Azeem Azhar and Gerd Leonhard: Artificial Intelligence – heaven or hell?

Hello, this is Gerd Leonhard Futurist, CEO of the Futures Agency. Today in a podcast conversation with Azim Azhar. Azim is also a member of the Futures Agency and the two of us are going to tackle the topic of artificial intelligence. Azeem, hello. And just say a few things about yourself.

Absolutely. Good. Good morning to you. So yes, Azeem Azar here. Really pleased to be speaking with you. Amongst other things, I write a very popular newsletter on the impact of AI and other exponential technologies on society. It's called Exponential View, you can just find it on Google. And I am also an early stage investor looking for interesting founders doing smart things with technology for major goals. And I advise a number of AI companies and AI organizations on where the world is going over the coming years.

Great.

Yeah, I think the two of us have met in sort of the context.

Of speaking about AI and the future and that stuff. So we figured we would do a podcast to share our views on this and we're also going to go out as a duo. So if you want to book a band for your gig, Azim and me, we'll pick up our instruments, our virtual instruments and play some music together. So let's dive into this. What is your general view, Azim? We're gonna just ping pong on this back and forth to artificial intelligence. The number one question I get all the time, is it really as dangerous as it sounds in the media? And of course, just a side note on that, we all know that of course the media is heavily distorting what AI is and what it can do. But what's your view on this?

Yeah, it's the number one question that I get. So much so that often when I'm speaking I by saying we're not going to talk about killer robots or the Singularity and inevitably you still get that question. And the way, the way I look at this is that there's a set of really quite important and big existential questions about what happens if these possible futures of super intelligent machines finally does arrive. And those questions I think are best dealt with by the philosophers and their groups like the Future of Humanity Institute and the Machine Intelligence Research Institute. Look at those questions and they aren't the sorts of things that we need to worry about over the next 5, 10, 15 years. Over the next 5, 10, 15 Years. I think the questions that we really have to ask are about how AI systems, particularly when they get connected with profit motivated capitalism, can construct unsatisfactory, negative or unethical outcomes for people's daily lives and as these AI systems become a key part of the interface, the way in which you or I or anybody listening to this gets access to train tickets or news and information or buying products, as AI becomes the kind of core layer that we have to walk through, how can poorly designed, poorly implemented AI actually result in negative consequences? So when I think of the threats, I really think about the things that are happening today and will happen over the next five to 10 years, which are related to some of the things I've just discussed. Rather than, you know, humans being superseded by machines, which is a question that I think is slightly further away.

Yeah, maybe in that context we should first zoom back to the definition of AI. And because that's also very confusing fact.

I always say in my speeches that most people think of AI, they think of X machine and transcendence and so on. And it's very confusing for it's one of the most confusing terms. In fact, I always say that artificial intelligence is actually quite a bad word.

Because it's neither artificial normal nor is.

It really intelligent in the sense of artificial. Well, every machine is artificial, so that's hardly a distinguishing fact. And intelligence, well, human intelligence, as many researchers have shown, consists out of eight or 10 different types of intelligence, including emotional, social, and so on. And the machine only has really one intelligence right now, which is computing right intellectual. So let's define artificial intelligence. I'll kick this off and then you can say what you think of that. First, I think AI. I like the definition by Demis Hassabis from DeepMind, who says that AI is when data and information is used by a machine to derive knowledge so that patterns can be recognized and context can be created and so on. And then I would say also we have to, of course, look at deep learning and machine learning, that the kind of two subsections of AI, really, because most progress in the last few years has been in this whole arena of machine learning. And that is really the stuff that we see.

So from my point of view, that.

Is a quite good definition, turning things into knowledge. I also like to say in my speeches that I usually differ between Intelligent Assistance IA and that's 98% of what we see today, smart software.

And then AI, which is the next.

Iteration, actually contemplating cognitive machines. And then AGI, which is the super intelligence, the artificial general intelligence, which in my view has to be regulated similar to other tremendous powers like nuclear energy and so on. So that's kind of my definition. And what's your take there?

So I love your point about everything being artificial. And also your observation that we're not very clear about what we mean by intelligence. The way that I look at the question of intelligence is that I think about it as the capability or the capacity for an entity to express a goal and activate resources towards achieving that goal. So what distinguishes a simple process from something that is intelligent is the ability to take stimulus from the environment, to process that in a kind of information context, then to get hold of some resources to achieve the goal that you want to achieve. And it's quite broad in that sense. And I think what it does is it reminds us that that is effectively what distinguishes humans from amoeba, or humans from inanimate objects, or indeed chimpanzees from inanimate objects.

That's a good point. You know, there was a guy the other day, I read, I forgot who it was who said that AI has to do with the capacity to increase capacity. And I like that because it's kind of like that's what humans do all the time. And that's certainly something we should think twice about, giving that to machines beyond a certain level. Of course, we give machines data to increase capacity, but would we give the machines unlimited access to processing power to increase the capacity? Is that really a good idea?

Well, that's a good question. But, you know, you identified that there were lots of different types of intelligence manifests itself in different ways. Emotional intelligence, cultural intelligence, and so on. And I think we also need to recognize that we have started to already externalize our capacities and capabilities beyond our individual human bodies. There's a famous experiment where a designer called Thomas Thwaites tries to build a toaster. And he goes off and says, look, I'm going to find the cheapest toaster I can find. He finds this five Euro toaster which you can just buy from like a catalogue. And he says, I'm going to make this toaster myself. I'm going to make the plastic, I'm going to make the iron by digging out iron ore. And it takes him a couple of years to do this. The thing looks really ugly and catches fire immediately when he sets it going. And it cost him thousands and thousands of dollars to build this thing. Now the question is, how was it were we able to actually deliver a working toaster for $4 where this or $5 where this amazing human couldn't deliver a working one for 20,000 working on their own? And the reason is that, you know, we, through our ability to collaborate, through our ability to create group intelligence and group capabilities, we're actually able to deliver complicated things like toasters or even more complicated things like iPhones at really reasonable costs that are well beyond the capabilities of any individual human to actually, you know, deliver that. And I think when we think about types of intelligence, somehow we have to capture what is this group intelligence, this network intelligence, this intelligence that lives outside of us individually, which we are now suggesting we're going to bring, you know, a new class of actor in which is these brilliant machines.

Yeah, I think that, you know, a lot of arguments have been made about human intelligence versus machine intelligence. My view is that, of course, machines will eventually be able to simulate human intelligence, like social intelligence, emotional intelligence, musical intelligence. And I think it was Nick Cave who said the other day that a machine is perfectly capable of writing a good song, but it will never write a great song because it doesn't have the guts. And that was a very good quote. You know, what is missing in machine intelligence is the context is the sentience is the human agency. And that is also a certain kind of intelligence that we have. And I think that part of my conversations, I have a lot with people about AI is that people are confusing an intelligent machine, parenthesis, Like IBM Watson, debater. Right. The machine that can debate. They're confusing that machine with the person because it sounds so much like a person. But in the end, IBM Watson is pulling up a huge amount of preset responses and possible jokes and combining them at the speed of light, so to speak, kind of like a human. But it's still missing the actual awareness of what it says. Right. And that to me is one of the key differences that we shouldn't confuse. And also that I don't think we should, even if we could, I don't think we should let machines advance to.

That level of where they can acquire.

This, to the level of us getting really confused as to what is what.

Well, that. Wow, there's so much packed in there. I mean, I. First of all, I completely agree with you about IBM Watson. The IBM Watson is an automaton. It is like a circus puppet that is kind of controlled by some mechanical pulley system. It's smarter than that, but it's more sophisticated than that. But it doesn't exhibit the things that we would think of as intelligence at all. And I think we have to start to think of the machine learning tools and the natural language processing tools that exist today as exactly that. They are like tools and tools shape how humans can think. They shape how humans can create, they shape how humans can express themselves. And you, only you can do this in a very simple way, by going to an art gallery and looking at, you know, pre Renaissance paintings and Renaissance paintings and then look at the art that we have today and see how much more, how many more different ways we have of expressing ourselves now that we have got better tools. A lot of modern art today has fluorescent pink and fluorescent neon and fluorescent green colors. These were not colors that Michelangelo or Leonardo da Vinci could use because the technology to deliver them didn't exist. And I think we're in danger because it's in the interests of technology marketers of mistaking what is a better hammer or a better spreadsheet or a new colour of paint for something that could replace humans. I don't think that's the case. I think that the technologies that we have today actually allow us to express ourselves much, much more capably than we were able to 10 or 20 or 30 years ago. So we have more degrees of freedom, more places that we can explore and we can kind of run some examples. But then I think you've made a very good challenge, which is, you've made this challenge, which is to say, what if we can't distinguish what's going on and what if the machines able to do everything that humans can do and more, do we feel comfortable with that?

Well, I think the key question here really is I think that technology companies that offer this, like IBM and others and Microsoft, they should stop talking about thinking, you know, thinking machines, because we immediately think of ourselves as thinking and.

Then we say, oh, this machine can think. Oh, damn. You know, they're going to take over. And it's actually not that they don't think like we do at all, because how can you, how can you think like a human when you don't exist, when you can't suffer, right? When you, when you don't have emotions? It's, you know, I think some of the most famous psychotherapists in the world have said that we don't think with the brain, we think with the body. We are much more holistic than that. And if we are a machine, as Ray Kurzweil is saying, or others, if we were to actually be a machine, I think we're infinitely complicated machine. That will take at least 50 years to figure out if that is the case, which I don't believe. But I think this is one of the key points about artificial intelligence. Let's go back to saying, let them do the work, the monkey work, the routine, the things, the heavy lifting that humans are not really fit for and that we can drop, as I like to say, at the end of routine. Right, Let them do that. We don't want them to be like a human. We don't want them to be a partner or robot or you know, a lifelong partner or an advisor for government. We, we want that to be humans because of who we are. And I think this is one of the key facts when that in the debate keeps coming up from people saying like, you know, we really are just all machines anyway. So, you know, if we connect with AI, that's our natural destination.

Yeah, well, that one.

No, I know. You know, you throw out 50 years as a time frame. I mean, maybe it's 50, maybe it's 500, right? But it doesn't feel like it's five. Well, what we do know, we do know certain things. We do know that as we start to apply the process of science and inspect what's going on inside the human, we are finding more and more natural rather than supernatural explanations for why things are what they are. And I think there's a really interesting boundary here where we can start to look at AI based systems in mental health and psychotherapeutic conditions. And the reason I choose that one is because I think it's, it's kind of trivial to describe AI systems that can recognize whether there's a picture of a car shows damage to the bumper. Right? That's fine. That is just like a spreadsheet. That is just a tool. But what I have found interesting was that there is a lot of quite serious research going on within the mental health domain to say, can we build digital therapeutics that can replicate some aspects of talking therapies with a therapist to help people maintain their, you know, that maintain their mood in the case of mental, mental disorder. And the reason that's interesting is because that is such an intimately human condition to have, right to feel depressed or to feel anxious or to feel obsessive feels like it's as far away from, you know, engineering a bridge as anything that you could imagine. But it seems to be the case that serious doctors do believe that these things can be useful therapeutics. And to that extent, I think that we as people are getting uncomfortable with the idea that we might not be as special as we thought we once were. And that absence of specialness is part of human history. It's why Galileo got such a rough time because Galileo started to say, listen, we're not as special as we thought we were because we're not at the, you know, the Earth is not at the center of the solar system. The sun is at the center of the solar system. And then Darwin started to say, hey, listen, we're not as special as we once were because of this kind of, this process of evolution which, which connects us to the rest of the, you know, the living kingdom. And I think one of the reasons we are feeling a bit resistant to some of these advances is because it challenges our specialness and our centricity.

Well, I agree with that, but I think that I have this debate all.

The time with people and my view is that, okay, being special is one thing that we may imagine that we are, but to have a technological reduction of what we are is also a very bad thing. Basically saying, okay, the machine can understand the human and we can have a machine therapist and stuff is in my view, neither required nor good nor feasible. I think they have a toy that helps an autistic kid understand that there's a world outside of his head. That's a good thing. Right? But a healthy person in paramedicist with problems or with depression or whatever, to have a machine that analyzes that without the same background of sentience or consciousness awareness, contextivity and stuff is bound to be a huge failure if you ask me. I mean, it's kind of like.

I.

Think for some things that there are applicable because of some mechanical or sort of, you know, very simple issues like the autistic child has to recognize there's something else there and that's good, right? Having a pet for that. So that works. But beyond that, I think, I don't know. I think the idea of taking this shortcut the whole time and saying we don't need a therapist to spend 20 hours talking to me to get to a conclusion. We can just have Big Blue come up with a problem solving idea. That to me is reductionism. And that's really what I'm worried about. I'm not worried about us being special or not, even though I do maintain that we should try to keep ourselves special in the long run by having some distinction to the machines. I'm worried that we're going to trust machines that are reducing down our reality to like a 2% of the actual scope.

Yeah, I think that's a really interesting point. I mean, I think the danger of reductionism comes not so much from the, the technologies per se, but it comes from the priorities that are set by society. And so the way that I would look at any kind of AI, say within the medical or the healthcare domain is like, it's like a better stethoscope. You would never trust the stethoscope on its own to make a diagnosis. You want the stethoscope to be with the doctor. And doctors, for all of their intellectual training, when they're assessing you and working with you, do take into account your detailed, detailed history, and they take into account your gait and your mood as you walk into the office. They ask lots of questions about your family, the more complex the condition is. So it's a very holistic and human approach, combined with some technology, MRI scans, stethoscopes, and now AI. The danger of the reductionism is that we are living in a. We're living in a world where efficiency and productivity and profit are part of the mantra of how to run a political economy well. And once you start to think about efficiency and productivity, you start to care less about some of these other holistic components that you have talked about. So I feel right today that the risk is that we know how to innovate, we know how to do that very well. We know how to build efficiency in a very narrow context. But we don't use words like belonging or stewardship or contribution in our thinking about how we design our social and our economic systems. And if we started to use those words and started to care less about the shortcut and more about the journey and more about the contribution, I think we might prevent this creeping reductionism, which is a risk that you have raised, which I think is a significant risk.

I think this is one of the issues that I always relate to, the issue of AI. The problem being is that as long as we define our goal to be efficiency and optimization and all of the stuff that we need to achieve to make more money and to monetize, then we will use AI also to make things hyper efficient, which is a real temptation, for example, human resource analytics and so on. And then efficiency becomes more important than humanity. I think that is a very, very bad development because we should use technology not for efficiency, but for reinventing how we do things, to do them entirely differently and in the end, more human. So I use this header in some of my presentations and also in the book Technology versus Humanity. I talk about from efficiency to Human agency. That's really what we want. I mean, efficiency is useless as a measurement. It's great for the CFO and so on. And, you know, people don't like things that are not efficient. But does efficiency matter in a relationship? Does it really matter in an artistic process whether it's efficient or not? I mean, these are minor concerns for all things that really matter to us.

And actually it's often the reverse is true.

Right.

Often an artistic process is valuable or generates the right response because it was inefficient, not because it was efficient, because it was meticulous. And if you think about so many of the things that we remember and so many of the things that have emotional resonance to us, they are often things that are a bit ritualistic and things that do take a little bit of time. Even the process of wrapping up presents under a Christmas tree, which happens in many sort of countries with a Christian Christmas tradition, there's a much more efficient way of delivering the presents than to put them under a Christmas tree and wait for a particular day.

Yeah, I mean, this goes with this discussion of AGI versus AI, right?

Human intelligence versus artificial intelligence. Or as I like to say, smart computers or intelligent, rather ia intelligent assistants. Right. So human intelligence can be inefficient, can be messy, it can be organized. It doesn't show up on time. It doesn't. Maybe not doesn't show up at all and comes later. We change our mind. And I use this world, you know, machines are binary and we are multinary. You know, we can take in hundreds of things and then we just make a mess out of it, but somehow it's intelligent. And I don't think that's necessarily a magic sauce in the sense of supernatural. I think it's just a sauce that is very hard for a machine. I think Minsky said this already. Marvin Minsky, the founder of AI Already, pretty much gave the name to AI said already that whatever is very hard to do for a machine is easy for a human and vice versa. And I think this is something we need to keep in mind and also need to keep alive. The best possible combination for the future, in my view, is awesome humans, really amazing humans on top of amazing technology.

But not one replacing the other there.

We should be very careful with the promises that is made by a lot of AI companies of saying that this can be like me, this machine. That's not even a requirement in my view. If this machine can take care of the heavy lifting and the monkey work.

I'm a happy man.

Right?

It doesn't have to be like me.

It doesn't have to be like you. I think there is a tension there, though, because we talked about all the negatives of efficiency. But I also think about how efficiency and time saving is also part of the everyday of our experience. And I often think back to things that my, my, you know, that my mother would. Would do, you Know, she was of a generation where she was looking after the family and alongside her work. And she learned, she put together all sorts of tips and tricks that were more efficient in terms of how she should sharpen the knives or, you know, what was the best order in which to make the kids breakfast, given how we arrived downstairs. So there is a type of efficiency that we think of as part of family intelligence and just being human. But then there's this type of efficiency which makes us think about kind of robotic workers and working robots in an Amazon warehouse, which makes us think, hey, that doesn't feel quite right. And I'm curious about how you think about the division of those two. Right? What separates them?

Well, as everything. You know, this is not a black or white question. I mean, it's funny that these days I find myself answering saying, okay, it would be simple if it was black and white, you know, good or bad or yes or no. But it's usually a huge gray zone with like an iris, you know, going from black to white. And so the same applies for efficiency. I think efficiency is a worthy goal. It's something we need. We certainly don't need inefficiency, for example, with energy and so on. But if we keep it as the number one goal, then we would not exist because we are by nature inefficient, because we're not machines, machines. I mean, efficiency is for robots basically, and anything that you can robotize will be very efficient. So I think we should make sure that we achieve those things that are achievable. But for example, being an efficient musician, okay, doesn't mean I'm going to compete with an AI that writes a song. It's like that's entirely different if you want elevate on music or. So a machine can perfectly well do that if you want great music. Machine can never do that, I don't think. And we shouldn't ask it to because there's no point. So that's kind of my view that we need to get away from this thinking that everything has to be efficient because that goes against our very existence.

If we're going to hold that up.

I agree with you, I agree with you. But we also have to start to think critically about what we have been consuming. One of the things I found fascinating about the AI debate over the last couple of years has been that it's thrown a lot a light onto things that have been happening for years, but we haven't really given much thought to. One of my favorite is this thing called auto tune, which is used in the music industry to make people sound better. So when you hear a top pop star's record playing, you're actually not hearing their voice, you're hearing their voice run through a whole set of algorithms to make it sound much, much more consistent. And to what extent are we already living in the world that you are warning about? As you say, these things are not black and white. There are shades of gray in between. To what extent have we already got to that point?

Great debate. I mean, this is exactly the similar debate than drugs, for example, right? I mean, we don't outlaw alcohol or smoking per se, everywhere, but neither you nor me would drink a bottle of brandy for breakfast. And so there is a certain kind of social contract and agreement as we how we use that stuff. Right? And if you were a really lousy singer, you couldn't keep the pitch or you just weren't happening at all. Auto Tune couldn't help you. Or if it could, then you certainly couldn't perform live. You know, so there's a limit to.

How much we should use these available tools.

Just like in my view, there's a limit as to how much we should use AI to make up for allegedly human shortcomings or laziness and those kind of things. I think these are some really, really the discussion of digital ethics, I think that we're both involved in. Right. You know, I think we need to wrap up with the final statement and then maybe share a couple resources, because this is really the point as to, you know, this is what we're going to do on stage. Right. Let's not give it all away. So do you have a final statement on this or.

Yeah, I do. You know, I. Okay, so let's think about the final statement. I don't think we have answers right now. We have a lot of very important questions that we have to ask, and I don't think that any one of us will have the answers. What we can do is start to frame the questions and start to give people some triggers to think about what these questions are and then hear what they think about them. Because these issues are very, very important. And they're far too important to be left just to some technology companies who have an absolute urgency to sell some products.

Yeah, I agree. And my summary would be what I said in the beginning. I think that in my view, intelligent assistance is a good thing and that's really what we need. And there are huge ramifications on employment and those kind of things, but it's.

Not existential and that's where I would put my money right now to get that sort of assistance and everyday job, you know, smart everything parenthesis. I think AI as the next level, as really thinking machines. That is a question mark for me, but not necessarily evil parenthesis. But when I think about artificial general intelligence, the goal of making machines like us, I think we need to regulate that. That'll be just as difficult as nuclear bombs and to regulate and to control. But that's kind of my view of the land there. So if you're looking to do something with AI, I think focus on IA. Smart software, smart systems, smart cities, smart everything. That's the low hanging fruit. And the other stuff is also subject to many ethical debates. So in terms of resources, my resources will be tech versus human. My book technology versus humanity, that's techvs human.com I made a movie about half a year ago on AI. It's called we need to talk about AI and you'll find it on YouTube. But also just we need to talk about AI.com as a resource and you can download there as well. And if you're interested in Azim and myself appearing at your event, we have a website for my agency, thefuturesagency.com that's like future with approval thefuturesagency.com and just click on the page Azim and Gertrude to find out more. What are your resources? Azim?

Well, my best resource, very straightforward. You just need to Google Exponential View and you will find my newsletter. You'll also find the podcast season that I've run for the last couple of years. And of course it would be great to see any of you at an event with Gerd and I being the Simon and Garfunkel of AI.

Yes, and of course we also do media appearances and all that stuff. If you're interested, just send us a ping and we could talk forever, as you can see evidenced in this podcast. Thanks very much for tuning in and we look forward to connecting with you in the future.

Bye.