1: Thank you very much, Andy. Such a great way to wrap up the conversation. Very fantastic, you know, answers to to that multifaceted question. I appreciate you being on the show today. This has been quite engaging. And I'm sure a lot of our listeners and viewers would agree to the same. speaker-0: Appreciate it. Thanks for having me. I mean, like I said, this is the the most fun conversation and that's why, you know, I I watched a lot of your podcasts and when I asked, I'm like, Yes, I need to be on it. By the way, I if if you if if I may, I'll give a plug for my podcast too. Check out check out either my YouTube channel or the fuel cto dot com or follow me on LinkedIn. I have I have two podcasts that I have. One is called Tick Candid, where I have candid conversations on technology with the enterprise leaders. The other one is AI on AI, which is still being worked out, but you know, it'll start. And when it drops, the first episode drops, you would know why that particular name is AI on AI is a name. So thanks for having me. speaker-2: Very cool. Thank you, Andy. And thanks to everyone watching and listening. More conversations with people building, questioning, and shaping what's next in AI are on the way. We'll see you next time.