Future of maritime autonomy: cybersecurity, trust and mariner’s situational awareness

Key Question Being Explored in This Insight:

As the maritime industry moves toward fully autonomous vessels, the shift from onboard crews to remote operations introduces new risks, including reduced Situational Awareness (SA) and cyber threats. Kevin Jones highlights the need for future navigators to develop new digital skills for remote monitoring, supervision, and cyber incident management.

View the page of: Kevin Jones

Institution: University of Plymouth

Explore more insights in:

Electronic Engineering Engineering Professional Software Engineering

Media Literacy Toolkit

Faculti's Media Literacy Toolkit helps viewers critically engage with academic insights by analyzing the research context, identifying perspectives, and encouraging thoughtful evaluation.

Critical Questions to Consider ▼

What assumptions does the research make?
Are there alternative perspectives not explored?
What are the limitations of the research method?

Bias and Perspective Awareness ▼

This research comes from University of Plymouth . Reflect on how the institution's academic focus and research partnerships may shape the questions being explored.

Further Reading ▼

Read the Study

Image courtesy of interviewee. March 26, 2025

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

Language:

Export as PDF

X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:90000

1
00:00:00.065 --> 00:00:04.245
The IMO is aware that regulations as they currently stand,

2
00:00:04.425 --> 00:00:08.565
are very much based on the idea that there's a human

3
00:00:09.175 --> 00:00:12.085
who's effectively the master of the vessel and in command.

4
00:00:12.945 --> 00:00:15.165
Almost all the regulations for,

5
00:00:15.355 --> 00:00:17.525
well probably the past few hundred years,

6
00:00:18.395 --> 00:00:20.805
have assumed human oversight.

7
00:00:21.315 --> 00:00:24.925
Many of the regulations explicitly refer to an officer

8
00:00:24.985 --> 00:00:29.165
of the watch or a person sort of standing on the bridge

9
00:00:29.985 --> 00:00:32.525
Now for autonomous ships, of course that's not true.

10
00:00:33.505 --> 00:00:35.565
And one of the things that the IMO have

11
00:00:35.565 --> 00:00:39.045
to address is the principle underlying

12
00:00:39.045 --> 00:00:41.125
that which is appropriate oversight.

13
00:00:41.755 --> 00:00:44.005
What does that mean in an environment

14
00:00:44.015 --> 00:00:46.245
where the actual controller of the vessel

15
00:00:46.785 --> 00:00:50.365
is effectively a piece of software, an AI in the,

16
00:00:50.505 --> 00:00:53.205
in the sense of an autonomous, craft.

17
00:00:54.485 --> 00:00:55.965
I think they've been looking to address

18
00:00:55.965 --> 00:00:57.605
that in two different ways.

19
00:00:59.045 --> 00:01:02.525
Existing regulations, particularly those sort of overseen

20
00:01:02.525 --> 00:01:06.165
by the Maritime Safety Committee, can be modified

21
00:01:06.705 --> 00:01:11.165
to interpret oversight as being oversight

22
00:01:11.165 --> 00:01:13.045
of the controlling ai,

23
00:01:13.825 --> 00:01:16.805
or they can be interpreted to be oversight

24
00:01:17.265 --> 00:01:19.405
by a remote human operator.

25
00:01:20.605 --> 00:01:23.445
I think most of the work to date has assumed

26
00:01:23.915 --> 00:01:28.045
that a remote human operator will still fundamentally

27
00:01:28.625 --> 00:01:31.805
be in, in ultimately in control of the vessel.

28
00:01:32.625 --> 00:01:35.725
Um, regulations have have been

29
00:01:35.785 --> 00:01:38.685
and are being rewritten to specify

30
00:01:38.955 --> 00:01:41.125
that remote op remote operation

31
00:01:41.695 --> 00:01:44.805
still allows the appropriate oversight of a master,

32
00:01:45.615 --> 00:01:49.205
which means that instead of changing the regulations to date

33
00:01:49.265 --> 00:01:53.325
to allow no human oversight, they've been changed to

34
00:01:53.875 --> 00:01:56.285
effectively say there is oversight.

35
00:01:56.285 --> 00:01:58.725
Nothing has changed in the fundamental regulation,

36
00:01:59.145 --> 00:02:01.565
but that oversight may be remote rather

37
00:02:01.565 --> 00:02:02.765
than physically present.

38
00:02:03.545 --> 00:02:06.205
And I think that's probably the first step

39
00:02:06.745 --> 00:02:08.445
in safe integration.

40
00:02:08.955 --> 00:02:12.925
What it ultimately means is that it is still a human being

41
00:02:12.995 --> 00:02:16.765
that has final authority over the operation of the vessel.

42
00:02:17.585 --> 00:02:22.045
Now, of course, that partially solves the problem in

43
00:02:22.045 --> 00:02:24.245
that you no longer require physical presence,

44
00:02:24.545 --> 00:02:25.965
but it still means you have to have a human

45
00:02:25.965 --> 00:02:27.365
being involved in the loop.

46
00:02:27.365 --> 00:02:29.565
Somewhere as we move further

47
00:02:29.745 --> 00:02:33.325
and further into autonomy, we are going to want less

48
00:02:33.325 --> 00:02:35.085
and less oversight because

49
00:02:35.085 --> 00:02:38.845
after all, one of the purposes of autonomy is efficiencies

50
00:02:38.865 --> 00:02:42.485
of not needing human beings to be present, even if

51
00:02:42.485 --> 00:02:43.525
that presence is remote.

52
00:02:43.985 --> 00:02:47.045
So I think eventually there will have to be adjustment

53
00:02:47.045 --> 00:02:51.205
to regulation to effectively address the degree

54
00:02:51.205 --> 00:02:53.525
of confidence we have in the system

55
00:02:54.185 --> 00:02:56.205
as the system is making decisions.

56
00:02:56.905 --> 00:02:59.725
Um, the IMO have a committee working

57
00:03:00.325 --> 00:03:01.925
specifically on mass.

58
00:03:02.785 --> 00:03:06.205
Um, that committee feeds into the Maritime Safety Committee,

59
00:03:06.625 --> 00:03:08.925
so effectively is at the highest level

60
00:03:09.465 --> 00:03:13.325
of the IMOs operations to ensure safety at sea.

61
00:03:14.205 --> 00:03:15.805
I think we're seeing the beginnings

62
00:03:15.805 --> 00:03:17.645
of regulatory change today.

63
00:03:18.245 --> 00:03:20.045
I think we'll see over the course of

64
00:03:20.805 --> 00:03:23.245
probably the next decade to be honest, that

65
00:03:23.245 --> 00:03:24.845
that will continue to adjust

66
00:03:25.415 --> 00:03:27.965
until eventually we'll have removed a lot

67
00:03:27.965 --> 00:03:31.285
of the old bias from the regulations

68
00:03:31.745 --> 00:03:33.445
and replace them more with a sort

69
00:03:33.445 --> 00:03:38.205
of an operation in principle rather than sort of the letter

70
00:03:38.305 --> 00:03:40.405
of the law as it's been for a long time.

71
00:03:40.675 --> 00:03:42.525
It's very much a work in progress.

72
00:03:43.345 --> 00:03:46.325
It it's quite clear from, you know, years

73
00:03:46.345 --> 00:03:47.565
of practical operation,

74
00:03:47.705 --> 00:03:50.885
but also experiments done in various control settings

75
00:03:51.435 --> 00:03:54.885
that it's not easy to quantify all the information

76
00:03:54.905 --> 00:03:59.205
that's used by a mariner, particularly in sort of situations

77
00:03:59.205 --> 00:04:02.765
of danger or emergency, that there are a lot of people

78
00:04:02.865 --> 00:04:05.925
who will say, we know what we do, we know what we look at.

79
00:04:06.375 --> 00:04:08.685
Turns out if you do a controlled experiment that's not true,

80
00:04:09.025 --> 00:04:10.805
people will tell you, you know,

81
00:04:10.805 --> 00:04:13.205
the most critical instrumentation, for example,

82
00:04:13.495 --> 00:04:16.045
electronic charting system, you know,

83
00:04:16.105 --> 00:04:20.685
second most critical information system radar in a condition

84
00:04:20.685 --> 00:04:24.165
of emergency, many crew members will then focus their

85
00:04:24.165 --> 00:04:27.285
attention on things like depth, sounders and compass.

86
00:04:27.985 --> 00:04:32.845
So there is quite a lot of unknown interactions

87
00:04:32.955 --> 00:04:35.005
between the human and the vessel.

88
00:04:35.615 --> 00:04:37.965
We're not entirely clear on exactly

89
00:04:38.515 --> 00:04:40.925
what situational information is necessary.

90
00:04:41.595 --> 00:04:43.325
When you go to remote operation

91
00:04:43.545 --> 00:04:46.525
or semi-autonomous, you are in, in a position

92
00:04:46.525 --> 00:04:49.165
where the only information you get is

93
00:04:49.165 --> 00:04:53.285
that which has been decided sort of a priority is needed,

94
00:04:53.745 --> 00:04:56.765
the feed coming from the, the vessel for example.

95
00:04:57.505 --> 00:05:01.285
Now we know that cameras of course make a huge difference,

96
00:05:01.795 --> 00:05:03.045
sort of self-evident,

97
00:05:03.545 --> 00:05:05.205
but what do you need to see through the cameras?

98
00:05:05.785 --> 00:05:08.885
Do you need just the view sort of over the bow

99
00:05:09.425 --> 00:05:12.165
or do you actually need as you would in sort

100
00:05:12.165 --> 00:05:14.725
of in physical presence to be able

101
00:05:14.725 --> 00:05:15.845
to look over your shoulder

102
00:05:16.425 --> 00:05:19.525
and see, I don't know, cables are flapping or something.

103
00:05:20.225 --> 00:05:22.525
So I think it's, it's very much a,

104
00:05:22.625 --> 00:05:26.605
an exploration at the moment to ensure that all

105
00:05:26.605 --> 00:05:29.765
of the information including the, the sort

106
00:05:29.765 --> 00:05:32.045
of almost implicit information,

107
00:05:32.425 --> 00:05:34.485
if you imagine you're standing on a bridge of a ship

108
00:05:34.545 --> 00:05:36.365
and something is not going right,

109
00:05:36.865 --> 00:05:39.965
you might hear a grinding coming from sort

110
00:05:39.965 --> 00:05:42.205
of the engine room or from the steering gear

111
00:05:42.225 --> 00:05:45.085
or something that might tell an experience.

112
00:05:45.235 --> 00:05:48.365
Mariner, I know the problem is physical.

113
00:05:48.885 --> 00:05:50.645
I know the problem is, for example,

114
00:05:50.705 --> 00:05:54.125
in the propulsion system, if all you have to work

115
00:05:54.125 --> 00:05:58.645
with is the instrumentation, it might be possible to deduce

116
00:05:58.645 --> 00:06:01.005
that same thing, but it might take longer.

117
00:06:01.465 --> 00:06:04.165
You might have to look at various pieces of information

118
00:06:04.665 --> 00:06:07.645
and sort of almost by inference say, okay,

119
00:06:07.685 --> 00:06:08.965
I think something is wrong.

120
00:06:09.475 --> 00:06:12.445
Sort of with some of the bearings on the main prop shaft

121
00:06:12.505 --> 00:06:15.845
or something, if you're on the ship you feel a

122
00:06:15.875 --> 00:06:17.485
vibration that isn't right.

123
00:06:18.065 --> 00:06:21.645
So I think because of that we are working towards

124
00:06:22.235 --> 00:06:25.285
what information do you need in a remote operation center

125
00:06:25.865 --> 00:06:27.965
to not lose the ability

126
00:06:28.345 --> 00:06:30.365
to make good decisions in a timely manner.

127
00:06:30.815 --> 00:06:33.045
There are obvious things that you need

128
00:06:33.305 --> 00:06:36.845
and most remote operation centers today will have,

129
00:06:37.745 --> 00:06:38.885
an instrument panel

130
00:06:38.955 --> 00:06:42.325
that effectively reflects everything you could see on the

131
00:06:42.325 --> 00:06:45.485
bridge that I think is clear and known.

132
00:06:45.895 --> 00:06:49.125
There are various cameras showing various different views,

133
00:06:49.455 --> 00:06:51.085
which again give you a lot

134
00:06:51.085 --> 00:06:53.325
of the information you would have inid

135
00:06:53.915 --> 00:06:56.325
what we're less clear on yet as sort of the

136
00:06:57.685 --> 00:07:00.025
the second order pieces of information

137
00:07:00.095 --> 00:07:03.345
that you pick up not realizing it's critical

138
00:07:03.805 --> 00:07:08.345
and I don't think we know yet exactly how to duplicate that.

139
00:07:09.325 --> 00:07:12.025
The second thing is when you're physically present,

140
00:07:12.765 --> 00:07:15.025
you have a fair degree of trust in what you see.

141
00:07:15.565 --> 00:07:17.745
The kind of classic mark one eyeball.

142
00:07:17.845 --> 00:07:19.825
You are looking out the window, you see it.

143
00:07:20.445 --> 00:07:23.185
If what you are seeing is an image on a screen in front

144
00:07:23.185 --> 00:07:26.265
of you, do you have the same level of trust?

145
00:07:26.925 --> 00:07:29.985
Do you believe that you were seeing everything?

146
00:07:30.645 --> 00:07:33.505
Do you believe that there's no compromise on

147
00:07:33.505 --> 00:07:36.745
that information and can you do that with the same level

148
00:07:36.745 --> 00:07:38.825
of confidence you could if you're physically present.

149
00:07:39.585 --> 00:07:42.785
I think at the moment the answer is to some degree yes,

150
00:07:43.205 --> 00:07:45.065
but to a complete degree, which

151
00:07:45.165 --> 00:07:48.025
of course matters more in times of emergency

152
00:07:48.365 --> 00:07:51.025
or sort of tight maneuvering or something.

153
00:07:51.635 --> 00:07:54.385
Maybe that's still slightly an open question.

154
00:07:55.045 --> 00:07:58.545
So I think that the issue we have at the moment is given

155
00:07:58.575 --> 00:08:00.985
that of course there's bandwidth required

156
00:08:01.045 --> 00:08:04.785
to transmit information remotely, what do we need to see?

157
00:08:05.455 --> 00:08:07.825
What do we need to see in normal operation?

158
00:08:08.215 --> 00:08:10.745
What do we need to see in emergency situations

159
00:08:11.125 --> 00:08:13.945
and how do we make sure there's no loss

160
00:08:13.965 --> 00:08:15.905
of fidelity in decision making

161
00:08:16.215 --> 00:08:19.945
because we are remote rather than physically present?

162
00:08:20.945 --> 00:08:23.225
I think there are whole classes of errors

163
00:08:23.575 --> 00:08:26.585
that will be removed by autonomous operation,

164
00:08:27.405 --> 00:08:30.785
but there are also classes of errors that'll be introduced

165
00:08:30.925 --> 00:08:32.385
by autonomous operation.

166
00:08:33.395 --> 00:08:36.745
Think in particular the kind of errors that are caused

167
00:08:37.005 --> 00:08:38.425
by being tired

168
00:08:38.685 --> 00:08:41.465
or just missing a piece of information, which

169
00:08:41.465 --> 00:08:44.985
of course occur a lot when humans are in charge,

170
00:08:45.225 --> 00:08:49.025
particularly sort of if they've been on duty for a long time

171
00:08:49.565 --> 00:08:52.985
or are slightly unwell, that whole class should go away.

172
00:08:53.315 --> 00:08:55.805
Those kind of errors should not be,

173
00:08:55.805 --> 00:08:57.805
committed by a machine.

174
00:08:58.745 --> 00:09:01.085
On the other hand, there are often situations

175
00:09:01.085 --> 00:09:05.245
where a human being will just say something is not right.

176
00:09:06.065 --> 00:09:09.845
You know, almost intuitively the pieces are not adding up,

177
00:09:10.035 --> 00:09:13.005
therefore we know something is not quite right.

178
00:09:14.285 --> 00:09:17.315
Mechanical systems are more properly, software systems

179
00:09:18.085 --> 00:09:19.875
don't make those kind of decisions.

180
00:09:20.225 --> 00:09:23.155
They literally follow the procedure

181
00:09:23.665 --> 00:09:26.115
that they have been taught is right.

182
00:09:26.775 --> 00:09:30.555
In normal circumstances that will be executed flawlessly.

183
00:09:30.935 --> 00:09:32.155
So life is better.

184
00:09:32.815 --> 00:09:35.035
But in circumstances that are anomalous

185
00:09:35.445 --> 00:09:39.395
where say the training set for, for an AI based system

186
00:09:40.135 --> 00:09:42.635
has not covered that properly

187
00:09:42.865 --> 00:09:46.355
because you know the classic once in a thousand years kind

188
00:09:46.355 --> 00:09:49.675
of situation, then autonomy might make the wrong decision.

189
00:09:50.375 --> 00:09:52.595
Now that's just pure decision making.

190
00:09:53.105 --> 00:09:55.795
When we start to introduce sort of malicious actors,

191
00:09:56.615 --> 00:10:00.915
if we look at things like cyber attacks on vessels, one

192
00:10:00.915 --> 00:10:02.035
of the easiest ways

193
00:10:02.175 --> 00:10:05.835
to introduce cyber vulnerabilities into a ship is

194
00:10:05.835 --> 00:10:09.955
to get a person to do the wrong thing, either accidentally

195
00:10:09.975 --> 00:10:13.075
by misleading them or deliberately by bribing

196
00:10:13.075 --> 00:10:14.075
or blackmailing them.

197
00:10:14.575 --> 00:10:17.395
So again, with an autonomous system, that class

198
00:10:17.415 --> 00:10:20.795
of things sort of don't happen in the same way.

199
00:10:21.655 --> 00:10:25.235
On the other hand, an experienced crew might recognize

200
00:10:25.835 --> 00:10:27.155
symptoms of a cyber attack,

201
00:10:27.155 --> 00:10:30.075
particularly if they've been properly trained to know what

202
00:10:30.075 --> 00:10:33.515
that looks like and take mitigation action much earlier.

203
00:10:33.945 --> 00:10:36.715
Whereas an autonomous system would again,

204
00:10:36.985 --> 00:10:39.675
only operate within its normal parameters.

205
00:10:40.375 --> 00:10:43.675
Now we can build in extra security up

206
00:10:43.675 --> 00:10:45.635
to whatever level one wishes to,

207
00:10:46.255 --> 00:10:47.875
but of course that adds cost

208
00:10:48.575 --> 00:10:50.955
and again, in general, one of the reasons for

209
00:10:51.675 --> 00:10:54.275
autonomy becoming more prevalent is to save cost.

210
00:10:54.695 --> 00:10:56.875
So there's a definite trade off here.

211
00:10:57.285 --> 00:10:58.435
There are new risks

212
00:10:58.935 --> 00:11:01.795
and in particular, I mean, interesting example,

213
00:11:02.455 --> 00:11:06.075
if you look at a chart, if you as a mariner look at a chart,

214
00:11:06.815 --> 00:11:10.075
you are seeing a collection of lines and colors

215
00:11:10.415 --> 00:11:14.395
and text which you interpret in a particular way.

216
00:11:15.615 --> 00:11:17.515
If I were to give you a chart

217
00:11:17.515 --> 00:11:20.435
of The Bahamas when you were sailing in UK waters,

218
00:11:20.885 --> 00:11:22.555
you'd say this is wrong.

219
00:11:23.725 --> 00:11:26.265
You wouldn't trust the information you're given.

220
00:11:27.185 --> 00:11:29.145
A proper chart for an autonomous vessel

221
00:11:29.725 --> 00:11:31.465
is a collection of ones and zeroes.

222
00:11:31.495 --> 00:11:32.825
It's binary information.

223
00:11:33.805 --> 00:11:36.665
It will take that information as is.

224
00:11:37.685 --> 00:11:39.775
There's no sort of meta knowledge of

225
00:11:40.285 --> 00:11:43.615
this is not right in the way that a human would do.

226
00:11:44.275 --> 00:11:47.615
The other problem there is you would not be able to read

227
00:11:47.615 --> 00:11:49.735
that chart and so no,

228
00:11:49.845 --> 00:11:51.935
that it's not the right chart for the mission.

229
00:11:52.475 --> 00:11:55.085
That's a class of problem we don't have with people.

230
00:11:55.905 --> 00:11:59.925
So I think overall we will probably get less

231
00:12:01.375 --> 00:12:03.335
accidents with autonomous

232
00:12:03.335 --> 00:12:06.095
Systems, But there are different

233
00:12:06.095 --> 00:12:09.335
Classes of vulnerabilities to malicious action

234
00:12:09.875 --> 00:12:10.935
and of course we'll have

235
00:12:10.935 --> 00:12:12.575
to look at those in a different way.

236
00:12:13.155 --> 00:12:15.695
The question even today is how much do you

237
00:12:15.695 --> 00:12:17.775
Trust what the system is Telling you?

238
00:12:18.635 --> 00:12:22.415
And an observation that we've made doing a number of sort of

239
00:12:22.565 --> 00:12:24.095
Experiments with mariners

240
00:12:24.205 --> 00:12:26.935
Ranging from cadets right through

241
00:12:26.935 --> 00:12:30.295
to very experienced captains, they actually have different

242
00:12:30.295 --> 00:12:32.455
Levels of trust in different parts of the system.

243
00:12:33.455 --> 00:12:35.895
A cadet particularly Sort of, you know, somebody who

244
00:12:35.945 --> 00:12:38.255
Today is 18 to 20 years old

245
00:12:38.885 --> 00:12:40.535
Will often believe in the

246
00:12:40.535 --> 00:12:42.295
System as is shown on the screen

247
00:12:42.795 --> 00:12:43.895
As if it was the truth.

248
00:12:44.245 --> 00:12:47.095
Whereas an older mariner is much more likely to say,

249
00:12:48.155 --> 00:12:49.295
I'm not sure about

250
00:12:49.295 --> 00:12:50.295
That. You know, the

251
00:12:50.295 --> 00:12:51.695
classic statement is,

252
00:12:51.695 --> 00:12:52.815
well, I'm gonna look out the window

253
00:12:52.995 --> 00:12:54.655
And just see if this makes sense.

254
00:12:55.235 --> 00:12:56.255
So there is trust

255
00:12:56.655 --> 00:12:59.775
Involved even in, you know, human

256
00:12:59.785 --> 00:13:01.255
Based operations today.

257
00:13:01.915 --> 00:13:03.735
Um, one of the worrying things

258
00:13:03.795 --> 00:13:05.415
Is there is more likelihood

259
00:13:05.435 --> 00:13:08.015
to trust systems without verification

260
00:13:08.435 --> 00:13:09.775
In less experienced people.

261
00:13:10.475 --> 00:13:14.335
Um, and I suppose over time that will kind of balance

262
00:13:14.515 --> 00:13:16.255
Out as people get more experience.

263
00:13:17.035 --> 00:13:19.415
Um, as we move into more autonomy,

264
00:13:19.925 --> 00:13:22.575
there's a different kind of trust that's

265
00:13:22.895 --> 00:13:27.095
Required Because there is none of this, you know, kind

266
00:13:27.095 --> 00:13:30.895
of intuitive holistic view that a human being has.

267
00:13:31.305 --> 00:13:32.655
Human beings are very good

268
00:13:32.655 --> 00:13:35.055
At picking up patterns, even if they don't quite know

269
00:13:35.085 --> 00:13:36.935
What or why most

270
00:13:36.935 --> 00:13:38.935
Systems are not, they just follow the rules

271
00:13:39.035 --> 00:13:40.175
as given to some extent.

272
00:13:40.875 --> 00:13:44.615
Um, I think that confusing a human

273
00:13:45.515 --> 00:13:48.525
is possible confusing a system

274
00:13:49.585 --> 00:13:52.485
is in some ways easier if you have access.

275
00:13:53.455 --> 00:13:54.805
We've done experimentation

276
00:13:54.805 --> 00:13:56.645
where we've introduced things like

277
00:13:56.845 --> 00:13:58.925
GPS spoofing usually in simulators,

278
00:13:59.465 --> 00:14:01.165
And we spoof the GPS signal

279
00:14:01.345 --> 00:14:03.085
to see if we can misplace the ship.

280
00:14:03.875 --> 00:14:05.725
It's interesting to see how far

281
00:14:06.265 --> 00:14:08.685
you can get the ship to move from

282
00:14:08.825 --> 00:14:12.645
Its assigned course Before somebody says, hang on,

283
00:14:12.675 --> 00:14:14.045
something's not quite right.

284
00:14:15.265 --> 00:14:16.805
The reason they usually say

285
00:14:16.805 --> 00:14:17.845
Something is not quite

286
00:14:17.845 --> 00:14:22.085
Right is something else is telling them that

287
00:14:22.595 --> 00:14:24.125
this, this is not correlating

288
00:14:24.125 --> 00:14:25.125
With reality.

289
00:14:25.625 --> 00:14:28.765
Now we tend to look for correlation of information

290
00:14:29.275 --> 00:14:31.245
when we suspect something is wrong.

291
00:14:32.025 --> 00:14:35.245
You know, you're following, an actus chart

292
00:14:35.865 --> 00:14:38.565
and you have a navigation line and everything looks fine,

293
00:14:39.465 --> 00:14:40.685
but if you notice

294
00:14:40.875 --> 00:14:43.605
That the depth you were seeing on a depth sounder

295
00:14:43.985 --> 00:14:46.525
Is not Correlating to what you'd expect on the chart,

296
00:14:47.025 --> 00:14:49.445
You then begin to look differently

297
00:14:50.025 --> 00:14:51.605
To bring in Extra information

298
00:14:51.625 --> 00:14:53.165
to correlate what you think you're seeing.

299
00:14:53.865 --> 00:14:58.045
An autonomous system doesn't have that ability to say,

300
00:14:58.825 --> 00:15:01.205
I'm not sure I can trust this anymore.

301
00:15:01.865 --> 00:15:05.765
So something like GPS spoofing, whether it's done sort

302
00:15:05.765 --> 00:15:08.365
of radio frequency to change the input

303
00:15:08.825 --> 00:15:11.485
or whether it's done by malware on the vessel

304
00:15:12.025 --> 00:15:13.685
to give misleading information,

305
00:15:14.745 --> 00:15:18.165
an autonomous system would be much more likely to take that

306
00:15:18.305 --> 00:15:21.245
as the truth than a human system would.

307
00:15:21.625 --> 00:15:23.445
You know, going to an extreme example,

308
00:15:24.105 --> 00:15:26.405
if you see you're about to run a ground

309
00:15:26.865 --> 00:15:29.965
and you a charting system tells you you're in the middle

310
00:15:29.985 --> 00:15:32.085
of the channel, you are going

311
00:15:32.185 --> 00:15:34.045
to not believe the charting system.

312
00:15:34.865 --> 00:15:38.005
An autonomous system would be much more likely

313
00:15:38.505 --> 00:15:41.565
to take the information as given within the system

314
00:15:42.185 --> 00:15:44.485
and less of course it's much more sophisticated

315
00:15:44.945 --> 00:15:48.725
and has some kind of crosscheck correlation.

316
00:15:49.345 --> 00:15:53.125
So I think we have to think about building those things in

317
00:15:53.705 --> 00:15:57.405
as we develop the kind of autonomous system that we want

318
00:15:57.405 --> 00:15:59.565
to have more genuine autonomy.

319
00:15:59.945 --> 00:16:01.965
So higher degrees of autonomy using the,

320
00:16:02.065 --> 00:16:05.725
the amount terminology to support remote operation

321
00:16:06.305 --> 00:16:10.085
is different from the training required to be sort of a,

322
00:16:10.345 --> 00:16:11.725
an officer of the watch on the bridge.

323
00:16:12.345 --> 00:16:14.365
It has some points of commonality

324
00:16:14.945 --> 00:16:16.205
and I'm sure the sort

325
00:16:16.205 --> 00:16:19.325
of the maritimes mariners in the world would absolutely

326
00:16:19.505 --> 00:16:22.645
insist that all of the prerequisites for

327
00:16:23.205 --> 00:16:25.805
shipboard operation still have to be covered.

328
00:16:26.145 --> 00:16:29.445
And I think that's true, but there are other things in

329
00:16:29.445 --> 00:16:33.005
particular, as we talked about earlier, the level of trust

330
00:16:33.065 --> 00:16:35.725
and the information you're given, being able

331
00:16:35.745 --> 00:16:38.045
to recognize when there are particular kinds

332
00:16:38.045 --> 00:16:40.845
of compromise in that information stream.

333
00:16:41.655 --> 00:16:44.925
Those have to be built into the training at quite a

334
00:16:44.925 --> 00:16:45.925
fundamental level.

335
00:16:46.335 --> 00:16:47.645
We've been doing some work

336
00:16:47.755 --> 00:16:51.205
with the MCA in the UK Maritime Coast Guard agency,

337
00:16:51.675 --> 00:16:55.765
basically on the training syllabus for, you know,

338
00:16:55.765 --> 00:16:59.795
maritime operators in the current standards.

339
00:17:00.055 --> 00:17:01.715
Things like cybersecurity,

340
00:17:02.095 --> 00:17:05.275
remote operation are at best barely mentioned

341
00:17:06.245 --> 00:17:07.355
after going through

342
00:17:07.375 --> 00:17:09.915
and thinking about where should

343
00:17:09.985 --> 00:17:12.395
that really occur for the future.

344
00:17:13.255 --> 00:17:16.475
It turns out that in almost every part of the syllabus

345
00:17:17.085 --> 00:17:20.835
there needs to be a mention of cybersecurity

346
00:17:21.015 --> 00:17:24.675
or trust in instrumentation just so that that,

347
00:17:24.675 --> 00:17:27.995
that mariners are aware of the possible,

348
00:17:28.775 --> 00:17:30.595
symptoms, if you will.

349
00:17:31.045 --> 00:17:33.235
Being able to tell something is not right.

350
00:17:33.905 --> 00:17:37.435
Also being aware of the possible consequences of

351
00:17:38.155 --> 00:17:41.795
a cyber attack or even if just instrument failure

352
00:17:42.415 --> 00:17:45.435
or instrument degradation in a way that

353
00:17:45.435 --> 00:17:48.035
of course doesn't occur with physical presence again.

354
00:17:48.575 --> 00:17:50.645
So I think, think that we need

355
00:17:51.145 --> 00:17:54.125
to revise the entire training framework

356
00:17:54.745 --> 00:17:58.325
to not lose the essential elements of maritime operation,

357
00:17:58.945 --> 00:18:02.845
but to add in the essential elements of remote

358
00:18:02.945 --> 00:18:05.925
or autonomy at the moment I have a colleague who's working

359
00:18:05.945 --> 00:18:10.765
on building a training framework, particularly around sort

360
00:18:10.765 --> 00:18:14.485
of, integrating human

361
00:18:14.705 --> 00:18:19.525
and autonomous systems training, the situation

362
00:18:19.785 --> 00:18:24.205
of sort of, you know, genuine human autonomy teaming

363
00:18:24.785 --> 00:18:29.205
so that we are not trying to make either side independent

364
00:18:29.865 --> 00:18:33.565
but bring them together so that we can, if we get it right,

365
00:18:33.585 --> 00:18:35.205
get the best of both worlds.

366
00:18:35.595 --> 00:18:38.365
Because it's also necessary, I think to remember that

367
00:18:38.465 --> 00:18:43.165
for the foreseeable future, any training framework is going

368
00:18:43.225 --> 00:18:47.645
to have to address autonomous vessels, crude vessels,

369
00:18:48.025 --> 00:18:52.565
remote operated vessels, all in the same framework

370
00:18:52.715 --> 00:18:55.245
because they will all be interacting with each other.

371
00:18:56.005 --> 00:18:58.445
I think it will be a quite significant change

372
00:18:58.745 --> 00:19:01.765
to the material people are expected to know,

373
00:19:02.545 --> 00:19:04.645
but probably also the mindset

374
00:19:05.505 --> 00:19:09.365
and the awareness of what could go wrong,

375
00:19:09.795 --> 00:19:13.205
because that's changed quite dramatically once you've gone

376
00:19:13.225 --> 00:19:14.965
to sort of an autonomy environment.

377
00:19:15.425 --> 00:19:18.685
And most existing crews wouldn't know

378
00:19:19.315 --> 00:19:22.725
what bad looks like, so wouldn't know when

379
00:19:22.785 --> 00:19:24.645
to take appropriate operation.

380
00:19:25.485 --> 00:19:28.845
I don't think it's necessary for every mariner

381
00:19:29.145 --> 00:19:33.005
to be a cybersecurity expert or an autonomy expert,

382
00:19:33.545 --> 00:19:37.205
but what we do need are triggers built into processes

383
00:19:37.205 --> 00:19:39.965
and procedures so that are various points.

384
00:19:40.465 --> 00:19:43.765
An appropriate crew member can say, effectively,

385
00:19:44.115 --> 00:19:46.925
I've now moved outside my domain of competence.

386
00:19:47.415 --> 00:19:49.045
Maybe I can recognize

387
00:19:49.075 --> 00:19:51.805
that something is failing in the autonomy space,

388
00:19:52.305 --> 00:19:53.525
but I don't know what to do about it.

389
00:19:53.905 --> 00:19:58.125
But I do know who to contact to get that level

390
00:19:58.185 --> 00:19:59.965
of advice to move forward.

391
00:20:00.785 --> 00:20:03.925
In particular, if I use cyber as an example, you know,

392
00:20:04.095 --> 00:20:06.725
today something goes wrong with a a piece

393
00:20:06.725 --> 00:20:09.445
of bridge instrumentation, there's a likelihood

394
00:20:09.445 --> 00:20:12.165
that somebody may turn it off and turn it back on again

395
00:20:12.165 --> 00:20:13.885
because of course that's what you do.

396
00:20:14.505 --> 00:20:16.445
If on the other hand, we're proper training,

397
00:20:16.625 --> 00:20:20.325
you recognize a particular kind of cyber attack, you don't

398
00:20:20.395 --> 00:20:23.245
because you wanna freeze the forensic information

399
00:20:23.795 --> 00:20:26.005
that when you contact an expert,

400
00:20:26.805 --> 00:20:29.045
probably a sure side incidents response team,

401
00:20:29.595 --> 00:20:32.445
they have information available to know what

402
00:20:32.445 --> 00:20:34.845
to do next in the immediate situation,

403
00:20:35.465 --> 00:20:38.125
but also how to prevent that going forward.

404
00:20:38.745 --> 00:20:42.245
So I think it's, it's a level of awareness that has

405
00:20:42.245 --> 00:20:44.965
to permeate the Hulu maritime domain.

406
00:20:45.345 --> 00:20:50.285
So the remote operation becomes one more thread of basically

407
00:20:50.845 --> 00:20:52.845
maritime safety in general.

Quick Answers

00:00:00 - The IMO is aware that regulations as they currently stand, are very much based on the idea that there's a human who's effectively the master of the vessel and in command. Almost all the regulations for, well probably the past few hundred years, have assumed human oversight. Many of the regulations explicitly refer to an officer of the watch or a person sort of standing on the bridge Now for autonomous ships, of course that's not true.

00:00:35 - to address is the principle underlying that which is appropriate oversight. What does that mean in an environment where the actual controller of the vessel is effectively a piece of software, an AI in the, in the sense of an autonomous, craft. I think they've been looking to address that in two different ways. Existing regulations, particularly those sort of overseen by the Maritime Safety Committee, can be modified

00:01:11 - of the controlling ai, or they can be interpreted to be oversight by a remote human operator. I think most of the work to date has assumed that a remote human operator will still fundamentally be in, in ultimately in control of the vessel. regulations have have been and are being rewritten to specify

00:01:41 - still allows the appropriate oversight of a master, which means that instead of changing the regulations to date to allow no human oversight, they've been changed to effectively say there is oversight. Nothing has changed in the fundamental regulation, but that oversight may be remote rather than physically present. And I think that's probably the first step in safe integration.

00:02:12 - that has final authority over the operation of the vessel. Now, of course, that partially solves the problem in that you no longer require physical presence, but it still means you have to have a human being involved in the loop. Somewhere as we move further and further into autonomy, we are going to want less and less oversight because after all, one of the purposes of autonomy is efficiencies of not needing human beings to be present, even if

00:02:43 - So I think eventually there will have to be adjustment to regulation to effectively address the degree of confidence we have in the system as the system is making decisions. the IMO have a committee working specifically on mass. that committee feeds into the Maritime Safety Committee, so effectively is at the highest level of the IMOs operations to ensure safety at sea.

00:03:15 - of regulatory change today. I think we'll see over the course of probably the next decade to be honest, that that will continue to adjust until eventually we'll have removed a lot of the old bias from the regulations and replace them more with a sort of an operation in principle rather than sort of the letter of the law as it's been for a long time. It's very much a work in progress. It it's quite clear from, you know, years

00:03:47 - but also experiments done in various control settings that it's not easy to quantify all the information that's used by a mariner, particularly in sort of situations of danger or emergency, that there are a lot of people who will say, we know what we do, we know what we look at. Turns out if you do a controlled experiment that's not true, people will tell you, you know, the most critical instrumentation, for example, electronic charting system, you know, second most critical information system radar in a condition

00:04:24 - attention on things like depth, sounders and compass. So there is quite a lot of unknown interactions between the human and the vessel. We're not entirely clear on exactly what situational information is necessary. When you go to remote operation or semi-autonomous, you are in, in a position where the only information you get is that which has been decided sort of a priority is needed,

00:04:57 - Now we know that cameras of course make a huge difference, sort of self-evident, but what do you need to see through the cameras? Do you need just the view sort of over the bow or do you actually need as you would in sort of in physical presence to be able to look over your shoulder and see, I don't know, cables are flapping or something. So I think it's, it's very much a, an exploration at the moment to ensure that all

00:05:29 - of almost implicit information, if you imagine you're standing on a bridge of a ship and something is not going right, you might hear a grinding coming from sort of the engine room or from the steering gear or something that might tell an experience. Mariner, I know the problem is physical. I know the problem is, for example, in the propulsion system, if all you have to work with is the instrumentation, it might be possible to deduce

00:06:01 - You might have to look at various pieces of information and sort of almost by inference say, okay, I think something is wrong. Sort of with some of the bearings on the main prop shaft or something, if you're on the ship you feel a vibration that isn't right. So I think because of that we are working towards what information do you need in a remote operation center to not lose the ability to make good decisions in a timely manner.

00:06:33 - and most remote operation centers today will have, an instrument panel that effectively reflects everything you could see on the bridge that I think is clear and known. There are various cameras showing various different views, which again give you a lot of the information you would have inid what we're less clear on yet as sort of the the second order pieces of information that you pick up not realizing it's critical

00:07:09 - The second thing is when you're physically present, you have a fair degree of trust in what you see. The kind of classic mark one eyeball. You are looking out the window, you see it. If what you are seeing is an image on a screen in front of you, do you have the same level of trust? Do you believe that you were seeing everything? Do you believe that there's no compromise on that information and can you do that with the same level

00:07:39 - I think at the moment the answer is to some degree yes, but to a complete degree, which of course matters more in times of emergency or sort of tight maneuvering or something. Maybe that's still slightly an open question. So I think that the issue we have at the moment is given that of course there's bandwidth required to transmit information remotely, what do we need to see? What do we need to see in normal operation?

00:08:11 - and how do we make sure there's no loss of fidelity in decision making because we are remote rather than physically present? I think there are whole classes of errors that will be removed by autonomous operation, but there are also classes of errors that'll be introduced by autonomous operation. Think in particular the kind of errors that are caused by being tired

00:08:41 - of course occur a lot when humans are in charge, particularly sort of if they've been on duty for a long time or are slightly unwell, that whole class should go away. Those kind of errors should not be, committed by a machine. On the other hand, there are often situations where a human being will just say something is not right. You know, almost intuitively the pieces are not adding up,

00:09:14 - Mechanical systems are more properly, software systems don't make those kind of decisions. They literally follow the procedure that they have been taught is right. In normal circumstances that will be executed flawlessly. So life is better. But in circumstances that are anomalous where say the training set for, for an AI based system

00:09:42 - because you know the classic once in a thousand years kind of situation, then autonomy might make the wrong decision. Now that's just pure decision making. When we start to introduce sort of malicious actors, if we look at things like cyber attacks on vessels, one of the easiest ways to introduce cyber vulnerabilities into a ship is to get a person to do the wrong thing, either accidentally by misleading them or deliberately by bribing

00:10:14 - So again, with an autonomous system, that class of things sort of don't happen in the same way. On the other hand, an experienced crew might recognize symptoms of a cyber attack, particularly if they've been properly trained to know what that looks like and take mitigation action much earlier. Whereas an autonomous system would again, only operate within its normal parameters. Now we can build in extra security up

00:10:46 - but of course that adds cost and again, in general, one of the reasons for autonomy becoming more prevalent is to save cost. So there's a definite trade off here. There are new risks and in particular, I mean, interesting example, if you look at a chart, if you as a mariner look at a chart, you are seeing a collection of lines and colors and text which you interpret in a particular way.

00:11:17 - of The Bahamas when you were sailing in UK waters, you'd say this is wrong. You wouldn't trust the information you're given. A proper chart for an autonomous vessel is a collection of ones and zeroes. It's binary information. It will take that information as is. There's no sort of meta knowledge of this is not right in the way that a human would do. The other problem there is you would not be able to read

00:11:49 - that it's not the right chart for the mission. That's a class of problem we don't have with people. So I think overall we will probably get less accidents with autonomous Systems, But there are different Classes of vulnerabilities to malicious action and of course we'll have to look at those in a different way. The question even today is how much do you Trust what the system is Telling you?

00:12:22 - Experiments with mariners Ranging from cadets right through to very experienced captains, they actually have different Levels of trust in different parts of the system. A cadet particularly Sort of, you know, somebody who Today is 18 to 20 years old Will often believe in the System as is shown on the screen As if it was the truth. Whereas an older mariner is much more likely to say, I'm not sure about

00:12:50 - classic statement is, well, I'm gonna look out the window And just see if this makes sense. So there is trust Involved even in, you know, human Based operations today. one of the worrying things Is there is more likelihood to trust systems without verification In less experienced people. and I suppose over time that will kind of balance Out as people get more experience. as we move into more autonomy,

00:13:22 - Required Because there is none of this, you know, kind of intuitive holistic view that a human being has. Human beings are very good At picking up patterns, even if they don't quite know What or why most Systems are not, they just follow the rules as given to some extent. I think that confusing a human is possible confusing a system is in some ways easier if you have access.

00:13:54 - where we've introduced things like GPS spoofing usually in simulators, And we spoof the GPS signal to see if we can misplace the ship. It's interesting to see how far you can get the ship to move from Its assigned course Before somebody says, hang on, something's not quite right. The reason they usually say Something is not quite Right is something else is telling them that this, this is not correlating

00:14:25 - Now we tend to look for correlation of information when we suspect something is wrong. You know, you're following, an actus chart and you have a navigation line and everything looks fine, but if you notice That the depth you were seeing on a depth sounder Is not Correlating to what you'd expect on the chart, You then begin to look differently To bring in Extra information to correlate what you think you're seeing. An autonomous system doesn't have that ability to say,

00:15:01 - So something like GPS spoofing, whether it's done sort of radio frequency to change the input or whether it's done by malware on the vessel to give misleading information, an autonomous system would be much more likely to take that as the truth than a human system would. You know, going to an extreme example, if you see you're about to run a ground and you a charting system tells you you're in the middle

00:15:32 - to not believe the charting system. An autonomous system would be much more likely to take the information as given within the system and less of course it's much more sophisticated and has some kind of crosscheck correlation. So I think we have to think about building those things in as we develop the kind of autonomous system that we want to have more genuine autonomy. So higher degrees of autonomy using the,

00:16:06 - is different from the training required to be sort of a, an officer of the watch on the bridge. It has some points of commonality and I'm sure the sort of the maritimes mariners in the world would absolutely insist that all of the prerequisites for shipboard operation still have to be covered. And I think that's true, but there are other things in particular, as we talked about earlier, the level of trust

00:16:35 - to recognize when there are particular kinds of compromise in that information stream. Those have to be built into the training at quite a fundamental level. We've been doing some work with the MCA in the UK Maritime Coast Guard agency, basically on the training syllabus for, you know, maritime operators in the current standards. Things like cybersecurity, remote operation are at best barely mentioned

00:17:07 - and thinking about where should that really occur for the future. It turns out that in almost every part of the syllabus there needs to be a mention of cybersecurity or trust in instrumentation just so that that, that mariners are aware of the possible, symptoms, if you will. Being able to tell something is not right. Also being aware of the possible consequences of

00:17:42 - or instrument degradation in a way that of course doesn't occur with physical presence again. So I think, think that we need to revise the entire training framework to not lose the essential elements of maritime operation, but to add in the essential elements of remote or autonomy at the moment I have a colleague who's working on building a training framework, particularly around sort

00:18:14 - and autonomous systems training, the situation of sort of, you know, genuine human autonomy teaming so that we are not trying to make either side independent but bring them together so that we can, if we get it right, get the best of both worlds. Because it's also necessary, I think to remember that for the foreseeable future, any training framework is going

00:18:48 - remote operated vessels, all in the same framework because they will all be interacting with each other. I think it will be a quite significant change to the material people are expected to know, but probably also the mindset and the awareness of what could go wrong, because that's changed quite dramatically once you've gone

00:19:15 - And most existing crews wouldn't know what bad looks like, so wouldn't know when to take appropriate operation. I don't think it's necessary for every mariner to be a cybersecurity expert or an autonomy expert, but what we do need are triggers built into processes and procedures so that are various points. An appropriate crew member can say, effectively,

00:19:47 - Maybe I can recognize that something is failing in the autonomy space, but I don't know what to do about it. But I do know who to contact to get that level of advice to move forward. In particular, if I use cyber as an example, you know, today something goes wrong with a a piece of bridge instrumentation, there's a likelihood that somebody may turn it off and turn it back on again because of course that's what you do.

00:20:16 - you recognize a particular kind of cyber attack, you don't because you wanna freeze the forensic information that when you contact an expert, probably a sure side incidents response team, they have information available to know what to do next in the immediate situation, but also how to prevent that going forward. So I think it's, it's a level of awareness that has to permeate the Hulu maritime domain.

00:20:50 - maritime safety in general.

Faculti AI can make mistakes

Research

Policy

Analysis

Data

Originals

Explainers

Podcasts

Watch the latest episode of The Exchange – the show where influential policymakers and leading researchers engage in insightful discussions shaping the future.

Episode: Brazil's Economic Resilience