Latest / Strachey Lectures / Strachey Lecture: Getting AI Agents to Interact and Collaborate with Us on Our Terms

Transcript

Strachey Lecture: Getting AI Agents to Interact and Collaborate with Us on Our Terms · Strachey Lectures

• 0:01
  Hi, everybody. My name is Mike Wooldridge and I'm a professor of computer science at the University of
• 0:05
  Oxford and I'm currently head of Department of Computer Science at the University of Oxford.
• 0:09
  And it's my very great pleasure to welcome you to our Hillary term stretchy lecture.
• 0:14
  The second straight lecture that we've done online, we hope it will be only the second or third and that by November we will,
• 0:22
  with luck, be back to being able to do them face to face.
• 0:26
  So before I begin, I would just like to I would just like to thank our sponsors and our sponsors are Oxford Asset Management.
• 0:38
  They've been supporting these lectures now for six years and they've completely transformed what we've been able to do with these lectures.
• 0:44
  Obviously, we've been in a lockdown now for the past year. But when we went into lockdown,
• 0:49
  they enabled us to bring speakers that we couldn't have imagined bringing from
• 0:53
  across the world and completely transformed our structured lecture series.
• 0:57
  So we are tremendously grateful to you all at Oxford Asset Management.
• 1:02
  I hope you will be. I hope you will be tuning in and joining us today and and for the future, I should say.
• 1:08
  They always ask me to say this. They are hiring. They are based in Oxford. If you're interested in machine learning.
• 1:14
  If you're interested in advanced computational techniques in Oxford, Asset Management is definitely worth checking out.
• 1:20
  And I'd given you the you are out there. So onto today's lecture.
• 1:28
  So the hour straight you lecture today is Subbarao come home Bartee.
• 1:32
  He is a professor of computer science. Arizona State University.
• 1:35
  And he studies fundamental problems in planning and decision making and is motivated particularly by the challenges of human aware A.I. systems.
• 1:42
  He holds a number of prestigious fellowships, including a fellowship from the Association for the Advancement of Artificial Intelligence,
• 1:48
  triple-A AI, the American Association for the Advancement of Science, the Association for Computing Machinery.
• 1:54
  And he was a NSF young investigator. He served as president of triple-A on the Association for Advancement of A.I. and has served as a trustee
• 2:03
  of the International Conference on A.I. and has been a founding board member of the Partnership for A.I.
• 2:08
  His views on research, as well as progress and societal impact on A.I. are featured in many multiple international media outlets.
• 2:17
  And in particular, he writes a column, a regular column full on artificial intelligence for the Hill,
• 2:22
  which is both very entertaining and extremely informative.
• 2:26
  And I recommend it if you're interested in learning more about the reality of A.I. today and in seeing some commentary on any AI developments.
• 2:38
  He's also on Twitter and is an extremely entertaining person to follow on Twitter,
• 2:45
  making lots of interesting remarks about current developments in A.I. so raw.
• 2:51
  It's all very great. Pleasure to welcome you.
• 2:53
  I'm sorry we can't welcome you face to face, but as I just said to you off line, we hope that we will be able to do so at some point in the future.
• 3:00
  And you would be most welcome. So, Ralph, over to you.
• 3:05
  Thank you. So thank you all for coming for this lecture from wherever you are.
• 3:17
  And thanks, Mike. And Oxford as this department for inviting me to give this lectured.
• 3:25
  So. I am going to be talking to you today about how to get a chance to interact and collaborate with us on our terms.
• 3:37
  So let me get started by thinking about how we basically CEI being shown being represented in popular culture.
• 3:51
  Typically, AI systems already from the beginning from the get go.
• 3:57
  I see it as things that are like companions to humans. Whether you start from like 2001, A Space Odyssey, a HAL, you know, hanging around with Dave,
• 4:07
  Samantha hanging around in HUD and also, of course, our in ex machina even.
• 4:14
  Obviously, the popular culture, the novels, you know, Ian McKellen and most recently Cosla should go to talking about robots that work,
• 4:23
  you know, the way they imagine the baby imagining agencies. They will be working with us.
• 4:28
  They'll be sort of companions. And, you know, you will almost treat them almost as if that humans at some level.
• 4:35
  So this is the I in popular culture. So you would think that, you know, instead of the gloom and doom about it,
• 4:41
  I see the areas you'd be seeing this kind of breaking news in popular culture.
• 4:47
  Technically, I that is helping old lady across the street.
• 4:51
  It plays with kids, cooks, food hacks at home and sense of drama.
• 4:55
  In case you're wondering, how come we never saw this kind of breaking news item on CNN?
• 5:01
  It's because I made it up. It's fake news.
• 5:04
  Basically, it turns out that we typically keep hearing about a I not in the sense of agents working with humans.
• 5:14
  So there's been essentially a sort of a curious ambivalence that my research has had with humans.
• 5:23
  Our system seems to be happy as to either far away from humans, like the spirit, opportunity, curiosity.
• 5:32
  I'd basically in some adversarial stems with humans essentially claiming the heck out of us in various kinds of board games.
• 5:41
  It's almost as if we wanted to help humanity.
• 5:46
  It's the people that we idee, such as just gonna stand paraphrasing something that John Lennon said.
• 5:54
  Right. So I was sort of. This is something. Been something that has been weighing on me for several years.
• 6:00
  And I've seen sort of my research programme in for the last several years has been in terms of what the challenges of human activity ISIS stems.
• 6:12
  How do you get human systems to come? You know, I collaborate with us as companions.
• 6:19
  So I give a talk at the U.S. back at Tripoli Eye that sort of set up as a magazine article on the challenges of Human A very I.
• 6:28
  And most recently, that whole thing that Mike mentioned,
• 6:33
  I wrote a column sort of giving the current state of the area in the I fight like Lape Oblique.
• 6:42
  What I want to do in today's talk is sort of give you some of the technical background.
• 6:47
  And as much as my group has been working on in some of making some of these challenges come to fruition.
• 6:54
  So I was sort of asking in the in my trip, I talk back in 2000.
• 7:01
  Why isn't human having any AI all over the place already, given that pretty much everybody thinks about the,
• 7:07
  I think, SBI agents as working with humans, at least outside of it, and why we should be pursuing it.
• 7:14
  And of course, you know, some of the challenges and one of the points I kind of made there,
• 7:21
  and that's what sort of repeating here is having for high research to take humans into account.
• 7:29
  So we have our agents work with humans in the loop.
• 7:33
  It's not sort of a dilution of our goals.
• 7:37
  It's really sort of an expansion of our goals of the scope and reach of the enterprise.
• 7:44
  After all, it's been said that we had developed evolutionarily these crazy large breeds.
• 7:52
  We have not to run away from the lands of the Savannah.
• 7:58
  Are the title staff loyal Bengal. But to really strategize deal with each other.
• 8:05
  Also, this social interaction is what really potently led to some of the brain size.
• 8:11
  And please, at least speculatively, often from an evolutionary theory. If you don't believe that, of course,
• 8:16
  we do know that we that we are always modelling each other's mental states and what
• 8:21
  the other people are thinking about me and what would I do to counteract that?
• 8:26
  There's a great piece from like French, the show that some of you may have,
• 8:31
  Watchmen, Phoebe, saying they don't know that, we know that they know that vino.
• 8:38
  So you can actually play with these mental models both to cooperate and to strategize against, you know, adversarial stances.
• 8:48
  I mean, obviously, human eye systems are like needed in many places from the very beginning.
• 8:55
  Actually, people didn't realise that things like intelligent tutoring systems on the social robotics,
• 9:01
  the researchers in those areas have certainly taken humans into consideration
• 9:06
  in designing their systems because that's the whole point of those systems.
• 9:10
  But then that's not the only pieces where you have you want your as systems to work with humans,
• 9:18
  pretty much even quotidian interactions between systems for humans,
• 9:22
  for example, assistance, human digital, personal assistants, office hospital assistants and AFCO in teaming.
• 9:30
  You know, some of that is possible, Miles, some probably more so in the near future, elbow to elbow teaming more cognitive deeming,
• 9:40
  such as search and rescue scenarios and even like factory robotics, which can actually explain their operation to new humans in the loop.
• 9:51
  OK. So in some sense, increasingly, if in fact, that you believe the hype that yeye is everything and I actually each computer science,
• 10:02
  then human computer interaction really will become human interaction.
• 10:07
  And so we really should understand some of the challenges there and how to solve those in
• 10:14
  within the United States that I'm pretty familiar with the U.S. national priorities for.
• 10:20
  I include like 10 areas. And number two is your money AI interaction.
• 10:25
  And then there, like several research programmes, for example, Darbar has assessed ADAPT and a more recent one, BGT and so on.
• 10:33
  Many other programmes looking at these kinds of challenges. So getting to this stock, what do I want to say?
• 10:40
  What does it. What I really want to get across is what does it take for any AI agent to show interpretable behaviour in the prisms of human agents?
• 10:53
  And the simple answer that it turns out is managing mental model.
• 10:57
  Some of you have said, obviously, that's kind of obvious, but some of you may not have thought about this.
• 11:03
  This is something that psychologists have always been talking about. You know, something like Sallyanne past that some of you may be familiar with.
• 11:11
  It points out that kids recognise that what they believe about the world is different from what others believe about the world.
• 11:21
  And this is when they realise that they actually can lie.
• 11:24
  You know, not only they can cooperate by actually making clear what they believe in, but also lie.
• 11:29
  And, you know, I joke around that. I was extremely happy when my son started telling his fossilise like 25 years back because
• 11:39
  I knew he had a working brain which essentially is able to model other people's mind,
• 11:45
  mental states and realise that what he believes in is different from what others believe in.
• 11:50
  So managing mental models becomes that important aspect.
• 11:54
  So that gets me to the overview of the talk,
• 11:58
  which is sort of a perspective on how to do this from a bunch of research that we have been doing in the past.
• 12:06
  You know, six or seven years. So so the all you have to talk is I basically want to get you to see that effective
• 12:14
  human interaction requires systems to be able to manage human mental models.
• 12:19
  These include include not only the humans model of their task, but also humans model of the agents of that robot.
• 12:28
  So this is at least a second level. In general, mental models can have infinite regress.
• 12:32
  But psychologists have basically talked about the fact that humans tend to use about three level nesting.
• 12:38
  And I will specifically focus today on the second level of nesting because that winds up leading to explicable behaviour as well as explanations.
• 12:48
  I'm going to get to that in a minute. And then managing mental models brings up both influence and learning challenges.
• 12:54
  I'll talk about how we deal with them and then frameworks for human interaction,
• 13:00
  because we are a bunch of computer scientists and we might think we are humans. We know what humans want.
• 13:05
  And so here is a solution that is not going to be up to snuff, because essentially, you know,
• 13:13
  you really need to have a systematic race of evaluating whether your activities offer assistance and so on.
• 13:21
  Are actually that acceptable to the humans through systematic.
• 13:26
  Subject studies, and then finally, I'll end it with by pointing out that mental modelling capabilities allow agents not only to cooperate,
• 13:36
  just like they allow kids not only to cooperate with the humans, but also to manipulate the humans in the loop.
• 13:43
  In fact, people are worried about deep six.
• 13:46
  Eventually, we have to start worrying about head fakes, which is the agent making you believe something that's not actually true.
• 13:53
  And using that to an advantage that's kind of not beneficial to you.
• 13:59
  Again, I'm not saying that you're going to be out there trying to design any agent that are.
• 14:04
  But they can be hijacked by somebody else. In fact,
• 14:06
  I would argue that huge amounts of problems with Facebook and other social media are caused by the fact that people
• 14:14
  didn't quite realise that they are talking to something like something and an agent that is actually profiling them.
• 14:23
  When we talk to other humans spent already on guard, people who are putting their hearts out to these Facebook, giving away lots of information.
• 14:30
  And then, of course, I got used misused in multiple ways as we found out.
• 14:35
  So it introduces several new ethical challenges that we need to be worried about.
• 14:41
  Before I go into details, let me show the people who actually do most of the work that I'm talking about, which is my group of students right here.
• 14:51
  And so most of the work that I'm talking about has done by one of them,
• 14:55
  in particular on a car who just graduated this different year at this is this week and
• 15:02
  shuddered and thought that that fellow who got in I tend to watch award this year.
• 15:09
  He graduated a couple of years back that they're actually done a lot of this work that I'll be discussing.
• 15:14
  Other people also. OK. So let me start with how we handle this sort of human interaction.
• 15:23
  And it starts with, as I say, it's about mental models. So let's start with three.
• 15:27
  A tale of three models. When I say I know for the rest of the talk, when I talk about models, I just want you to understand that.
• 15:35
  I'm not wedded to the specific language,
• 15:38
  although a lot of our work was done with relational representations in PDL are planning domain description language.
• 15:47
  They've also done work with other kinds of models, such as dynamic programming models have been used in MGP are not all communities.
• 15:55
  These are all connected. Anyway, the important thing is what's the content of these models?
• 15:59
  You would think you can think of the contents of these models essentially as including Inácio state of the world.
• 16:05
  The goal state the actions, the observation capabilities, as well as the current plan.
• 16:10
  If any of either of the agents and we have two agents that they're talking about, the robot and the human asset, said robot.
• 16:17
  But that doesn't. I don't particularly care about embodied A.I. systems alone.
• 16:21
  I will actually show you that we work also with us.
• 16:24
  Basically, our software AI systems, software artefacts which are supporting us may not be embedded.
• 16:30
  So I'll just use the word robot. And in what general sense? OK. So that's what you have to keep in mind when we're talking about models.
• 16:38
  This can be presented, obviously, PDA. But they can also be represented in your favourite languages.
• 16:44
  So let's start talking about, like, the simplest single agent which is living in the world.
• 16:50
  And then as the agent is actually just by itself, it really has some approximate model,
• 16:58
  an executable model of the world, what it hopes is the executable model of the world.
• 17:02
  OK. We'll call that and you can use that to make its own course of action and live in that world.
• 17:09
  So you basically given its reward metrics, given its goals.
• 17:13
  It can actually come up with courses of action that are consistent.
• 17:17
  I didn't do it on a model and then hope that they can be executed in the world and start executing them.
• 17:22
  So this is something that we know in a one on one. The moment you put a human in the loop, which is another agent.
• 17:30
  But it's a not another artificial agent. Is this human agent with all the human fables.
• 17:36
  So one of the first things that would happen is that the human race ends up having a model, MFH, for the task.
• 17:44
  That means they have their own goals, their intentions and their observational constraints and so on.
• 17:52
  And so the robot to be able to at least sort of it, you know, at least not get they may not get into there.
• 17:59
  We are hopefully try to help them. It needs to have some idea of this image and they'll call that image.
• 18:06
  OK. So this image is a robot of the eye injuries approximation of the humans model.
• 18:13
  That's the first of the mental models we're talking about. And you heard about this in many places.
• 18:20
  When you talk about basically robots, are agents anticipating.
• 18:26
  Human behaviour. This is what we mean. We are anticipating human behaviour by using MHR.
• 18:34
  This can lead to assistance that keeps getting out of the way.
• 18:38
  So why do the humans so that you don't get out of it. Don't get into that.
• 18:41
  And sometimes also to support Oeming, et cetera. So this has been actually done quite a bit in the community.
• 18:50
  In fact, I was looking at the strategy lectures, and I know that I think a couple of years back I started Glowers,
• 18:57
  gave it OK, and she was mostly a lot more interested in just sort of the human behaviour learning.
• 19:03
  So that is the cut off. You can think of it in terms of MHR. We are done some of the work in those directions.
• 19:09
  Do we know, for example, you can do intention, recognition of the human model,
• 19:18
  basically vete in this particular case anymore to brain computer interface?
• 19:24
  This is nowhere near as fancy as the neural link that Elon Musk is trying.
• 19:28
  But even a simple thing which can work. Leslie know without actual explicit communication, there are what is actually able to see in this?
• 19:39
  The one in the lower window that that such in was the subject that actually wanted to reserve
• 19:46
  one of the blocks and that it realised that it was actually using only the other blocks.
• 19:50
  So things of this kind. So basically intention, recognition and anticipating what people want and wide and getting out of the way can be done.
• 19:57
  We have looked at that, you know, in the same direction, actually, robots can project their intentions in to the humans either by actually speaking.
• 20:08
  But in this particular case, we're using augmented reality holo lens, which sort of provides to that human not only what they're seeing,
• 20:16
  but also additional information that the robot is projecting about its goals by pointing out two particular blocks and so on.
• 20:24
  OK. So these are things that are possible. But I'm not actually going to be talking as much about that part.
• 20:32
  The one that I want to talk more about is essentially the fact that the moment humans are interacting with any AI system,
• 20:41
  they will wind up making their own model of daddy AI system.
• 20:46
  So if they see a robot, they will essentially have some model of the robot MRI.
• 20:51
  Which is the task model that they attribute to the robot.
• 20:54
  Some of this can be very magical, some of them, if they're experts in the loop.
• 20:58
  They might have some really good sense of what the work schools are, what it's actually capable of.
• 21:02
  So there is MRI edge, which is that humans model and the agent, the robot essentially now needs to have an approximation of that, too.
• 21:12
  This is the second mental model that I will spend most of my time in the dark talking about how to exploit and leverage that.
• 21:19
  So the MRI it here allows the agent to anticipate human expectations in order to either conform to those expectations.
• 21:29
  If it cannot confirm, then explain its own behaviour in terms by by sort of modifying a model.
• 21:36
  So tell the human you have to change your expectations on me in the following ways so that my behaviour makes sense to you.
• 21:43
  So to me, explanation really is this dialogue that the agent has with the human changing their model, which is their model.
• 21:52
  Updated. OK. I will mention that MHR and marriage are different from MH and Emaar.
• 22:00
  Imagine my father, human and father Robot Sackhoff execution models.
• 22:05
  These are models that they're hoping at executable in the world that actions without real preconditions.
• 22:10
  And, you know, hopefully, in fact, the application is satisfied. There should be a high probability that the action actually works.
• 22:16
  Unlike those image. Ah. And I'm on the edge of expectations.
• 22:21
  That really expectations on models. We mostly for this part of the stock.
• 22:25
  We'll talk about like the maximum likelihood expectations maybe.
• 22:30
  But in fact, they can even be like a distribution or what, the bardos.
• 22:34
  And they don't have to have any execution semantics at all. They are just being used by the agent to make sense of the other agents behaviour.
• 22:42
  So it's worth keeping that in mind. So given doors, it's basically the V, the longitudinal human A.I. Interaction Cycle works in its real assume.
• 22:53
  And I'm mostly focussing in this particular case in terms of the M.R. Edge, which is the human model of the robot's task model.
• 23:04
  And so you start with an initial estimate of haemorrhage.
• 23:08
  That is, the robot starts with the initial estimate of haemorrhage.
• 23:11
  And I should mention here that from my perspective, from any AI researchers perspective,
• 23:16
  if an agent is working with a human, the only controllable variable really is the eye agent.
• 23:23
  The most obviously controllable variable is the. And because you can't you can possibly tell the human, please do these.
• 23:31
  But they don't necessarily have to actually take your advice. So in essence, much of what I would be discussing is from the perspective of the robot.
• 23:38
  So not only does it needs to anticipate what the humans are thinking of for its capabilities, it also needs to.
• 23:47
  And I understand what the humans think of its capabilities.
• 23:51
  And I'm so used that that is our marriage and we assume that we start with our initial estimate of our marriage.
• 23:57
  And I talk to you in some cases, they're stocked with a shared model initially.
• 24:02
  Other cases you can actually learn from existing behaviour traces.
• 24:05
  I'll get to that part little later in the talk and then you sort of have this loop.
• 24:10
  There are two pieces in the loop.
• 24:12
  One is the model following where if the agent thinks that there is a sort of an understanding between what the human thinks its model is,
• 24:21
  that then it can just conform to it and marriage and conforming to it is really explicable.
• 24:28
  And I talk about it and I'll have it in a minute.
• 24:31
  So an agent basically making its behaviour such that it is in accordance with what a human agent expects it to do.
• 24:39
  That is explicable. There is another notion that other notions related to model following, which such as a predictability,
• 24:48
  which is sort of locally this behaviour should be consistent with what the human expects.
• 24:54
  Then that is the model. Communication is sometimes actually conforming to the human model, winds up being extremely costly.
• 25:05
  And so in those cases, essentially, that robot basically decides to do what is optimal for it.
• 25:10
  I'll give you an example of this in a minute.
• 25:12
  And then it realises upfront that since this behaviour is different from what the human would be expecting,
• 25:18
  given their model of its behaviour, it needs to change their model.
• 25:24
  And so that's what explicable the explanation part starts explanation really is communicating the changes to our marriage,
• 25:31
  either as it heads up it upfront or as a post facto explanation after the behaviour is done.
• 25:37
  There are other ideas such as legibility, which is sort of the explanation, except typically with implicit actions and design,
• 25:47
  which is sort of changing the environment in such a way that the environment itself gives you a heads up to the human.
• 25:54
  As to what you are. A marriage is also touch on a little bit on the design factor.
• 26:00
  So that's basically the longitudinal human interaction cycle. And I will sort of show this flash to slide a couple of times in the next few sites.
• 26:08
  So let's talk about model following part. So as I mentioned, I we actually use two kinds of use cases.
• 26:16
  One is on the left hand side where basically you have embodied robots.
• 26:21
  And in this particular case, where the robot, the Fitch robot, is going around in a mock search and rescue scenario with the human commander,
• 26:31
  are safely ensconced in some some other place and having like a camera, a dialogue with their robot on the other side.
• 26:39
  On the right side, we are looking at a planning decision, support system.
• 26:44
  If you have a mission plan, the Mars mission planner or somebody, they are trying to make the plans themselves.
• 26:49
  And the attention support system provides suggestions about how to complete their plans.
• 26:55
  So these two are the two general kinds of scenarios that we look at, the challenges that they look very different syntactically.
• 27:03
  What the challenge is about, how an agent interacts with the human wind up being quite similar in both cases as it is.
• 27:11
  I'll show you as we go forward. OK. Does other use cases.
• 27:15
  So and then, you know, given that I love social, mostly the robot use cases in the demos because those are more fun to watch.
• 27:22
  So in the case of model differences with the human in the loop,
• 27:26
  that means that emerick's that the human has is different from MRV, that the robots actual model.
• 27:32
  So that what that means is the plans that are optimal to the robot according to its model,
• 27:39
  Imod might actually not be optimal according to the human's expectations of a robot model.
• 27:48
  So I am on edge. So what the plan that is optimal in that model may not be optimal in commodities.
• 27:52
  That's all I'm saying.
• 27:53
  And when that happens, there's a surprise on the human side because they're basically seeing behaviour by the robot that they didn't expect.
• 28:01
  So this leads to inexplicability are inexplicable behaviour.
• 28:05
  At that point, the robot has two options. It can from in the model following is to change the model that the humans have.
• 28:14
  That is a marriage, which is the explanation part.
• 28:17
  And I'll talk to both of you, talk about both of them in the context of this search and rescue scenario.
• 28:23
  So imagine you have the you know, in this scenario, you have the human and the agent,
• 28:29
  the robot essentially start with the correct map of the environment in this particular case.
• 28:34
  This is the fifth floor right outside my door. And in computer science department right here is you.
• 28:40
  And presumably that is sort of a. So the model of the map of the environment is known upfront.
• 28:46
  But. It turns out that there is possibly some sort of a disaster.
• 28:50
  And so there possibly are rubble, there may be parts that are closed, maybe actually be passed out are opening up,
• 28:57
  which means that the Internet age and the robot, which is on the floor, might actually wind up realising that the map has changed.
• 29:04
  But the human still thinks that the map is the same as before.
• 29:08
  And so essentially, there's already a difference between marriage and a model.
• 29:12
  And then that needs to be reconciled through, you know, through this interaction process,
• 29:17
  either by the robot being conforming to what the human expects or by changing the model.
• 29:24
  So in the context of conforming to the human model in this particular case, in this picture,
• 29:30
  what's happening is that particular path that would have been the shortest path is closed because there is an obstacle act.
• 29:39
  Think of it as some kind of some huge double date. So if you don't want the human to be surprised at your behaviour because they're
• 29:46
  actually expecting perhaps that you would be coming out of this particular kind of dud,
• 29:50
  then you have to conform to the humans model.
• 29:54
  So in this particular case, it wound up removing the rubble.
• 29:58
  So it's costlier than just taking the next best path. That's what is explicable behaviour.
• 30:03
  So to be able to provide explicable security, given a goal,
• 30:06
  the objective normally would have been to just find a plan that is optimal with respect to its own model.
• 30:14
  But now it needs to find a plan that's not only opportunity inspectorates model,
• 30:17
  but it's close to what is expected by the human based on dead model, which is there at marriage.
• 30:24
  And so it becomes sort of a multi-modal planning problem.
• 30:29
  And then this can be done.
• 30:30
  I'll talk about this become sort of a more complex optimisation problem, you know, to actually come up with this kind of an explicable plan.
• 30:40
  And what we're trying to do is sort of reduce the distance as much as possible.
• 30:44
  Sometimes making the distance zero actually might be impossible because you must might be expecting magical properties that the robot doesn't have,
• 30:52
  such as completely removing the rubble from a particular candidate.
• 30:57
  OK. So now conforming to the explanations is costly.
• 31:01
  As I said something, you actually have to remove rubble or something in this particular example.
• 31:05
  So then you might want to communicate changes to am on edge to the human.
• 31:13
  And that would be the explanation. But in this case.
• 31:18
  If you could hear it, it was basically yelling at the top of the wires that a particular part is actually blocked.
• 31:25
  And because it's this other plot that is being taken.
• 31:29
  So in essence, it's just changed the model,
• 31:31
  just partially Delta changed to the model that the human and if the human actually pays attention and takes that into account,
• 31:39
  then all of a sudden the robot's behaviour is as explicable with respect to the changed model.
• 31:44
  So the explanations really has been talked about a lot in explainable A.I. systems and explanations have been talked about a lot.
• 31:53
  But when human A.I. agents are interacting, really it's best to think of explanations essentially as this constellation constellation of models.
• 32:04
  This is sort of close to psychological value here. You do as I talk about it later.
• 32:08
  But in essence, what we are seeing is that an explanation.
• 32:12
  Epsilon, given a plan by that robot, has essentially decided to act on is is such that if you add this Epsilon to current Modrich of the human,
• 32:25
  then this plan by will be actually optimal in this changed model.
• 32:32
  And it's also already optimised with respect to the robots model. That's why the robot again, I'm using the word optimal.
• 32:40
  It's easier to think in terms of essentially optimality from a theoretical analysis point of view.
• 32:47
  But in practise, essentially, you're more or less trying to say, I'm doing this and you're expecting a different plan.
• 32:53
  And this is the changes that I need to make to your model such that my plan would be in the Top Gear.
• 32:58
  Plans that you would see me doing about optimality allows for a much better theoretical analysis.
• 33:04
  OK, so that's the explanation. Size model reconciliation. It turns out that, in fact,
• 33:08
  computing these explanation's winds up being a metasearch in the space of the models
• 33:13
  themselves actually skipped a little bit about the representations of the models values.
• 33:20
  You know, in much of this,
• 33:21
  what we wound up using planning domain description languages which have actions and with the preconditions and effort in symbolic terms.
• 33:30
  And, um, so you if you have a model with a bunch of actions and the implications and effects of the model that the human has with a, you know,
• 33:37
  either similar odd different actions with their preconditions,
• 33:41
  what the robot is trying to do is essentially modify the human model, Emerich, that they are attributing to the robot.
• 33:50
  This will involve, for example, changing the preconditions,
• 33:53
  changing the effects of changing the costs of the actions that the humans have attributed to the robot.
• 34:00
  And this search essentially has all sorts of interesting properties. Using this model space search.
• 34:05
  You can come up with different kinds of explanations,
• 34:07
  both minimal explanations that would just make the human understand why this behaviour is fine in this case,
• 34:15
  but might question this explanation little later because you do something that's inconsistent with this explanation.
• 34:22
  You can also come up with monotonic explanations which will essentially not cross that confusion.
• 34:28
  Monotonic explanations will be costlier to compute.
• 34:31
  And one of the cases you are searching from the human model in the model space works, the robots model.
• 34:37
  In either case, you're searching from the other side. The technical details of this search are actually in this HK 2000 17 paper.
• 34:46
  That would be you might want to look at it, even though I sort of mentioned explicable.
• 34:53
  And explanation as if one is the model following and one is the model communication.
• 35:01
  You can actually combine both of them from a planning perspective, from a reasoning agent's perspective into a single planning framework.
• 35:09
  You know, we have a paper. Actually, I did this last year just before everything closed in play 20.
• 35:17
  There's a paper on expectation we're planning which essentially considers both on peak actions, the kinds of actions that change the environment.
• 35:27
  Mostly those are climate effects are changing the environment. Epistemic actions, which are mostly their effect, is to change the mental models.
• 35:37
  That means that communicative actions and sometimes you have actions and take actions can have epistemic effects.
• 35:43
  That's how you wind up doing implicit communication just by doing actions that have physical effects and then on rent.
• 35:51
  But they also are providing mental effects, a mental model changes to the human side.
• 35:58
  So the robot has both the standard Ontake actions and explanatory actions.
• 36:02
  It can model the effects of its actions on both states as well as the human's mental state.
• 36:06
  It sort of becomes for those of you and especially people like Mike. I mean, this is essentially epistemic planning, epistemic reasoning.
• 36:14
  So here, planning becomes a multi model and.
• 36:17
  Is it uses its both its model and the human's expectation model to generate a course of action that contains both explicitly unwanted actions.
• 36:25
  And so the explanatory actions are such that those are making changes to the marriage, that the human has an offer.
• 36:33
  Those changes are made. Then the rest of the Ontake actions that the plan that the robot is actually exhibiting,
• 36:42
  that behaviour is exhibiting would be optimal in this changed model.
• 36:45
  So you're doing both in the same environment.
• 36:48
  And the difference between applicability and explanation just becomes whether that's planetary message is a the E side pot is empty,
• 36:58
  in which case it's just explicable planning.
• 37:01
  And if it's not empty, then you are asking that you want to change that mental model first before understanding your behaviour.
• 37:07
  So that becomes explanation. But it turns out, actually epistemic planning has been studied before.
• 37:12
  Do one of the nice things is for just the two level nesting that we're considering that are efficient compilations to single agent planning.
• 37:23
  And so that's the trick that we use to make this doable, tractable.
• 37:28
  So the details of how the compilation works here are industry 2020.
• 37:33
  Although I looked at the robot changes, basically that the embodied robot scenarios,
• 37:41
  as I said, the same kinds of explanations can be used in the decision systems.
• 37:46
  So this is a radar system that we use to kind of provide vision, planning support to humans in the loop.
• 37:56
  We had some work at NASA as well as some building work. And now when our Office of Naval Research, in both cases,
• 38:02
  people are interested in making plans and people want to be in charge and the systems should be helping them, the A.I. system should be helping them.
• 38:09
  And the explanations, when you provide a solution, why did you provide this solution?
• 38:13
  If that human were to ask, you should be able to provide an explanation.
• 38:16
  And that is something that systems are able to do using these kinds of theories.
• 38:22
  OK, so coming back to the longitudinal human interaction cycle.
• 38:27
  Now, what I want to do is that having talked about Explicable Ety an explanation, I want to talk about the learning bot.
• 38:35
  So basically, where did that what did you want and what which model come. Sometimes it is as I said, it maybe did as a shared motivationally.
• 38:42
  Sometimes it has to be learnt. So in some cases, such as that USAF scenario human and the agent will both start with the same shared model.
• 38:50
  So all that is needed is tracking the model drift.
• 38:54
  Even if the robot doesn't know the model and Modrich with certainty, it can jism with multiple possible models.
• 39:01
  If you are not sure which is the specific model the robot a human has.
• 39:04
  If you don't know Emerich specifically, do you think that it's one of these scale models?
• 39:09
  You know, in the most general case, you have some Bayesian distribution of what these models,
• 39:13
  especially the case where you have kear different models.
• 39:16
  We can actually reason with these different models and provide conferment explanations of conditional explanations.
• 39:23
  With respect to these models, six models that there's a paper back in 2008 didn't come back.
• 39:30
  In other cases, in the end, you actually have to let alone these models from scratch.
• 39:36
  Some behaviour traces. If it's not being provided by upfront, by knowledge engineering aspects.
• 39:42
  So in that sense, the agent would need to learn the human and mental models from the traces.
• 39:47
  One thing that we do want to keep in mind is unlike MHR, which is anticipating humans,
• 39:54
  which can be learnt from human behaviour, traces that have been cached or at large amounts of time.
• 40:00
  You've just seen too many humans acting on the problem in over a long period of time.
• 40:04
  And you just use that odd edge which is requiring basic that is humans take on what the robot is doing.
• 40:13
  And so that doesn't just come from the human behaviour. Oftentimes we have different expectations of the agent than what we are capable of doing.
• 40:23
  So, for example, if you're walking to a room and you see a robot that starts dancing,
• 40:29
  you would be surprised because you don't expect robots to be able to dance, except maybe in a couple of, you know, videos that you may have seen.
• 40:37
  So whereas humans, you do expect them to be able to dance. So that MRI really requires places that human likes or the traces of the robot's behaviour.
• 40:46
  So that's a slightly harder learning problem. And sometimes you may actually have to deal with vocabulary differences, too.
• 40:53
  And I'll talk about that.
• 40:54
  I mean it with respect to one of the points that I want to mention here is that, as I mentioned, MHR and Edmar age are both expectations on models.
• 41:08
  And so they really don't have to be in any specific format, such as being little Fotomat that a modern MHR.
• 41:17
  Rewards. Because they actually have to plan with them. But that much out and am I right?
• 41:21
  I only SRF expectations and they can be represented in many more flexible ways.
• 41:26
  One of the things we did essentially is to learn human preferences of the robot behaviour as
• 41:33
  labelling functions during a training phase and use these labelling functions internally.
• 41:39
  But the robot then learns this labelling function.
• 41:43
  There's a paper in 2017 which sort of treats learns this as a sort of CRF and uses this labelling function as part of its behaviour,
• 41:53
  its planning process.
• 41:54
  So it computes this distance between the plan it is making and the plan the human expects in terms of this labelling procedure that it learnt.
• 42:05
  And so that is a way you can be explicable. It turns out that the same idea for learning.
• 42:11
  So you're essentially in this particular case, you're learning a marriage not by learning it in explicit actions,
• 42:18
  clear conditions affect state, but in terms of just a labelling procedure.
• 42:22
  You could do the same thing even when you are not doing explicable. But you're also providing explanations.
• 42:27
  In that case, all you need to do is improve the training scenario,
• 42:32
  expand the training scenario such that you are showing a behaviour as well as some explanatory
• 42:37
  messages next to them and asking you humans to say whether or not the behaviour makes sense.
• 42:44
  Which parts of the behaviour is making sense. Which doesn't. And this is the training phase.
• 42:49
  And having used this,
• 42:50
  then you can then have a labelling procedure which can then be used to compute the explanations on demand without having a an explicit EMERICH,
• 43:01
  which is in terms of actions, pretensions and effects. So this is why it's important to understand that Modrich and my child are just expectations.
• 43:07
  They don't have to be executable models. Having said all of this about explanation,
• 43:13
  some of you probably have heard a lot about explain ability and explain will I in the context of machine learning systems.
• 43:21
  I want to kind of make a connexion between all I discussed and some of the body of the work.
• 43:26
  They're both similarities and differences. And it's what understanding XY is hot, but mostly as a debugging tool for inscrutable representations.
• 43:38
  So, for example, oftentimes you lined up the system basically says this particular particular dog is an Alaskan husky.
• 43:49
  And you ask, why do you think it is an Alaskan husky?
• 43:52
  And the system points out that the snow part of the pig cells that the saliency region is, that's not big cells.
• 43:59
  And then you might say, aha, you don't really understand Alaskan husky at all.
• 44:03
  You're just understanding snow and the correlation of snow being present next to the dog.
• 44:09
  But notice, first of all, that this is a debugging explanation and it's a pointing explanation.
• 44:15
  And really, you can't actually point to many things.
• 44:19
  You know, oftentimes the kinds of behaviour I'm talking about, the sequential behaviour as to why did you do this decision?
• 44:27
  You you want to point out, you know, do a pointing explanation there.
• 44:32
  You need to point to a space time, too, which is extremely hard and very unwieldy,
• 44:37
  which is why civilisation progress by sort of developing these symbolic vocabularies through which we provide explanations.
• 44:43
  And that's the kind of model based explanations that I've been talking about till now.
• 44:48
  In fact, if you have looked at this adversarial example scenario that, you know,
• 44:53
  many people have probably seen about that school bus on the left hand side of a little bit of noise becomes an ostrich on the right side,
• 45:01
  you and you don't see it as an ostrich. You know, our system, a classification system might very well say with high confidence that it is an ostrich.
• 45:08
  And if you ask it, tell me which part of this makes it. Moscovitch, point to me what parts makes an ostrich.
• 45:14
  That would be a useless explanation because it'll just show a couple of several pixels,
• 45:20
  you know, all over the picture, which don't have any rhyme or reason from your point of view.
• 45:25
  But that's what the system is concerned that made that class for this on ostrich.
• 45:30
  So the point being that pointing explanations of the primitive explanations that
• 45:34
  are mostly useful only for debugging explanations are critical for collaboration,
• 45:39
  but they're not really a solid luckly by the agent. It's not just talking to itself, which is trying to make sense to the human agent in the loop.
• 45:47
  And that would require a lot more than pointing pixels.
• 45:50
  And the model, the Constellation view that I've been talking about here views closer to psychological theories of explanation,
• 45:57
  such as the ones that not Labruzzo uncle talked about.
• 46:01
  That's a useful thing to keep in mind. So. One other thing I want to mention when I'm talking about here is that handling differing vocabulary's.
• 46:15
  Is another important issue that could wind up happening.
• 46:18
  We know we did not talk about whether the models are on completely different vocabularies at modern Modrich.
• 46:25
  Then you have an inscrutable system which essentially, basically is learning its own representations.
• 46:31
  And so there is no direct connexion to any concepts that the human understands.
• 46:37
  This is where you were hoping that you'd be able to just point to the pixels of
• 46:40
  something that is common between us and point to that particular substrate.
• 46:44
  But really, if you have things like windshear vision problems,
• 46:48
  you will need to provide the explanation in terms of concepts that humans can understand.
• 46:53
  And so ongoing work that we have in a lab, as well as some other work that some other people like Kim have done, involves essentially.
• 47:01
  Translating the explanation that you have into the vocabulary that the humans understand and to do that translation,
• 47:08
  you need to first learn the translation so you understand mappings from simple to big cells in this particular case.
• 47:14
  And then can work, can hold the explanation in to these symbols.
• 47:20
  And when you do that, it will be possibly an approximate explanation.
• 47:23
  But that's way better than just sort of throwing a big place off your reasoning to the humans and say this is the reason I did it.
• 47:31
  Actually, I should mention that that's trivial. Violist farm off explanations.
• 47:36
  Are those where you say you will understand what I did. If you take my brain here, take it.
• 47:40
  And that is one of the cubicle farms. And we understand we assume that that that's not a good explanation at all.
• 47:46
  That's why we actually search for smaller pieces of changes to the marriage.
• 47:51
  That will be enough that human to understand your decisions, just to put this in a classification kind of scenario.
• 48:00
  We have some recent work which shows that sort of thinking about mental models,
• 48:06
  even if it's not sequential additions, making problems can be quite useful.
• 48:09
  So, for example, in many classification systems are typically evaluated only in terms of their accuracy.
• 48:16
  How often? Not correct. But I would argue that really it's equally important to see when they are wrong, how badly off are they?
• 48:25
  So a system which makes an egregious misclassification. Humans, we lose trust in it.
• 48:31
  By the way, the word plussed is the first time I mentioned it in the dog, but.
• 48:35
  The interaction between an agent and a human humans might kind of, you know,
• 48:40
  might engender trust in humans if in fact they can make sense of the decisions that the agent is making.
• 48:46
  So in this particular pictures that they're showing,
• 48:49
  we essentially about this system where it not only takes into account the classification accuracy,
• 48:55
  but that misclassification egregiousness and using that,
• 48:59
  you know, our system, for example, will be able to come up with better kinds of these classifications,
• 49:06
  even if it is asked to misclassify than the existing system.
• 49:10
  So, for example, some of the examples there are showing that, you know, if you have something like a Dabby on the very first top left gardener,
• 49:21
  you know, existing systems might just when they make a mistake with this, they might just say it's a remote control.
• 49:27
  And, you know, our system will say it's a tiger cat. So because in essence, it's also looking for the cost of that misclassification.
• 49:34
  And that's something that you can do with respect to mental models. So there's a bunch of things that are that can be extended from this.
• 49:42
  For example, communicating the model.
• 49:44
  Myard may not guarantee portability unless the human has the inferential capacity to compute optimal behaviour from the model.
• 49:53
  I sort of, you know, pushed out of the bag earlier.
• 49:56
  It turns out that you can use file based interactions with a bunch of what that we have done their way into humans.
• 50:03
  Just ask why not this other behaviour?
• 50:05
  How do you only provide reason as to why that other behaviour is not as good as the behaviour you have sheets shown.
• 50:12
  And this will be easier start off to understand than just dealt us to the model with
• 50:18
  respect to which you need to be able to see that the behaviour is actually optimal.
• 50:22
  And similarly, when you provide information to the humans,
• 50:26
  the real question is also a question is also whether the agent is actually
• 50:30
  humans are paying attention to the communication and that involves two parts.
• 50:35
  Is the communication actually perceivable, at least in theory, at least?
• 50:38
  So, for example, if I was getting this talk on radio talk radio, if I started showing you slides, holding them in my hands,
• 50:46
  the joke is on me because I'm not reasoning about the fact that you won't be able to see what I am sure you.
• 50:53
  So that is basically reasoning with the communication, possible perception modalities.
• 50:59
  The other is, even if you see, you may not be attention to it.
• 51:03
  That is attention management. That is something we haven't done much.
• 51:07
  But in fact, it's been done by others, such as the incorporates know in the context of the all Clippy, the Microsoft assistant,
• 51:16
  which is basically figuring out what is the best time to tell humans anything so that they're likely to pay attention to it.
• 51:21
  So in terms of the file based explanations, for example, we can you know, here's a bunch of papers that we have made.
• 51:28
  The humans basically ask, why not this behaviour? That is, they'll provide a plan, a partial plan of the company then and say,
• 51:35
  why aren't you doing this one against the one second is the one you are currently doing.
• 51:39
  And you can use that question itself to figure out what level of abstraction and the abstraction hierarchy
• 51:46
  of models that human side in and use that knowledge to provide that kind of explanations to them.
• 51:52
  That's like a file based explanation. We actually did this in the context of this sub radar system that we're talking about,
• 52:01
  where we provide a longitudinal dialogue between the agent and the humans and the human actually provides the file.
• 52:08
  Not only can we provide an explanation as to why that file won't work,
• 52:12
  but we also realise that that question means that the humans really would like the plan to be closer to the file.
• 52:19
  So then we do have a planning scenario where we will bind up coming up with a plan that's closer to the file, even if it is a little less optimal.
• 52:27
  And there's a demo CHIPLEY. I know that that talks about this.
• 52:32
  We also extended these models to also consider multiple humans in the loop.
• 52:37
  So essentially providing customised explanations for different humans at different times based on their mental models.
• 52:45
  And then I talked about this controllability of zero controlled observability planning problem.
• 52:51
  This is something that Unida has done there. Essentially,
• 52:55
  the system can reason about what can and cannot be seen by the human and use that both to make sure that they see that the help is being given.
• 53:04
  And also, if it if it's so, please,
• 53:06
  it's up Outfest gate to people in the loop for it doesn't want them to understand that this help, it can obfuscate.
• 53:14
  So that's also possible. That's actually a 19 paper.
• 53:19
  And then more recently,
• 53:21
  it has been looking at this proactive assistance part where essentially you are reasoning about the fact that help should not only be given,
• 53:28
  but should be seen to be given. So at least you should be able to reason about the possible upsetter.
• 53:34
  Aspects of the model observability aspects and make sure that the human understands
• 53:40
  that you have provided to help either by explicit communication in Oneida's work,
• 53:45
  it's actually by implicit communication. So you wind up showing that you are transporting second objects and that will sort of,
• 53:52
  you know, provide to communicate certain messages to the humans.
• 53:58
  So I just want to quickly go what this and maybe in the couple of minutes, I know that they're very close to hand.
• 54:04
  So the framework for human interaction really cannot just be me.
• 54:08
  I'm saying I did what because we are humans and we think it works for us.
• 54:12
  So our solution for this has been interdisciplinary collaboration with the human factors such as Nancy Cook.
• 54:18
  Sean Did is a colleague here, and she has actually been she was a past president of the Human Factor Society.
• 54:25
  And so we work with hard and hard group in setting up systematic human subjects,
• 54:31
  studies to actually cheque whether the kinds of explicable behaviour as well as explanations
• 54:37
  actually are coming up with makes sense to realise humans in the loop and doing human subjects.
• 54:43
  That is pretty painful. But I would also say that anybody who says that into human interaction but haven't heard of the word I ought to be,
• 54:52
  which is institutional review board certification,
• 54:54
  I'm probably making it all up because without that, you have no clue as to whether or not these things are actually working.
• 55:00
  So, in fact, there's papers in both H and I.
• 55:02
  And it's a journal that show how that kind of explanation and that we come up with are seen to be useful in collaborative behaviour with humans.
• 55:16
  A very lasting mental model capabilities allow agents to manipulate the humans in the loop.
• 55:22
  As I mentioned, they can allow for head fakes.
• 55:25
  This sort of leads to a bunch of ethical quandaries far any work which starts thinking in loving human agent agents,
• 55:33
  true mental model humans evolutionarily mental modelling allowed us to both cooperate as well as compete about each other.
• 55:41
  But really, you can't just stop your agents from having that ability because without the mental model, they won't cooperate with you.
• 55:48
  Even so, you will have to essentially deal with the Pandora's box like that has to be open you and systems with mental more linkable.
• 55:56
  It is being ethical boundaries, additional ethical quandaries beyond the usual.
• 56:01
  Oh my God. Autonomous robots are going to kill us all.
• 56:04
  Such as, for example, automated negotiating agents that misrepresent their intentions to gain material advantage.
• 56:10
  Your personal assistant that tells you white lies to get you to eat healthy.
• 56:16
  I end up, as I already mentioned, people would have been actually more guarded with social media if they realised that those platforms that
• 56:22
  actually actively profiling that we are already unmoored off our guard when we're dealing with other humans.
• 56:28
  But sometimes we don't realise that Facebook is actually an agent.
• 56:32
  It just doesn't look like a human. But it has some of the same capabilities in terms of profiling.
• 56:36
  So, you know, in particular that with the social intelligence scenarios, humans example, closer sentences are far more pronounced.
• 56:44
  And so we need to be very careful about how to deploy these technologies.
• 56:49
  As I said, head fakes really are a lot more powerful than deep fakes, and we don't need to worry about them.
• 56:55
  But then again, every tool is actually a weapon, too. If you just hold it right.
• 56:58
  And that's true also for these kinds of things.
• 57:01
  You know what what we actually have shown that model, the Constellation, can be used to tell lies, lies of commission as well as less commission.
• 57:09
  And then this work by Unagi that I mentioned earlier, that actually shows, for example, that that.
• 57:20
  In this particular case, in one case, the robot is making sure that the guy sees what it is doing.
• 57:25
  Watch what the robot is doing. Novik, is it actually dyspraxia to do something else?
• 57:31
  And so basically it's about manipulating the perception model. And so this is like signalling France and taking to enemies simultaneously.
• 57:39
  These are things that can be done. The purpose on this in 2009 interview a monster 10 20.
• 57:45
  Finally, you know what we did when I mentioned that lying robot study saying when can I box like we started under what conditions?
• 57:53
  People are willing to be told how light lies. And there's a paper on that.
• 57:58
  Yes. System conference. So I'll stop here.
• 58:03
  So somebody basically what I was trying to get to you is that effective human
• 58:07
  interaction requires human systems to be able to manage your own mental models.
• 58:12
  But the model that you want to model up their task as well as the humans model of the robot,
• 58:16
  at the least you could do crazy things with infinite regress.
• 58:20
  But in fact, you can do a lot of useful things with just the second level models.
• 58:24
  That's what I've tried to tell you. Managing mental models brings up inference as well as learning challenges.
• 58:30
  And these frameworks for human interaction have to be evaluated with actual human subjects.
• 58:35
  Studies and then mental modelling capabilities do open up new ethical problems.
• 58:41
  And we have to be careful about them. There's a paper in my magazine which sort of gives an overview of some of these issues.
• 58:48
  And then we also just wrote sort of a draught of a book that's under review, not explainable human interaction, planning perspective.
• 58:56
  With that, I'm going to stop and thank you for your attention.
• 59:07
  Thank you so much for that, rather, it was fantastic. So we are ready for questions, so please enter some questions in the Q&A.
• 59:16
  There is if you look on your teams, there is a button on the top right foot, which is cute.
• 59:24
  And there is a little question mark of speech bubble with a question mark on.
• 59:28
  If you click on that, that will get you to the questions and answers. Let me kick things off.
• 59:33
  So I absolutely see where you're coming from or where you want to go with this.
• 59:38
  But how far can you get with this towards, you know, to kind of human level modelling?
• 59:46
  How far will this take you? Otherwise, there's something that's missing. What do you think?
• 59:50
  This is all that we're going to need? Meaning what? Human level modelling.
• 59:54
  And this is the way we deal with each other. Yeah, yeah.
• 59:58
  I mean, this is partial steps towards it. And it's very useful in actual quotidian interactions between human and A.I. systems.
• 1:00:08
  I do realise that we do allow far sort of large amount of nesting when we design with each other.
• 1:00:18
  You know, oftentimes, in fact,
• 1:00:19
  one of the interesting things which I didn't talk about it is you can engender trust in somebody by doing what they expect you to do.
• 1:00:27
  And once they start trusting you, you can actually use it.
• 1:00:30
  You are at one day. So, in fact, we have a paper where this sort of reason about how much trust the labour written or robot should engender.
• 1:00:38
  So that it's not required to be explicable in the sense are not being monitored as closely.
• 1:00:45
  So that it turns out that given that people understand this already, you know, when you and I interact,
• 1:00:51
  if you always do things that I expect you to do, I start wondering, you know, stuff not being surprised.
• 1:00:57
  I mean, in the beginning, I'm not surprised because you're doing what I expecting you to do, but you're always doing what I'm expecting you to do.
• 1:01:04
  Then I wonder that you how other al-Qaeda goals and then so that will basically bring in additional nesting off the mental models,
• 1:01:14
  which we don't deal with right now. I would suddenly think that is something that will be important in bringing it closer.
• 1:01:21
  The other thing, of course, is really how we wound up human to human communication.
• 1:01:27
  We wound up developing this sort of a common shared vocabulary, you know, over a period of evolution as well as our number going up.
• 1:01:35
  And this is something that was very important.
• 1:01:39
  Far, far. Yeah, yeah. Agents in particular, those which are learning their own representations,
• 1:01:46
  in particular because the representation they learn may not have any connexions to what humans understand.
• 1:01:51
  And so this idea of translation, this is something that's going to be quite important.
• 1:01:56
  So I don't really know if this is still a big issue as to whether my thinking is in the same words that I'm talking to.
• 1:02:02
  You are I'm thinking in a different way and I'm talking to you in words that you will understand.
• 1:02:07
  And this is something that the agents have to deal with, like more.
• 1:02:10
  And this translation problem is something that I mentioned some of it earlier, but that's going to be another big issue in getting there.
• 1:02:19
  OK. So we have a question that's come in from Max, rankly. Max.
• 1:02:23
  Yes. Thank you for a brilliant lecture.
• 1:02:24
  How is this approach similar to the ways that intelligent tutoring systems model learning skills and understandings
• 1:02:31
  and then came to lessons and follow on skills reinforcement appropriately other existing similarities?
• 1:02:37
  And you think your approach will enable intelligent tutoring systems to be better tutors?
• 1:02:42
  Yeah, that's very much online. So in fact, I should have mentioned that when I said human interaction,
• 1:02:49
  I didn't quite explain who is helping whom and who is trying to get out of who's way in the way we do the work.
• 1:02:57
  Actually, the robot could be the teacher, the robot could be the student.
• 1:03:01
  That or what could be peer to peer and ideas basically is the robot as the tutor to the human.
• 1:03:08
  And in many of these issues, actually, in fact, as I mentioned in the very beginning on the human eye, very AI application slide,
• 1:03:15
  people who have always given let's speak to humans in the loop are intelligent tutoring systems, people who like out on land.
• 1:03:22
  My colleague in that field here and other ideas, people, though, always that they didn't have the luxury of saying let's ignore humans, basically.
• 1:03:30
  The whole point is trying to teach the humans.
• 1:03:33
  And so, yes, there's a very significant connexion between the ideas of literature and the work that we are doing.
• 1:03:40
  We actually have papers on applying some of this model reconciliation framework in teaching scenarios that,
• 1:03:50
  you know, this is an AI caps workshop paper that's available from our homepage.
• 1:03:55
  In fact, one of my students, our current students, Suchin,
• 1:03:58
  was actually in one of the earlier videos where he tries to get away from the robot hand he worked with Good.
• 1:04:03
  And he's working with me. And so there is this obvious kind of. That idea systems and that human interaction are very much connected.
• 1:04:12
  One typical issue has been that idea systems tended to be sort of more samite automated oftentimes.
• 1:04:21
  And the level of explanations that they provide are somewhat different from the kinds of explanations we tended to provide in the what.
• 1:04:30
  But they're very much complementary and it would be less than that.
• 1:04:34
  We try to learn from the ideas, literature and also fought back. Yeah.
• 1:04:38
  And suddenly that's a clear cut case. In fact, I why has this new project coming up got into a perception guided task training?
• 1:04:49
  That's something where the robot will teach humans how to do a factory.
• 1:04:53
  What does any agent teach factory worker? How to do new physical activities by example.
• 1:04:58
  So it's a robot coach and that winds up having all these issues.
• 1:05:03
  You know, you need to have a marriage. You need to also have event shot and use them to provide these kinds of support.
• 1:05:11
  Okay. So. Well. So everybody out there please ask your questions.
• 1:05:15
  We've got another few minutes while we're waiting Sittler. Any other questions. So one other one for me.
• 1:05:19
  So so I was really pleased to see you allude to the work on the social brain hypothesis,
• 1:05:26
  which is this idea that we have big brains, which enables everything because we have large social groups.
• 1:05:32
  So we need to keep track of complex social relationships. And the famous work there, I think is work by Robin Dunbar at Oxford.
• 1:05:39
  Evolutionary psychologist, which is lovely, lovely work.
• 1:05:43
  It's the kind of work that I look at and think, well, I wish I did that kind of research. You know, that's.
• 1:05:49
  So what do I think about that work?
• 1:05:51
  I mean, the the example that I use to illustrate it is this six word dialogue that's due to Steven Pinker where where Bob says,
• 1:05:59
  I'm leaving you and says, who is she? Right. And I love it because it's so crisp.
• 1:06:04
  Six words. And again, it's incredibly rich. And we all understand what's going on there.
• 1:06:09
  And we can all fill in all the details about what's going on in the relationship between, you know,
• 1:06:15
  we can draw rich, rich, rich pictures of these two individuals based on just those six words.
• 1:06:22
  And that a very important part of that understanding is our kind of common sense theory of mind.
• 1:06:28
  We all have. So what I'm puzzling over and this is entirely independently of your view of your talk is is it a target?
• 1:06:38
  Should it be a target of a A.I. to build machines that can understand that dialogue?
• 1:06:43
  I mean, you imagined as a robot in the room with with ball and should it understand should what should it do modelling and understand that dialogue?
• 1:06:51
  I think it's certainly my say. I said basic.
• 1:06:56
  I mean, so what would the worry, presumably, that you have is that it might lead to other unethical uses.
• 1:07:03
  I mean, I think that ship has sailed my sentences and everything.
• 1:07:08
  I mean, that's what I think. I mean, every technology is dual use technology and intelligence is the ultimate technology.
• 1:07:14
  So everything we do in the context of intelligence is going to be dual use.
• 1:07:19
  You cannot stop developing it. You have to dial up with complete care.
• 1:07:24
  They can to make sure that, you know, some some sorts of ethical framework is being sort of followed, even flying them.
• 1:07:32
  So my science and my sense of as an AI problem, as a problem of not understanding intelligent interaction.
• 1:07:38
  This is very much relevant to actually fill up these blanks.
• 1:07:43
  I mean, you know, using the significant amount of common sense background knowledge,
• 1:07:46
  what actually has happened as to how those sorts of things lined up, becoming a problem, of course.
• 1:07:55
  Everything I talked about almost can be misused.
• 1:07:59
  And so that's that's just the issue.
• 1:08:04
  Yeah. So I would say it suddenly is what worth doing and in fact, part of the idea.
• 1:08:09
  They will be dealing with inferences in the context of rich background knowledge.
• 1:08:14
  I've sort of made it simpler by saying the model only with respect to the model.
• 1:08:17
  The inference is being done, but model plus everything else you know about the word will come into the picture.
• 1:08:23
  Doing those influences. OK. OK. So we go.
• 1:08:27
  I think one last question. So the question is human working relationships require time to establish after a few months with a new colleague.
• 1:08:35
  One starts to learn whether competence can be relied upon. I think there's a backstory there.
• 1:08:39
  And for example, jargon develops.
• 1:08:41
  To what degree are your eyes learning about individuals of different categories of human behaviour and building these relationships over time?
• 1:08:50
  So that's a fantastic question. So as I mentioned, this one slide about longitudinal picture of model following and model communication.
• 1:09:01
  This is something that, you know, I've sort of used to motivate a lot of our work, more stuff or beginning stuff.
• 1:09:09
  I work up until, let's say, last two years back was in the fast shot explanation.
• 1:09:15
  But longitudinal explanations lined up being changing your mental model of the human as well as JD noticing that data on commodities changing.
• 1:09:25
  A classic example is this idea that if the robot is doing something inexplicable in front of it, who won the first time?
• 1:09:35
  They'll be surprised. The third time they will still be supplies by 17th time.
• 1:09:39
  The have nots are placed on by the 18th time and the robot actually realises that it's making
• 1:09:44
  a mistake and starts doing the more explicable thing that humans do against uprights,
• 1:09:48
  because by this time they went around changing their model of how the human robot marriage.
• 1:09:54
  And this is something that happens in longitudinal scenarios.
• 1:09:58
  In that cluster work that I mentioned, we are actually trying to use that to see that if you engender enough trust,
• 1:10:05
  then you will basically not how to continually show explicable behaviour.
• 1:10:11
  So if you're shown enough explicable behaviour, some number of times, humans, basically how I trust them.
• 1:10:16
  So they have less of a monitoring going on on you, at which point you can live your life and do more things that are more optimal.
• 1:10:23
  Again, when you're doing this, whether in fact that's going to be unsafe or not, that's it.
• 1:10:28
  Actual separate issues or whether that robot is doing something deliberately unsafe with a human.
• 1:10:33
  Is that something that we have to still deal with?
• 1:10:39
  No, but the idea that our mental models will be changing during the interaction and so explanations
• 1:10:47
  will be changing as the mental models change is very much part of the work that we are doing.
• 1:10:52
  The more recent papers, in fact, is one that I didn't get into,
• 1:10:57
  which talks in terms of a Bayesian account of all these things where you think of what the human essentially has at the Bayesian prior or what a
• 1:11:07
  marriage that is a D of a possible different models that are what has and then
• 1:11:11
  Ventilla that humans are looking at what the robot is doing effectively, that the prior is being updated to be posted here.
• 1:11:19
  And so there are new probability masses, all of this.
• 1:11:23
  And in fact, there's always this possibility that one of the models that the humans have is, oh, my God, I have no idea what the robot is doing,
• 1:11:29
  which is what we call this Amistad, which means this is you admit to yourself that I have no clue what this person is doing.
• 1:11:37
  And I just hope that that would be the least likely explanations for the behaviour.
• 1:11:42
  And the more that becomes the most likely explanation, the more the inexplicable the behaviour is.
• 1:11:48
  And so updating this distribution or that longitudinal interaction is a great way to answer this gentleman's
• 1:11:55
  question as to what how do you learn that the robot model is that marriage is changing or the interaction?
• 1:12:05
  Okay. We do. We do still have like a minute. We have one other question that's come in.
• 1:12:10
  Could this kind of modelling be used for machine to machine interactions?
• 1:12:13
  If so, would you need to change the way you form models of what the other agent expects and have written?
• 1:12:18
  I'm really sorry if this is a stupid question. We know exactly is our life is spent dealing with stupid questions.
• 1:12:24
  That's fine. Actually, it's actually a great question and I thank you for asking this.
• 1:12:31
  Yes, it can be done, but no, I won't do it that way because machine and machine interaction,
• 1:12:37
  I would say that the thing that we are missing really and I hope Elon Musk won't
• 1:12:41
  be successful in changing this is we don't have a USP connected to our brains.
• 1:12:45
  So there is no brain to brain Grantland's. Okay, so you can only interact with these humans in this sort of just like we can't eat pizza or e-mail.
• 1:12:54
  We can't learn each other's mindsets by anything other than sort of watching their behaviour and making hypotheses.
• 1:13:02
  And so that's. What makes that human interaction more complex than human human interaction also has been more complex because of that.
• 1:13:10
  If I was designing robots, in fact, I would say they will have a hive mind.
• 1:13:14
  You know, if I have a whole bunch of robots that have to walk in the factory, there's no real reason why they have to deal with it.
• 1:13:20
  Am I being extra able to do that? I guess I'll have the same mind. I'll just make sure that you have the same mind.
• 1:13:25
  And then maybe somebody from North Korea will have a different hive mind. You just have to deal with that back.
• 1:13:30
  But there are cheaper solutions if it is machine to machine interaction.
• 1:13:34
  That's some pops up. This is all relevant.
• 1:13:36
  But there's a lot more that is really because of human in the loop here, because I still think that I'm confused.
• 1:13:44
  I'm not. This is the problem, right? What? The pandemic. We did everything on Zoome except eating pizza, which we had to do physically.
• 1:13:52
  And we are sort of inefficient that way. And same thing in terms of understanding each other.
• 1:13:56
  We have to do this in this inadequate way. So that makes things harder.
• 1:14:00
  But machine and machine, you can still do it in the hardware. But they are also easier ways because you can design the machine.
• 1:14:06
  And I'm hoping that we won't design that human ward.
• 1:14:09
  And I'm hoping that neural link doesn't become successful, at least not in my lifetime.
• 1:14:14
  I don't think I thing. I'm with you on that one.
• 1:14:19
  Thank you so much for a wonderful lecture. We very much appreciated it.
• 1:14:23
  And let me just say again, you are most welcome physically in Oxford when all this is over.
• 1:14:27
  We would love to have you. And I think you've you've now got a whole bunch of people who would love to meet you in person.
• 1:14:32
  So you'll be most welcome. Now, in the meantime, stay safe. And thank you again for those that haven't.
• 1:14:37
  I really urge you to look round pieces in the hill and also follow him on Twitter if you're
• 1:14:42
  interested in A.I. and for a humorous and informed commentary on the current state of A.I.,
• 1:14:48
  there's no better place to look, I think. So thanks again, Raph, and take it.
• 1:14:52
  Thanks again for the invite. Mike.