 Hi everyone, so I'm Elinda and I'll be talking about unartificial intelligence. So, ever since the Industrial Revolution, we've had a fascination with stories about AI. Just thinking of recent movies, we've got all these different variations of the same theme. And how does actually having AI change our world? Now, I'm a huge movie geek, but for my day job, I'm a developer at FutureLearn. And we're a London-based startup and it's a social learning platform. So we work together with universities and cultural institutions to deliver online courses. Now, what that means, though, is that our team is encouraged to learn more about the theory and the principles of pedagogy and how to build great learning experiences. Now, my own background is in AI. I did it back in university. And what I realized from learning about human learning experiences is that how machines or the artificial learn is very, very similar to how people or the unartificial learn. So that's what I'll be talking about here, explaining some concepts from AI and link them to unartificial intelligence. So before we can look at artificial and unartificial intelligence, we need to define intelligence. How do we define, what makes something or someone intelligent? So I did what every geek would do and I looked it up in the Dungeons and Dragons manual. Now, the parts about wizards and spells aren't really relevant to us here, but these parts are. How well your character learns and reasons and having a wide assortment of skills. So here's another proper definition. So intelligence is not just about having the knowledge or the skills, it's about knowing how to obtain them, how to reason about them and how to use them. So what do we mean with artificial intelligence? Well, actually we use the phrase for two different things. On one hand, it's the actual intelligence of machines or software, but we also use the term for the actual research field looking at creating intelligence within machines. And within that field, we basically have four different approaches as to how to implement AI. And these are kind of the four main areas. And today we're only looking at one of them, systems that act rationally. And the system is rational if it does the right thing. How do we define what the right thing is? So this is very much the concept of intelligent agents. So an intelligent agent is one that acts to achieve the best outcome in all situations. So we can create the simple, well, reasonably simple diagram of a simple reflex agent. So an agent lives in an environment, it has sensors with which it can observe the world, and it has effectors with which it can take actions on the world. And it basically creates a state of what the world is like now, and then has a set of if then rules. And together these two are used to determine what action the agent's next takes. So when we're looking at the unartificial, we can apply the same diagram. Our sensors are basically our five senses, taste, touch, et cetera. And our effectors are whatever we use to take actions on the world. So our voice, our hands, our movements. And this becomes most apparent when looking at the research of Pavlov. So he was a Russian physiologist known for his work in classical conditioning, where he basically trained dogs to associate the sound of a buzzer with food. Now these are my two cats, Casey and Dusty, and they really, really get annoying when they're hungry. So I thought I'd try to do the same thing as Pavlov did. Could I actually train them? So they started out like normal cats. Whenever they smelled food, they'd jump up and rush to the kitchen knowing that they'd soon get to eat. And I then started training them with a standard iPhone alarm. And I basically only stand up and feed them if that alarm went off and if they heard it. So eventually they started associating that sound with food. It was a bit of a failed experiment though because it didn't mean that they stopped being hungry the rest of the time, so they were still pretty annoying. But even now, and it's about three, four years later, they'll still recognize that iPhone alarm whenever it's used in a TV show or in a movie, and they'll rush to the kitchen expecting that they would get food, which can get pretty annoying. Because there's quite a lot of movies out there that still use that bloody iPhone alarm. So we're not that different from cats and we use the same principles on ourselves to form habits. So each morning when I hear my alarm clock, I know that means I need to wake up and get out of bed. And as a developer, I know that when I see failing tests, I should fix them. And these are very, very simplified loops, but how do we actually learn these new rules about what actions to take? How do we learn new things? So here's a bit more of a complex diagram of a learning agent. So just like before, we have sensors and effectors which we can observe and take action on the environment. And, but now we have a couple more elements. The main element here is the performance element. And this is basically the entire agent that you just saw before. It has, again, that bit that can create a state of the world. It has the if-then rules. But now we have a couple more components that can influence and change what this performance element does. So the main thing we're interested in is the learning element. So what this does is that based on observations that it gets from the critic about past actions and environment, it gets feedback and then can make changes to the performance element. So this changes the way we make decisions. So how does it learn what actions, what changes to make? And this is where learning algorithms come in. So there are a couple of known learning algorithms and I'll only highlight a couple of the more familiar ones here. So there's supervised learning and in this case the feedback is basically all up front. So you get a bunch of label training data and you're basically trying to infer the rules for which input belongs to which label. Then there's unsupervised learning and in this case the feedback is just a bunch of data. So the learning algorithm has to identify the patterns or the structure from the inputs even though there's no specific output values. And then there's reinforcement learning and rather than having correctly labeled data in this case the feedback with reinforcement learning is actual proper feedback. So the agent makes a decision and then gets the feedback whether it's right or wrong. So it's much more general but at the same time the agent has to have a much better understanding of how the environment works and it needs something or someone to tell it whether it's right or wrong. And in the same way that machines learn through different types of algorithms, humans basically learn from different types of activities. So these are, this is an overview of the 16 different types of learning activities and I'd highlight just a couple of them here because we don't have time to go over all 16 of them. But you kind of, but all of these different ones appear when you're trying to learn something. So the first thing I wanna tell or talk about is delivered and in this case learners are presented with information. So this is again very similar to supervised learning. You're presented with the content which contains all the information that you need. And as developers we do this when trying to learn something new. We basically read books, we watch videos, we attend conferences. It's all about consuming content. The next one I wanted to talk about is conversational and collaborative. So here it's about learning through conversing with others and by constructing a shared understanding. So this is very much like unsupervised learning where the learning comes from the structure of what is being learned. And again as developers we do this when pairing with others. Together we form a shared understanding of what we're working on. And then it's assessing and in this case it's about receiving constructive feedback. So this is very much like reinforcement learning. You get the feedback and from the feedback you learn. And as developers we again do the same thing basically with pull requests. We learn from the feedback we get from others. So again the overview of 16 of the learning activities. And I just wanted you all to take a very brief moment to just think about the last time you actively learned something. When was the last time? What type of activities did you actually do? So what makes us different from machines? Well for starters we are very contextual. We know what we're learning in what situation. Unlike machines we're not really bound to one purpose or one domain. We're then constantly learning. We don't have an off switch for learning. We might not be consciously learning new skills or knowledge but we process everything that's around us. Whereas machines are very much state based. And then prior knowledge we have a huge backlog of other things we know and we can make associations between the different types of information we find but this is something that not necessarily other people might have picked up on. And next to that we're emotional. We attach value to certain skills, information and experiences and we need that emotion to make proper decisions. And finally with social we learn from others and we need that social interaction like here at a conference to gain better learning. So we started creating machines that have all these different abilities but not all of them combine. And machines need to be able to do all of these in a generalized way before we consider them learning the way we as humans do. And I don't think we're that far off. I'm not the first to say this but I think we will have artificial intelligence within the century. And that's not gonna be the type of AI that is putting out downfall and causing the end of the world but rather we'll have machines that can learn and reason about skills and knowledge as humans do. And I think that's when things really get interesting because in a world where humans and machines learn the same way, does that mean they'll learn together? Will our schools become places where both humans and machines learn? So when we're thinking of web developments of the future, think about this. Will what we develop for humans also actually be used for machines? And will what we develop for humans also actually work for machines? So thanks for listening. And I'm Luna.