1
00:00:00,040 --> 00:00:01,400
Speaker 1: So a fun fact for you, each year of the

2
00:00:01,480 --> 00:00:05,360
Speaker 1: airline industry globally loses thirteen billion dollars to fog. So

3
00:00:05,400 --> 00:00:08,840
Speaker 1: we now have technology that predicts fog, predicts fog and

4
00:00:08,880 --> 00:00:12,360
Speaker 1: prevents flight delays. Local tech piper Vision are. The founder

5
00:00:12,400 --> 00:00:13,480
Speaker 1: was Emily blythe who's with us?

6
00:00:13,520 --> 00:00:16,200
Speaker 2: Emily, good morning, Good morning, Mike cal Y.

7
00:00:16,440 --> 00:00:18,920
Speaker 1: I'm well, thank you. So this is in the sense

8
00:00:18,960 --> 00:00:21,320
Speaker 1: that you're predicting, not trying to clear, right, So there's

9
00:00:21,360 --> 00:00:24,040
Speaker 1: two ways of going about this, that's correct.

10
00:00:24,120 --> 00:00:26,920
Speaker 2: Yeah, predicting is going to be our first stage an approach.

11
00:00:27,000 --> 00:00:29,440
Speaker 1: Now it's where are you at in the process. I mean,

12
00:00:29,480 --> 00:00:31,200
Speaker 1: if you've got a product set to go, you close

13
00:00:31,240 --> 00:00:32,080
Speaker 1: to where are you at?

14
00:00:32,960 --> 00:00:37,400
Speaker 2: Yeah, So we've built an initial forecasting model that can

15
00:00:37,440 --> 00:00:40,480
Speaker 2: predict fog much more accurately than what traditional models can.

16
00:00:41,479 --> 00:00:43,080
Speaker 2: It's still got a lot to go in terms of

17
00:00:43,200 --> 00:00:46,400
Speaker 2: lifting performance. We need additional data sets to plug inn there,

18
00:00:46,400 --> 00:00:48,880
Speaker 2: and we've turned up with the team ad Attentive Technologies

19
00:00:48,960 --> 00:00:52,440
Speaker 2: to colleck that data, which is really exciting. But we

20
00:00:52,440 --> 00:00:55,440
Speaker 2: should be ready to have or start making impact for

21
00:00:55,520 --> 00:00:57,600
Speaker 2: the public early twenty twenty.

22
00:00:57,360 --> 00:00:59,760
Speaker 1: Six with what level of accuracy.

23
00:01:01,240 --> 00:01:05,080
Speaker 2: So at the moment, with existing data only, we're performing

24
00:01:05,280 --> 00:01:08,000
Speaker 2: at a fifty percent jump on traditional models, So New

25
00:01:08,080 --> 00:01:12,759
Speaker 2: Zealand accuracy is around nineteen percent. Traditionally we're at thirty

26
00:01:12,800 --> 00:01:16,440
Speaker 2: two percent with existing data only, but we're going to

27
00:01:16,440 --> 00:01:19,240
Speaker 2: be adding in a lot more spatial awareness at ground

28
00:01:19,319 --> 00:01:21,759
Speaker 2: level two which will help lift that performance much better.

29
00:01:21,840 --> 00:01:24,399
Speaker 2: So I'm hoping to get way better than a queen flip.

30
00:01:25,600 --> 00:01:28,760
Speaker 1: So nineteen percent currently, In other words, you've got a

31
00:01:28,800 --> 00:01:30,560
Speaker 1: far greater chance of being completely wrong.

32
00:01:32,080 --> 00:01:36,240
Speaker 2: Yeah yeah, gosh, So on those days where fogs yeah

33
00:01:36,280 --> 00:01:36,720
Speaker 2: due to form?

34
00:01:36,800 --> 00:01:39,920
Speaker 1: Yeah yeah yeah. Is generative AI involved that? Does that

35
00:01:40,040 --> 00:01:42,640
Speaker 1: solve everything? Or not?

36
00:01:42,640 --> 00:01:46,240
Speaker 2: Not quite. What we're using is we're wanting to see

37
00:01:46,360 --> 00:01:50,160
Speaker 2: Traditional models are using numerical weather models to predict fogs,

38
00:01:50,160 --> 00:01:53,240
Speaker 2: so that relies on our own scientific understanding of how

39
00:01:53,360 --> 00:01:56,720
Speaker 2: fog forms at a local level, and when our own

40
00:01:56,800 --> 00:02:00,240
Speaker 2: understanding is not very great, the calculations and mess that

41
00:02:00,320 --> 00:02:03,680
Speaker 2: what the model can achieve isn't isn't very good either.

42
00:02:04,440 --> 00:02:08,120
Speaker 2: So by switching through to a machine learning model, we

43
00:02:08,200 --> 00:02:11,280
Speaker 2: can actually start to kind of harness AI to teach

44
00:02:11,400 --> 00:02:14,000
Speaker 2: us what are the patterns that we're recognizing, and then

45
00:02:14,000 --> 00:02:17,120
Speaker 2: we can intupulate that data to actually learn more about

46
00:02:17,120 --> 00:02:18,040
Speaker 2: sold longer term.

47
00:02:18,720 --> 00:02:22,399
Speaker 1: How in advance can you potentially do it? Because it's

48
00:02:22,400 --> 00:02:25,400
Speaker 1: the in advance but that I'm assuming allows an airport

49
00:02:25,480 --> 00:02:26,440
Speaker 1: or an airline to react.

50
00:02:27,400 --> 00:02:30,239
Speaker 2: Yeah, so the airline schedules are set up to respond

51
00:02:30,360 --> 00:02:36,119
Speaker 2: super super quickly, which is great. And we're aiming initially

52
00:02:36,280 --> 00:02:41,000
Speaker 2: for a three hour forecast horizon with real time data,

53
00:02:41,000 --> 00:02:44,200
Speaker 2: so providing sort of minutely updates coming through just to

54
00:02:44,240 --> 00:02:47,200
Speaker 2: give a bit of context there. Current forecasts are only

55
00:02:47,280 --> 00:02:50,880
Speaker 2: updated every three to six hours and so that all

56
00:02:50,919 --> 00:02:52,960
Speaker 2: gives them a lot better information around whether or not

57
00:02:53,000 --> 00:02:55,760
Speaker 2: those flights can change, and within the New Zealand market

58
00:02:56,919 --> 00:03:00,000
Speaker 2: it should. It will give the airlines in our time

59
00:03:00,040 --> 00:03:02,720
Speaker 2: in order to make those flight changes needed, and we'll

60
00:03:02,760 --> 00:03:03,960
Speaker 2: slightly grow that up from there.

61
00:03:04,160 --> 00:03:05,919
Speaker 1: Super exciting. I'm going to get you back in twenty

62
00:03:05,919 --> 00:03:07,320
Speaker 1: twenty six and we'll have a good talk about it

63
00:03:07,360 --> 00:03:08,160
Speaker 1: when it's all set to go.

64
00:03:08,200 --> 00:03:10,560
Speaker 2: How's that sounds great? Thank you?

65
00:03:10,720 --> 00:03:13,519
Speaker 1: Hardly to talk to Emily go well, Emily blythe I

66
00:03:13,560 --> 00:03:15,040
Speaker 1: don't know how old she is. I do know how

67
00:03:15,080 --> 00:03:17,079
Speaker 1: old you. She's in a twenty she's another one of

68
00:03:17,120 --> 00:03:21,760
Speaker 1: these young new Zealand freak shows your parents would be

69
00:03:21,840 --> 00:03:24,760
Speaker 1: unbelievably proud of. But talk about rip the top off

70
00:03:24,760 --> 00:03:28,160
Speaker 1: that we've all always suspected, which is what they haven't

71
00:03:28,160 --> 00:03:31,079
Speaker 1: got a clue. They haven't the slightest idea. Next time

72
00:03:31,080 --> 00:03:34,320
Speaker 1: you see somebody predicting exactly there might be fog patches

73
00:03:34,360 --> 00:03:37,760
Speaker 1: tomorrow morning, nineteen percent, just turn off your TV exactly right.

74
00:03:38,240 --> 00:03:40,600
Speaker 1: You think about the amount of broadcasting time that's been

75
00:03:40,600 --> 00:03:43,960
Speaker 1: given to what we now know to be literally nothing

76
00:03:44,000 --> 00:03:46,960
Speaker 1: more than guesswork, and nineteen percent charts a big It's

77
00:03:47,000 --> 00:03:50,840
Speaker 1: about as accurate as the profit made by the Prime

78
00:03:50,880 --> 00:03:54,800
Speaker 1: Minister on a house. Welcome to Media, twenty twenty four.

79
00:03:54,840 --> 00:03:58,560
Speaker 1: Styles For more from the mi Casking Breakfast, listen live

80
00:03:58,680 --> 00:04:01,560
Speaker 1: to news talks it'd be from six am weekdays, or

81
00:04:01,600 --> 00:04:03,520
Speaker 1: follow the podcast on iHeartRadio