1
00:00:01,520 --> 00:00:16,889
Know. Okay.

2
00:00:16,890 --> 00:00:19,799
Good afternoon, everybody. My name is Mike Wooldridge.

3
00:00:19,800 --> 00:00:25,650
I'm headed Department of Computer Science and it's my very great pleasure to welcome you to the Trinity Term Strategy Lecture.

4
00:00:26,070 --> 00:00:28,110
And in fact, as you will surely have noticed,

5
00:00:28,110 --> 00:00:36,390
we have an exciting double bill this afternoon because the Straight G lecture will be followed by after a half hour break by the Lovelace Lecture.

6
00:00:37,300 --> 00:00:45,370
But first the Straight U lecture and it's my enormous pleasure to welcome my colleague Sari Krauss from Bar-Ilan University in Israel.

7
00:00:45,790 --> 00:00:53,110
Sara gained her Ph.D. in 1992, one of the earliest PhDs in the area of multi-agent systems.

8
00:00:54,190 --> 00:00:57,370
And it was a system that played the game of diplomacy, as I recall,

9
00:00:58,060 --> 00:01:03,220
and this system established many of the techniques that went on to define so each career.

10
00:01:03,940 --> 00:01:10,989
Although her most cited work is in the area of non monotonic reasoning so-called claim semantics in my community,

11
00:01:10,990 --> 00:01:17,200
she's best known for her work on automated negotiation, automated bargaining,

12
00:01:17,200 --> 00:01:22,780
and this is an area which she has really identified and made has as her own.

13
00:01:22,960 --> 00:01:30,100
And over the last two decades, she's been the one person in that community which has really championed that area and pushed it forward very

14
00:01:30,100 --> 00:01:36,520
determinedly towards the goal of building agents that can negotiate and bargain proficiently with people,

15
00:01:36,520 --> 00:01:42,940
which is what she's going to tell us about this afternoon. She's won a string of awards, including computers and thought awards,

16
00:01:43,150 --> 00:01:49,450
which is the most prestigious award that can be given to an AI researcher under the age of 35.

17
00:01:50,110 --> 00:01:56,259
That was now some time ago. Sorry, we both got a bit older since then, but is then gone on to win a number of other awards.

18
00:01:56,260 --> 00:02:05,290
She's a Triple H fellow, a fellow of the European Association for A.I. and is a recipient of the ACM Autonomous Agents Research Award in 2007.

19
00:02:05,740 --> 00:02:11,050
She's a good friend and a very close colleague. We've published many papers together and I look forward to publishing many more.

20
00:02:11,060 --> 00:02:16,420
She suffers, as I do from from the current affliction of being a head of department.

21
00:02:16,600 --> 00:02:19,659
But when we're both through that, we'll both get back to it. So sorry.

22
00:02:19,660 --> 00:02:29,920
It's an enormous pleasure to welcome you to give us a straight G lecture. Thank you. Thank you, Mike, for your kind words.

23
00:02:30,820 --> 00:02:35,860
And I would like to tell you what I like to do. I like to build agent.

24
00:02:36,340 --> 00:02:40,810
I like to build Agent Z, interact proficiently with people.

25
00:02:41,590 --> 00:02:45,129
You know, in this swarm, I probably don't need to say what is agent,

26
00:02:45,130 --> 00:02:53,980
but I am looking at autonomous a intelligence systems that the plan is adapted to learn and cooperate with.

27
00:02:53,980 --> 00:03:02,260
People can face adversary, but especially I'm interested in these agents that interact proficiently with people.

28
00:03:02,710 --> 00:03:07,420
Now, what can this agent do? They can help people.

29
00:03:07,810 --> 00:03:12,400
They can replace people. And they can be used to train people.

30
00:03:12,520 --> 00:03:17,350
For example, supporting people. We have a project with General Motors.

31
00:03:17,590 --> 00:03:28,749
Well, the agent was a helping and supporting a driver in electronic cars to reduce the use of energy

32
00:03:28,750 --> 00:03:36,670
by helping set the air conditioning and the climate control system in intelligent way.

33
00:03:36,940 --> 00:03:39,690
We had the system resembling Tampa.

34
00:03:39,820 --> 00:03:53,530
That is still since 2007 in the L.A. airport, helping people to decide where to put a random checkpoints in to save, to know, talking and world.

35
00:03:53,560 --> 00:03:57,100
We got an award from the Homeland Security of the system.

36
00:03:57,100 --> 00:03:59,140
I'm really proud of it. They are really using it.

37
00:04:00,000 --> 00:04:08,639
And then on the other hand, I would like an agent that will whisper in my ear and help me convince my son to live.

38
00:04:08,640 --> 00:04:16,500
He's a smartphone and go to do something. But I really like this agent to replace people.

39
00:04:17,490 --> 00:04:27,480
For example, I would like to have an agent that will convince me of Asal not to eat the cake, but always something healthy.

40
00:04:28,050 --> 00:04:32,070
Oh, we built four with a Sheba hospital in Israel,

41
00:04:32,400 --> 00:04:42,330
a system that replaces a speech therapist in helping people with brain damage to train and improve their speech capabilities.

42
00:04:43,320 --> 00:04:47,160
And we can also use our agent to train people.

43
00:04:47,700 --> 00:04:49,739
For example, I'll show you in a minute.

44
00:04:49,740 --> 00:05:01,320
We built an automated system that can be used for training people that want to get ready for an interview on in the job market.

45
00:05:02,470 --> 00:05:10,209
Or we build an agent for the Israeli police to train people in interviewing athletes.

46
00:05:10,210 --> 00:05:19,840
It's especially young personnel to train A to interview a suspect or other people.

47
00:05:19,870 --> 00:05:24,070
Let me just show you a short video of this. So it's in Hebrew.

48
00:05:24,280 --> 00:05:29,500
Why? Because the Israeli police busted them.

49
00:05:29,890 --> 00:05:40,920
But you have subtitles. Mushrooms was a phenomenal fuel with the issue of having mushroom and leaving home.

50
00:05:44,600 --> 00:05:51,079
Okay. And so we build the agent that plays the role of the virtual suspect.

51
00:05:51,080 --> 00:06:00,020
And his answers depend on the questions that they are young that a police personnel is asking.

52
00:06:00,020 --> 00:06:14,839
And this led to a huge help pain wise in 2020 project when we try to play to train people to collaborate together in interviews while others,

53
00:06:14,840 --> 00:06:23,600
you know, the adversaries are getting are collaborating and are we need also the good guys need to train themself how to collaborate together.

54
00:06:24,470 --> 00:06:31,850
So these are many project where computer system interact proficiently with people.

55
00:06:32,180 --> 00:06:35,300
What is common to all these systems?

56
00:06:35,900 --> 00:06:45,560
So the main issue is that in all this system we need to predict the user behaviour in order to build a good agent.

57
00:06:45,830 --> 00:06:50,690
I need to predict whether a human will accept or not accept an offer.

58
00:06:50,990 --> 00:06:55,430
I need to know how they, my son,

59
00:06:55,430 --> 00:07:06,160
will respond to an argument that I say in the in the discussion I need to predict what will the adversary will attack?

60
00:07:06,170 --> 00:07:17,840
I need to predict. Well, how it drivers in the world in another project will react to setting a a car race in a location and settle.

61
00:07:18,470 --> 00:07:22,400
Now, why is it so difficult to predict people?

62
00:07:23,550 --> 00:07:29,640
Decision making? Well, I always say because people are upsetting their own experiments.

63
00:07:29,850 --> 00:07:33,960
But really because they don't maximise expected utility.

64
00:07:34,740 --> 00:07:37,860
Why they don't maximise expected utility, I'm not sure.

65
00:07:38,220 --> 00:07:42,020
But probably because they are sensitive to context.

66
00:07:42,030 --> 00:07:46,770
They like to know left of their own preferences. They affect by complexity.

67
00:07:47,040 --> 00:07:50,460
They'll have problem of self control and sitter.

68
00:07:51,000 --> 00:07:57,020
So that made predicting human decision making very difficult.

69
00:07:57,180 --> 00:08:09,480
Know I would prefer to assume that people are following really nice equilibrium strategy, but this is rarely happens, at least in my experience.

70
00:08:09,930 --> 00:08:17,820
So what can we do? So we first start with some data data about decision making of people,

71
00:08:19,140 --> 00:08:24,690
and they sometimes we have a lot of data, but in many Google are collecting data for us.

72
00:08:24,990 --> 00:08:35,370
But in many cases, we don't have that many exemplars, as you will see, because collecting decision making of how drive is,

73
00:08:35,370 --> 00:08:42,180
how they respond to our suggesting how to save energy is more difficult.

74
00:08:42,480 --> 00:08:49,650
So in some cases we are using human behaviour models that were developed by social

75
00:08:49,650 --> 00:08:58,740
sciences is the way to to shape our models and to find good features in the properties.

76
00:08:59,070 --> 00:09:02,640
Then with this we are using some machine learning.

77
00:09:02,790 --> 00:09:09,330
So many times we need to develop our own machine learning algorithm because of the constraints of the problem.

78
00:09:10,110 --> 00:09:12,120
So we have a human prediction model.

79
00:09:12,330 --> 00:09:21,750
And then when we have a specific human with a specific data, even if it's evolved over time, we can enter it to the model and get some prediction.

80
00:09:21,930 --> 00:09:27,370
But this is just the beginning. Having the prediction or a model of the human is just.

81
00:09:27,370 --> 00:09:31,890
The prediction is this is just the prediction is just the beginning of the story.

82
00:09:32,130 --> 00:09:35,250
Then the main issue is that we have some.

83
00:09:36,420 --> 00:09:47,249
Optimisation problem or some game theory setting to to solve because our agent do does have some problems that it does have

84
00:09:47,250 --> 00:09:54,510
an optimisation problem to solve does have a goal that we would like it to maximise expected utility or to satisfy the goal.

85
00:09:54,900 --> 00:10:06,360
So the idea is to integrate the prediction model of the human into some optimisation or some game theory model such that after we solve it,

86
00:10:06,510 --> 00:10:12,720
we have some action that the agent should take in the interaction with the human.

87
00:10:13,140 --> 00:10:19,379
And usually if you have some nice virtual human around, it really helps the interaction with people.

88
00:10:19,380 --> 00:10:36,240
But this is not our main issues. So for example, if we are thinking about the automated agent negotiate that negotiated with the people,

89
00:10:37,470 --> 00:10:46,890
we had this aspect, we have data that we use both to predict whether people will accept an offer or not.

90
00:10:47,730 --> 00:10:55,230
And also we use a data because people with using chat we need to recognise what they are saying in the negotiation.

91
00:10:55,320 --> 00:11:03,299
We also use a data in machine learning, but then we had some optimisation problem to solve.

92
00:11:03,300 --> 00:11:07,080
What will be the best strategy of the negotiation?

93
00:11:07,200 --> 00:11:12,360
So our agent will get the best result from our point of view.

94
00:11:12,750 --> 00:11:19,829
And we also use the human behaviour models so such that it will help in some

95
00:11:19,830 --> 00:11:25,080
decision of the agent that we can solve by solving an optimisation problem.

96
00:11:25,950 --> 00:11:29,670
So if we just want to see our system.

97
00:11:30,480 --> 00:11:38,940
So I was offered a job at a role and I am going asked to come in and discuss the terms of my planning.

98
00:11:39,450 --> 00:11:43,260
Okay, so let me just show you what is doing.

99
00:11:44,220 --> 00:11:52,380
The logic behind negotiating is based on two values rational theory anchoring and aspiration adaptation theory.

100
00:11:57,480 --> 00:12:02,730
At the start of this negotiation session, negotiators present the whole offer.

101
00:12:03,900 --> 00:12:13,320
While we don't expect the human counterpart to accept this offer, it does provide an anchor or a basis for further negotiation offers.

102
00:12:18,790 --> 00:12:29,229
The agent incrementally focuses on one issue at a time based on which issue is the next most important or aspire to the employers.

103
00:12:29,230 --> 00:12:48,010
Most important value is salary. Negotiators showed flexibility in negotiating for an inspired value salary and tries to find a counteroffer

104
00:12:48,370 --> 00:12:54,730
that yields a similar utility value to itself while incorporating the employer's request for salary.

105
00:12:56,340 --> 00:13:04,889
Okay. So that, you know, if you the young people, if you need to train for way negotiating about salary, you can log into the system.

106
00:13:04,890 --> 00:13:08,670
It's online and try and train yourself.

107
00:13:09,180 --> 00:13:13,079
Anyway, so that was one aspect in negotiation.

108
00:13:13,080 --> 00:13:17,940
In the cases I show you say domain was not that complex.

109
00:13:17,940 --> 00:13:25,200
We had it well in advance. But in other settings it's not just we need to predict the human behaviour,

110
00:13:25,470 --> 00:13:32,070
we need also to predict things about the environment and again we need the model.

111
00:13:32,070 --> 00:13:37,050
So we need the model for the human and we need the model for the environment.

112
00:13:37,290 --> 00:13:40,800
And again, we have data, we need to collect data on both.

113
00:13:41,010 --> 00:13:52,110
We have the machine learning method to build the models and then it goes into the optimisation of the agent and they the agent takes in actions.

114
00:13:52,680 --> 00:13:59,370
So I want to present in this talk two examples that demonstrate the use of this methodology.

115
00:14:00,120 --> 00:14:05,190
One is about providing arguments in discussion, and it's with my student.

116
00:14:06,060 --> 00:14:11,160
I really wasn't filled. And the story was like this.

117
00:14:11,610 --> 00:14:17,939
People from Intel came to me a few years ago and they wanted to develop an agent

118
00:14:17,940 --> 00:14:23,250
that will whispering in the ear of people when they are doing discussions.

119
00:14:23,790 --> 00:14:27,869
And they said, you know, you are you have this non-Newtonian reasoning.

120
00:14:27,870 --> 00:14:31,200
That was the basic full argumentation theory.

121
00:14:31,410 --> 00:14:35,850
Why want to use argumentation theory to build such an agent?

122
00:14:36,480 --> 00:14:49,110
And the student was very excited. And he went and out many, many, many papers about beautiful, really nice series about argumentation.

123
00:14:49,590 --> 00:14:52,799
And we read the papers and we said, okay, we'll use this.

124
00:14:52,800 --> 00:14:59,190
But before we use it, I told him, you know, I know that in game series it's not games.

125
00:14:59,190 --> 00:15:07,830
There is not a good predictive models for people behaviour before we are using argumentation theory for our model, let's check that.

126
00:15:08,730 --> 00:15:13,440
That people are following album one of this series of argumentation theory.

127
00:15:14,040 --> 00:15:23,550
So what we did, we collected, we took from the literature six fictional cases that appeared in the literature of argumentation theory.

128
00:15:23,910 --> 00:15:31,590
We put this, we collected people from Amazon Turk, and we let them negotiate.

129
00:15:31,830 --> 00:15:41,820
And this and you can see it in the, in the format of argumentation theory, you have nodes which are saying, well, is this thing.

130
00:15:42,720 --> 00:15:57,430
This appeared. I remind you have the Zen nodes are a the nodes of the argument and they can attack or support a.

131
00:15:58,780 --> 00:16:08,379
It's the argument and there's always some a ways in argumentation theory to decide what is a strong and strong argument.

132
00:16:08,380 --> 00:16:14,950
And he is the wife is saying, let's buy SUV and the husband or the other way round.

133
00:16:15,130 --> 00:16:18,670
Osborne said, Let's buy a movie. And the wife said, It's too expensive.

134
00:16:18,670 --> 00:16:24,890
And so we asked him, what will be the next argument? You will say, so you can say, Well, SUV safe.

135
00:16:25,420 --> 00:16:30,300
We said, Yeah, it's too expensive because it's taxes.

136
00:16:30,820 --> 00:16:36,379
Or you can say, I'm taking a loan or you can say, Well, it's acting, taking loans.

137
00:16:36,380 --> 00:16:41,860
That is high interest. Okay, so what will you say in such a situation?

138
00:16:43,080 --> 00:16:48,470
So we collected what people said, and most of them are choices.

139
00:16:48,480 --> 00:16:51,570
They say argument. You know, let's take a loan.

140
00:16:52,540 --> 00:17:03,850
And it turned out that any argumentation series that you use will give very low weight to this argument.

141
00:17:04,180 --> 00:17:07,420
And as I said, 35% of the people chose this one.

142
00:17:07,540 --> 00:17:11,199
But this is just an example, and this is really low regard.

143
00:17:11,200 --> 00:17:14,800
This was extension use or numbers or any.

144
00:17:15,160 --> 00:17:19,630
So that wasn't that nice. I was very disappointed. The student was extremely disappointed.

145
00:17:20,440 --> 00:17:26,090
So I said, Well, okay. He said, this is just fictional arguments.

146
00:17:27,160 --> 00:17:30,880
You know, why won't we go and look at real argumentation?

147
00:17:30,970 --> 00:17:41,590
So the student found a real found a database of transcripts of argumentation, people, arguments, people said about various topics.

148
00:17:41,590 --> 00:17:45,970
And it shows two topics capital punishment and trial by jury.

149
00:17:46,820 --> 00:17:56,320
And this was a database from 1995. And he checked whether the arguments that people said belong to extension extensions

150
00:17:56,320 --> 00:18:01,719
is a concept in argumentation theory of what are good arguments that will be,

151
00:18:01,720 --> 00:18:13,090
say, in the series. And it turned out that only less than 45% of the arguments said in the discussion belong to any extension.

152
00:18:13,480 --> 00:18:19,690
And that was extremely discouraging because this was really a real discussions.

153
00:18:20,590 --> 00:18:32,380
So yeah. So just to summarise this, we have fictional cases of 142 student people with transcripts of these people,

154
00:18:32,530 --> 00:18:44,409
and we have also students that did chat with a 72 people doing a chat and most of them didn't follow argumentation.

155
00:18:44,410 --> 00:18:47,970
Sui. So what can be done?

156
00:18:48,680 --> 00:18:52,570
Okay, I said, let's use our methodology. What's the methodology said?

157
00:18:52,590 --> 00:18:59,970
Let's try to predict what people will say in their argumentation.

158
00:19:00,000 --> 00:19:02,790
This will be the first time, first thing you know.

159
00:19:03,150 --> 00:19:14,130
So we collected data from Amazon Turk and also computer science student in Israel, and they did both this and others.

160
00:19:14,520 --> 00:19:20,190
And we tried to find features about this argumentation.

161
00:19:20,190 --> 00:19:27,410
And one thing that we did, the features that we used were based on argumentation.

162
00:19:27,420 --> 00:19:30,920
Issawi Like justification that was.

163
00:19:31,500 --> 00:19:40,350
But we also took feature form psychology and also new concepts that we developed, which called relevance.

164
00:19:40,470 --> 00:19:44,070
How this argument is close to the previous arguments.

165
00:19:44,640 --> 00:19:48,060
Look, this is a tree. Okay.

166
00:19:48,420 --> 00:19:52,040
And then we will able to do quite well.

167
00:19:52,050 --> 00:19:56,820
This is us. It depends how many. How many time I saw you.

168
00:19:56,850 --> 00:19:57,450
So if I.

169
00:19:57,460 --> 00:20:08,850
So, for example, full response of the of the person I quite well know how to predict the fifth time, the fifth argument that you will propose.

170
00:20:09,210 --> 00:20:13,530
And these are the other like a random or a.

171
00:20:14,770 --> 00:20:21,280
As a going according to the a major. This is Emily and we did quite well.

172
00:20:21,280 --> 00:20:25,209
Interestingly enough, people around we did it in two countries.

173
00:20:25,210 --> 00:20:30,820
People from both countries didn't do exactly the same.

174
00:20:30,820 --> 00:20:32,680
But I can use the model.

175
00:20:33,040 --> 00:20:44,469
Let's say the model I learned in the computer science student and I got a 77% accuracy to the Amazon tech and I got 72% accuracy,

176
00:20:44,470 --> 00:20:48,880
which is really nice that it transferable between countries.

177
00:20:50,780 --> 00:20:57,620
We did the same thing prediction in the capital punishment and we again got quite good the prediction model.

178
00:20:58,240 --> 00:21:01,970
So but as I told you, prediction is just the first step.

179
00:21:02,270 --> 00:21:12,860
How do I use a prediction to make our argument to to suggest to people which argument to say next in the deliberation.

180
00:21:13,700 --> 00:21:23,240
So. We said we look at different a possibility one just two or fields in the prediction.

181
00:21:24,500 --> 00:21:29,030
You know, I will see what is a best prediction and then I will choose one of them.

182
00:21:29,360 --> 00:21:36,940
Well, I was very reluctant to do it because I said, you know, if a person is thinking about an argument, I really like my argument.

183
00:21:36,950 --> 00:21:41,190
But, you know, I don't understand in people I just understanding predictions so.

184
00:21:41,210 --> 00:21:46,720
Okay. So we said the concept of relevance will clearly related.

185
00:21:46,730 --> 00:21:52,700
I was thinking that we look at the graph, each of them is based on some mathematical models,

186
00:21:52,700 --> 00:21:58,910
but I'm just summarising things that are far away from the argument because these are innovative arguments.

187
00:22:00,290 --> 00:22:09,830
We have some you always think of prediction proofs relevant. I said, okay, I would say something forms argumentation theory and see if it's left it.

188
00:22:10,070 --> 00:22:12,980
We compare it with not saying anything and with random.

189
00:22:13,820 --> 00:22:24,620
So if 204 participants in this experiment and they you can see this is the ones that did the best forms, the acceptance rates.

190
00:22:24,830 --> 00:22:32,770
If I give you a suggestion what to say in the discussion, will you accept my proposal?

191
00:22:32,780 --> 00:22:39,650
So this is the first, the average you said. So the prediction proofs relevance did the best.

192
00:22:40,130 --> 00:22:45,440
And so I do a prediction and I have several arguments that can be said.

193
00:22:45,440 --> 00:22:55,339
I chose according to what events and the and then there was only relevance it was doing well and also the prediction weakly related,

194
00:22:55,340 --> 00:22:57,530
which was for my surprise, this divorce.

195
00:22:57,800 --> 00:23:05,930
I was thinking if there is an argument that I didn't think of, I would like to hear it as a suggestion, but people didn't like it.

196
00:23:07,440 --> 00:23:15,890
And similarly, if we are looking at the satisfaction of the people from the they got again, they like this one.

197
00:23:16,400 --> 00:23:26,780
Okay. So I was quite happy with this approach that we can help people in the discussion, but really what they want.

198
00:23:26,990 --> 00:23:31,220
I want an agent that can convince people to do something.

199
00:23:32,190 --> 00:23:41,100
As I said, you'll remember the cake and say, okay, so we said we have, we have a methodology.

200
00:23:41,640 --> 00:23:45,330
We, we need human argumentative behaviour.

201
00:23:46,350 --> 00:23:55,979
I said model for me, argumentation. We build a optimisation problem and let's see how we do it.

202
00:23:55,980 --> 00:23:59,309
So we started with the argumentation series.

203
00:23:59,310 --> 00:24:02,880
This is a we combine here two models.

204
00:24:03,180 --> 00:24:14,250
One is the most theory and one is also based on weights and and each of them give different possible conclusions.

205
00:24:14,520 --> 00:24:17,820
So we chose one of them according to our experience.

206
00:24:18,090 --> 00:24:20,640
So this is how the model is looks.

207
00:24:20,940 --> 00:24:31,230
So we have a arguments in the notes and we have a attacks in support, but we have also weights on the attacks and support.

208
00:24:31,620 --> 00:24:43,530
And we also had weights on the argument if it stayed by itself, that is its critical model and it has some advantage of this way that we build it.

209
00:24:43,800 --> 00:24:50,340
Now, if we want to have it with people, we need to know where these numbers will come from.

210
00:24:51,350 --> 00:24:55,710
So. So we took two domains.

211
00:24:55,920 --> 00:24:59,310
One was about convincing computer science masters.

212
00:25:00,370 --> 00:25:03,400
Computer Science Undergraduate Student two.

213
00:25:04,720 --> 00:25:09,220
To do a master degree in Israeli fields, do a master's degrees and a Ph.D.

214
00:25:09,490 --> 00:25:17,890
So that was a good day. And also this sick issue, just that, you know, in this experiment, people came to the lab,

215
00:25:18,160 --> 00:25:26,410
they said what they want a a a a bar, a healthy bar or a chocolate cake.

216
00:25:26,420 --> 00:25:30,940
And then we tried to convince them the other way around. And when they left, they got one of the things.

217
00:25:30,940 --> 00:25:34,419
We brought them to the lab and they got the cake all they had.

218
00:25:34,420 --> 00:25:37,300
See? Well, I'm not. So is that healthy? But never mind.

219
00:25:40,420 --> 00:25:50,650
So first we collected human dialogues A to build these, say, possible trees or graphs for argumentation theory.

220
00:25:51,040 --> 00:26:02,019
And then we also collected question we, we gave people questionnaires about how strong these things attacks, what diseases sing about argumentation.

221
00:26:02,020 --> 00:26:14,400
And we collected a lot of data to build this is. Interestingly enough in this say augmentations it's let them just to collect data.

222
00:26:14,640 --> 00:26:27,080
It turns out that they people use 33% of the student used in the argumentation at least two arguments that attacked one another.

223
00:26:28,840 --> 00:26:36,820
What can I say? So we have argumentation framework 12 human argumentation, argumentative behaviour.

224
00:26:37,450 --> 00:26:48,490
And then we used also machine learning to predict what some distribution of the arguments that the person will say next given the last day,

225
00:26:48,700 --> 00:26:51,970
a is a sequence of arguments that he said so far.

226
00:26:52,270 --> 00:26:55,360
So we use this data also for prediction.

227
00:26:57,100 --> 00:27:01,990
So we have all this report now. We came to the optimisation problem.

228
00:27:02,110 --> 00:27:05,540
Now, this was tough. What do we have?

229
00:27:05,560 --> 00:27:12,910
What do we know about these people? We almost don't know anything about them because what is their argumentation framework?

230
00:27:12,940 --> 00:27:15,440
We don't know what is their lie.

231
00:27:16,090 --> 00:27:30,480
So we build pom pom the p that the nodes are states of our belief of states as they are argumentation graph of the persons.

232
00:27:31,000 --> 00:27:34,000
And we updated it over time.

233
00:27:34,300 --> 00:27:40,670
And they because we have observation. The observation is what the argument that the person really said.

234
00:27:40,690 --> 00:27:55,690
So this is a good indication about his model of argumentation and that we used all the prediction and probabilities for buildings upon dip.

235
00:27:56,500 --> 00:28:02,950
And now given that wave bomb the P, we solved it and we got a well, is it safe to solve it?

236
00:28:02,950 --> 00:28:08,290
Because it was not easy to solve, you know, these huge bomb dips.

237
00:28:08,680 --> 00:28:18,009
And we use the Monte Carlo search tree and some smart optimisation problems, solutions to solve the problem.

238
00:28:18,010 --> 00:28:27,160
The P and Z gave us the strategy of the agent in our mentation, convincing people.

239
00:28:27,280 --> 00:28:32,259
Did it help? Well, we ran this experiment with new people.

240
00:28:32,260 --> 00:28:38,110
Of course not. The people that we checked earlier and they are lucky to us.

241
00:28:38,680 --> 00:28:52,240
We were able to show that in the master degree we did as well as the people that try to convince their friends.

242
00:28:52,690 --> 00:29:00,370
And actually in the chocolate cake, we were doing better, significantly better that using the baseline,

243
00:29:00,730 --> 00:29:13,090
the ones it was following some argumentation theory that was the purpose for agents because no one did with humans didn't do very well.

244
00:29:13,450 --> 00:29:24,040
I must say that in general, as you noticed, the percentage of our ability to convince another person to change his mind is extremely difficult.

245
00:29:24,520 --> 00:29:32,200
And they, while we are doing at least as good as people trying to convince people to change their mind is not that easy.

246
00:29:32,410 --> 00:29:39,130
Now we have a project where we have sleep people and each try to change a person that doesn't have an opinion.

247
00:29:39,430 --> 00:29:51,110
So if I don't care if I have a cake or a a bar, it's easier to convince me to take one of them, but they so will try.

248
00:29:51,160 --> 00:29:55,690
So this is the ongoing work and we'll see how we are doing in this case.

249
00:29:56,950 --> 00:30:00,040
So that was one example.

250
00:30:00,040 --> 00:30:05,980
I wanted to show you how we develop and use this methodology that we had.

251
00:30:06,460 --> 00:30:10,990
I want to say to show another a project.

252
00:30:11,290 --> 00:30:15,040
Well, we had also to model the environment.

253
00:30:15,580 --> 00:30:19,629
And this has to do with a also with robots.

254
00:30:19,630 --> 00:30:27,370
And this is the work of also a Rosenfeld and Lagman and Oleg Maxime of Amazonia.

255
00:30:28,390 --> 00:30:32,170
So think about now. Move to another things.

256
00:30:32,410 --> 00:30:35,290
Forget about argumentation and discussion.

257
00:30:35,620 --> 00:30:45,940
Just think about one person trying to manage some whole boats so it can be a soldier, but it can be a drones.

258
00:30:46,750 --> 00:30:58,750
And they these days with drones usually and these robots, you have one person on one drone or one person that manage if they are doing small things.

259
00:30:58,870 --> 00:31:07,060
Even if the drones are autonomous UAV, I don't know if you know they are two people on one UAV these days.

260
00:31:07,450 --> 00:31:14,770
And the challenge was, can we put one person on autonomous robots that are cheap?

261
00:31:16,020 --> 00:31:26,280
And the it's really problem. If you see this is my lab and the robots are moving around and you should come to my lab and try to manage ten robots.

262
00:31:26,550 --> 00:31:31,830
The robots are looking for health, for green balls. So you say this is autonomous robots.

263
00:31:32,070 --> 00:31:36,890
Why do I care? Legit robot move around. But the robot's the problem also.

264
00:31:37,110 --> 00:31:42,240
But, you know, backing to some places and sometimes, you know,

265
00:31:42,510 --> 00:31:48,600
these robots like to get into the ladies room in my lab and then going out is quite difficult.

266
00:31:49,030 --> 00:31:52,829
And while while they can do it,

267
00:31:52,830 --> 00:32:01,350
eventually a human operator can easily manoeuvre them out and they are on their way to continue searching for the green balls.

268
00:32:01,830 --> 00:32:04,980
Oh. Sometimes they went out of battery.

269
00:32:05,130 --> 00:32:16,100
Sometimes. And in addition, still, there are several decisions that people don't want the robot to make, but want that as a person.

270
00:32:16,620 --> 00:32:17,909
When they find the green birds,

271
00:32:17,910 --> 00:32:26,640
they want it to get a confirmation that they say this is really a green ball and not just a box, a green box or something.

272
00:32:27,540 --> 00:32:36,960
So the problem is that the resistant robot's moving around and the person that is trying to operate them is getting overwhelmed.

273
00:32:37,800 --> 00:32:50,040
And given that, the question is, can we have an agent that will support the operator and will help the robot's ends operate to get better results?

274
00:32:50,040 --> 00:32:53,340
In our case, find more green balls.

275
00:32:54,540 --> 00:32:59,130
So this is just that. You see the robots. These are the robots are.

276
00:33:02,180 --> 00:33:05,510
And I was out looking for this kid involved in my lab.

277
00:33:07,050 --> 00:33:11,129
And they are very cheap. They cost less than 1500.

278
00:33:11,130 --> 00:33:23,120
The. And this is the interface of the.

279
00:33:24,930 --> 00:33:29,690
Operator. Okay. We'll go back to these sailboats in a minute.

280
00:33:30,110 --> 00:33:33,970
There would be another movie. So engine design.

281
00:33:33,980 --> 00:33:38,060
We need a model for the humans. Who is a woman? He's the operator.

282
00:33:38,450 --> 00:33:41,930
We need the model for the robots because he's the environment.

283
00:33:42,140 --> 00:33:47,550
We don't know how we will behave. So we collected data on the human behaviour.

284
00:33:47,570 --> 00:33:49,550
We collected data on the robot performance.

285
00:33:49,890 --> 00:33:58,880
This the idea was that it's really difficult to collect data when the robots are moving around, running out of battery and.

286
00:33:59,630 --> 00:34:03,530
So the interesting thing here is the machine learning point of view was that

287
00:34:03,530 --> 00:34:09,380
we collected data from the simulation and then we deployed it on real robots.

288
00:34:09,740 --> 00:34:21,290
So that was quite nice. And then we have an optimisation problem to solve, to set and solve, to give the best advice to the operator.

289
00:34:21,300 --> 00:34:28,520
What should we do next? That was the advice we gave him, and this turned out to be extremely useful.

290
00:34:29,360 --> 00:34:42,740
So we have 150 hours of simulation. And so human operators that came to the lab and then after we have the agent, we ran sleep experiments.

291
00:34:44,060 --> 00:34:54,080
We in the in the simulations. But the most interesting experiment was with bringing real people to the lab twice in a week in between.

292
00:34:54,350 --> 00:35:00,889
Once they managed the robots by themself, once they met the robot with the help of the agent.

293
00:35:00,890 --> 00:35:07,760
Have did this first have the Zaza sync first just to be fair to all of them and they.

294
00:35:08,980 --> 00:35:12,700
Uh, agent. Really? Hips are on average.

295
00:35:13,030 --> 00:35:18,280
Zay Roberts found seven balls when they operate.

296
00:35:18,330 --> 00:35:31,940
Operator didn't have the help of Zay. Of the agent and 14 balls when it did have the help of Z a operator as the agents operated the help of the.

297
00:35:32,420 --> 00:35:40,250
And it's not just that it was an average would it help all the people that came to the labs, the agent help all of them?

298
00:35:40,730 --> 00:35:50,340
And I must tell you, you know, I'm not z good. The manoeuvring robots, but we have also quite good operators and you can try it.

299
00:35:50,360 --> 00:35:53,690
It's really helping. Are people doing it?

300
00:35:54,170 --> 00:36:00,649
So let me show you another video demonstrating why I have this.

301
00:36:00,650 --> 00:36:10,160
I must tell you why we have this video. Because my student, they wanted to do a video to send to each guy competition on a video for robots.

302
00:36:10,550 --> 00:36:15,200
And I said, come on, you know, this has nothing to do with research.

303
00:36:15,650 --> 00:36:23,060
But they were very excited. I said, okay, I will give you some money, but don't come to me for anything about this stupid video.

304
00:36:23,270 --> 00:36:27,110
And they were in first place. So what can I say?

305
00:36:34,770 --> 00:36:38,940
This past week, there's a student missing in this section of the campus.

306
00:36:39,150 --> 00:36:43,290
Even the replication unit is largely.

307
00:36:59,240 --> 00:37:04,640
In disaster and violence. Both the victims and the rescuers lives may be in danger.

308
00:37:05,300 --> 00:37:07,700
Robot technology can provide solutions.

309
00:37:10,440 --> 00:37:17,820
Today's robots are semi-autonomous and require a human offering to help whenever they cannot handle the situation on their own.

310
00:37:18,480 --> 00:37:23,210
Agent three away from. Agent four moved. Agent four Away from agent and UAV drive.

311
00:37:23,310 --> 00:37:32,730
Robot six to a fast paced simultaneous streams of data from multiple robots can be overwhelming and can be a gruelling task for a single operator.

312
00:37:32,730 --> 00:37:35,100
The robots are too close together and spread them around.

313
00:37:35,490 --> 00:37:41,910
The robots are too close together and spread them around and usually drive robot snakes to a better position to agents.

314
00:37:44,370 --> 00:37:51,900
Here in Maryland University, we have developed an intelligent agent that supports the human multi robot team collaboration.

315
00:37:52,830 --> 00:37:56,150
The agent prioritises the different class structures.

316
00:37:56,160 --> 00:38:04,740
The robot ignores and generates learning functionality, offering situational awareness, thereby allowing you to make better decisions.

317
00:38:07,060 --> 00:38:12,070
The agent uses a novel methodology combining healing and optimisation,

318
00:38:13,120 --> 00:38:20,890
which accounts for both the operator's ability and robust performance in real time through extensive empirical evaluations.

319
00:38:21,370 --> 00:38:33,030
We have shown that our aim was able to significantly enhance the team's performance, which can translate into being loved, says some of the guys.

320
00:38:36,200 --> 00:38:42,570
I'll send you an application out. Just a shout.

321
00:38:53,520 --> 00:38:58,950
Okay. So we actually use this approach in another domain,

322
00:38:59,160 --> 00:39:13,410
which is a storage domain where you have a robot that brings the items to the operators that package them and heals the person,

323
00:39:13,410 --> 00:39:17,040
needs bots to pack and to manage the robots.

324
00:39:17,550 --> 00:39:20,960
And this is extremely difficult.

325
00:39:20,970 --> 00:39:30,420
We had the simulation about robot shelf boxes packing and the transit agent was helping the operator.

326
00:39:30,890 --> 00:39:39,240
And again, we did a we adhere to algorithms with solving the optimisation problem.

327
00:39:39,510 --> 00:39:51,180
And you can see that we are people are doing extremely better when they have the agent is advisor is us advising what to do.

328
00:39:52,500 --> 00:40:05,580
So if I will summarise we have automated agent can help people, can replace people and they can be used to train people.

329
00:40:05,970 --> 00:40:12,270
I demonstrate a few of these, a project that we did.

330
00:40:12,660 --> 00:40:18,160
It's a difficult task to build this agent and we need a lot of things.

331
00:40:18,180 --> 00:40:21,900
We need machine learning algorithms. We need optimisation.

332
00:40:22,080 --> 00:40:25,530
We need to understand interaction and use game theory models.

333
00:40:26,310 --> 00:40:30,090
But putting it all together. I'm having fun.

334
00:40:30,750 --> 00:40:31,290
Thank you.