WEBVTT

1
00:00:04.000 --> 00:00:09.919
This is Later with Lee Matthews the
Lee Matthews Podcast More What You Here weekday

2
00:00:09.919 --> 00:00:15.240
afternoons on the Drive. Doctor Bard
Costco, no relation to the Big Box

3
00:00:15.279 --> 00:00:19.160
Store, is an attorney and an
engineering professor at USC, also best selling

4
00:00:19.199 --> 00:00:23.640
author and world renowned scientists, scholar
and media commentator. He's also the author

5
00:00:23.719 --> 00:00:27.719
of the best selling book Fuzzy Thinking, probably the best primer around for learning

6
00:00:27.719 --> 00:00:32.240
what Fuzzy Logic is about, being
rereleased on a digital version next week,

7
00:00:32.280 --> 00:00:37.840
and we're having him on to talk
about artificial intelligence because doctor Costco, it's

8
00:00:37.920 --> 00:00:44.079
coming to kill us all. Well, maybe not right away, Okay,

9
00:00:46.280 --> 00:00:50.000
actually quite the opposite. And I'm
kind of on the fence with it.

10
00:00:50.320 --> 00:00:54.759
In general. I see the panic, but then I also see the benefits.

11
00:00:54.759 --> 00:00:59.079
And I'm trying to keep the attitude
that Eisenhower had about nuclear energy.

12
00:00:59.479 --> 00:01:03.920
Yes, we be concerned, but
we mustn't be hysterical. Absolutely, it's

13
00:01:03.959 --> 00:01:07.319
exactly right. And there's many types
of AI, like anything else. What

14
00:01:07.439 --> 00:01:12.560
you're hearing a lot about is the
linguistic type like chat GPT, and that's

15
00:01:12.640 --> 00:01:17.159
something that's got some kinks in it
to be worked out. That was released

16
00:01:17.400 --> 00:01:23.719
chat GPT four in March by Open
AI and its competitors have raised and release

17
00:01:23.840 --> 00:01:26.680
other versions. They're not quite ready
for prime time, and they have a

18
00:01:26.719 --> 00:01:30.439
lot of problems. But there's other
types of AI, but those are more

19
00:01:30.439 --> 00:01:34.640
generally what are called black box AI. Though they learned patterns real well.

20
00:01:36.000 --> 00:01:41.439
But the trouble lead is that they
also make things up and hallucinate and like

21
00:01:41.519 --> 00:01:45.840
for example, you might have seen
the sixty minutes article last episode last week

22
00:01:46.400 --> 00:01:49.760
where they went to Google, which
has a massive system called Palm that's got

23
00:01:49.799 --> 00:01:56.200
over half a trillion parameters, each
parameter like a wire synaps in your brain,

24
00:01:56.920 --> 00:02:01.280
and they asked the question about inflation
of a great answer cited five economics

25
00:02:01.280 --> 00:02:05.680
texts, but it made them up. There aren't such economics texts, and

26
00:02:05.680 --> 00:02:09.159
it didn't tell the user that it
literally lied it call it hallucinated. So

27
00:02:09.400 --> 00:02:15.520
hallucinated, so you have to be
very careful demand an explanation. In contrast,

28
00:02:15.599 --> 00:02:20.360
if you take an AI system which
is mainly statistics and a lot of

29
00:02:20.400 --> 00:02:24.599
computing power and apply it to medical
diagnoses, like what a radiologists would look

30
00:02:24.639 --> 00:02:30.159
at for cancer detection or just whether
you have a retinal problem, it's fantastic

31
00:02:30.240 --> 00:02:34.199
and it's relatively easy to validate those
systems, so we don't have this what

32
00:02:34.240 --> 00:02:38.560
we call XAI or explainable AI problem. Now, fuzzy logic is more often

33
00:02:38.719 --> 00:02:44.120
traditionally applying to control systems, like
for example, I should say fuzzy logic

34
00:02:44.240 --> 00:02:46.479
is thinking with shades of gray,
and it has a rule based structure,

35
00:02:46.840 --> 00:02:51.520
so you get and input output,
and you get a mapping and get some

36
00:02:51.599 --> 00:02:54.240
kind of audit trail and even a
confidence measure which you don't get with the

37
00:02:54.280 --> 00:02:59.039
others. But I have a Subaru
a car, and it uses fuzzy logics

38
00:02:59.080 --> 00:03:02.840
and transmission so many times per second, taking censor debt and making decisions about

39
00:03:02.879 --> 00:03:07.439
whether or how to shift, and
so it doesn't it doesn't get feel a

40
00:03:07.439 --> 00:03:10.639
one shot thing like cat GPT,
and there's many many versions of that.

41
00:03:10.680 --> 00:03:15.120
You probably have a gadget in your
house like that has fuzzyologic in it,

42
00:03:15.159 --> 00:03:19.759
whether a washing machine or even something
as humbles, a microwave, oven,

43
00:03:20.080 --> 00:03:23.159
thermostat. I think my thermostat claims
to have AI, but it's I know,

44
00:03:23.240 --> 00:03:28.439
it's very basic. Yeah, it
could easily have that, and that's

45
00:03:28.439 --> 00:03:30.960
all well to the good. But
on the other hand, when we look

46
00:03:30.000 --> 00:03:36.919
at the use of these systems by
companies, and I'm very concerned I am

47
00:03:36.960 --> 00:03:40.919
a privacy attorney about the effect on
our privacy, especially from our government let

48
00:03:40.919 --> 00:03:46.719
a loan other governments, and the
laws have not been updated as they should

49
00:03:46.719 --> 00:03:50.400
have been. The laws they were
updated, I think in a bad way

50
00:03:50.479 --> 00:03:53.919
after nine to eleven, and we
haven't come to grips of that. So,

51
00:03:54.000 --> 00:03:57.800
for example, the FBI, we
found out from the Federal Court a

52
00:03:57.840 --> 00:04:01.800
couple of months ago, has violated
the rights without the search warrant of the

53
00:04:01.919 --> 00:04:08.039
PHISER Rights, an intelligence surveilling fact
of over two hundred and seventy thousand citizens

54
00:04:08.120 --> 00:04:12.000
US citizens. That's in the last
four years, just just what we know

55
00:04:12.039 --> 00:04:15.199
of. And now that key provision
is called seven oh two of PHISER Actor

56
00:04:15.279 --> 00:04:18.319
is up for renewal this year.
So that's an issue. We have to

57
00:04:18.319 --> 00:04:24.519
address that in many others, but
not just that. Leads the big agencies

58
00:04:24.639 --> 00:04:29.120
have computers and software that every day
gets better. So even if we don't

59
00:04:29.199 --> 00:04:31.720
change the laws, their ability to
monitor what we have or now we learned

60
00:04:31.800 --> 00:04:36.079
last week that the FBI goes online
and buys our own private data that it

61
00:04:36.079 --> 00:04:40.199
couldn't get without a search warrant.
And what they can do with that is

62
00:04:40.199 --> 00:04:44.360
shocking. And what other actors can
do with that sort of thing for the

63
00:04:44.399 --> 00:04:46.240
next election, Wow, it's going
to be interesting to see. I think

64
00:04:46.240 --> 00:04:49.639
we can call that, will likely
call that the AI election in twenty twenty

65
00:04:49.680 --> 00:04:55.319
four uncharted, not just to predict
how people will vote, but to manipulate

66
00:04:55.360 --> 00:04:58.240
them. Well, that's what concerns
me more than anything. And I don't

67
00:04:58.279 --> 00:05:01.480
even I'm not even too worried about
the government. Doctor Bart costcos with us

68
00:05:01.480 --> 00:05:05.079
and he's his number one selling book, is Fuzzy Thinking. He's also an

69
00:05:05.120 --> 00:05:09.600
AI expert. My concern isn't what
the government's going to do with this.

70
00:05:09.759 --> 00:05:14.439
My concern is what commerce and the
marketplace will do with this, and pollsters

71
00:05:14.720 --> 00:05:17.319
that work for a particular political party
who are not bound by the constitution.

72
00:05:18.360 --> 00:05:23.519
Already, already we are so monitored
by the Googles and the facebooks and the

73
00:05:23.560 --> 00:05:28.560
twitters of the world right, and
then there's been some interaction we've been learning

74
00:05:28.639 --> 00:05:31.399
thanks to the Twitter files and others, with these government agencies and lee.

75
00:05:31.560 --> 00:05:35.279
Beyond that, there's these other actors
out there. We learn from the Snowden

76
00:05:35.360 --> 00:05:39.959
cases, and I recommend Oliver Stone's
movie If People haven't seen it. It's

77
00:05:40.000 --> 00:05:44.959
not just the abuses we learned.
There were entire programs, surveillance programs in

78
00:05:45.079 --> 00:05:48.920
ways to circumvent our Fourth Amendment,
which protects right of search and seizure by

79
00:05:48.959 --> 00:05:55.160
simply getting our buddies in England or
New Zengland or New Zealand or even Canada

80
00:05:55.319 --> 00:05:59.120
to do the dirty workforce. And
then there are the other governments who are

81
00:05:59.160 --> 00:06:01.839
doing a lot with that. So
the profiles being built about you and the

82
00:06:01.920 --> 00:06:08.120
like you don't know, and just
the small, comparatively small problems of deep

83
00:06:08.160 --> 00:06:12.279
fakes are out there. There could
be stuff going on with your likeness and

84
00:06:12.519 --> 00:06:15.800
voice. You may never know it. You may not get a job offer

85
00:06:15.879 --> 00:06:17.879
because of something you're a deep faked
on, and that's a real problem.

86
00:06:17.879 --> 00:06:21.600
We have a problem with hallucinations,
We have a problem with privacy. I

87
00:06:21.639 --> 00:06:26.079
think we'll address it. But at
the same time, computing powers growing so

88
00:06:26.160 --> 00:06:30.160
fast and the big tech companies are
racing so fast we don't quite have time

89
00:06:30.199 --> 00:06:33.879
to catch up. Well, that
seems to be in a lot of industries.

90
00:06:33.959 --> 00:06:39.079
But I've been trying doctor Bard Costco's
with as fuzzy thinking as his book,

91
00:06:39.360 --> 00:06:42.800
and it's about to be reissued digitally, and he's an AI expert.

92
00:06:43.120 --> 00:06:47.800
I try to focus on what this
massive amount of computing can do for the

93
00:06:47.839 --> 00:06:53.199
medical industry. For example, it
is probably not out of the realm of

94
00:06:53.240 --> 00:06:59.600
possibility for there to be a microbot
developed that can have artificial intelligence that will

95
00:06:59.639 --> 00:07:03.600
know where the cancer cell is and
can go and destroy it. And things

96
00:07:03.639 --> 00:07:09.600
like that are what excite me.
Yeah, absolutely. I mean one consequence

97
00:07:09.720 --> 00:07:13.199
of the coronavirus vaccines. I know
there are a lot of problems with them

98
00:07:13.199 --> 00:07:18.480
and the issuance, but for thirty
years folks have been working on these Manolipo

99
00:07:18.639 --> 00:07:23.680
systems like that for just what you're
talking about, to teach the system some

100
00:07:23.879 --> 00:07:27.560
signal detection capable. It may not
have an AI computer that actually may be

101
00:07:27.720 --> 00:07:33.079
kind of housed in its biologies,
but to detect very specifically that particular cancer

102
00:07:33.120 --> 00:07:36.439
cell, or not just cancer cell, it may be the zombie cells or

103
00:07:36.439 --> 00:07:40.360
things like that. You will see
a lot of that. I think lots

104
00:07:40.360 --> 00:07:43.399
of cremic courts in all those cases
lead. I don't think you want to

105
00:07:43.399 --> 00:07:46.800
be the first person to experiment.
Oh no, no, the third or

106
00:07:46.800 --> 00:07:49.720
fourth pass it. Maybe give it
a go, but that's all to the

107
00:07:49.759 --> 00:07:56.319
good as you throw more statistics,
optimization algorithms and computing power at that kind

108
00:07:56.319 --> 00:08:00.480
of thing, a classic signal detection. I think the benefits will be marvelous.

109
00:08:00.480 --> 00:08:05.319
Plus the effects of implanting chips to
resource sight and a lot of other

110
00:08:05.319 --> 00:08:07.639
problems, it's already starting to happen. We do some of that at USC

111
00:08:07.800 --> 00:08:11.120
Actually. Ah, excellent, Yeah, because I mean that's the new that

112
00:08:11.160 --> 00:08:16.519
seems to be the new things like
glaucoma and retinal deterioration. That seems to

113
00:08:16.519 --> 00:08:18.560
be the new threat that. No, but we can't. We don't have

114
00:08:18.560 --> 00:08:22.560
anything to do about We can't.
We can't do anything about it. We'd

115
00:08:22.560 --> 00:08:26.920
like to grow new eyes at some
point and in the interim back up what

116
00:08:26.040 --> 00:08:30.639
you've got with your eyes. All
those sorts of things are coming very doable.

117
00:08:30.680 --> 00:08:33.559
They don't violate any law of physics. It's just a matter of stealing.

118
00:08:33.879 --> 00:08:37.639
And if you just look at just
this past century of technology again,

119
00:08:37.720 --> 00:08:41.679
nothing that goes first goes well,
whether it's the first car or the first

120
00:08:41.720 --> 00:08:46.320
airplane or first cell phone, but
man, does it come in time.

121
00:08:46.559 --> 00:08:50.840
What you want is something that has
something like Moore's law of the doubling of

122
00:08:50.919 --> 00:08:56.000
chip density on your side, and
AI has that signal detection has that those

123
00:08:56.039 --> 00:09:01.519
things will only get better and much
faster. Doctor Bardcosco for someone listening and

124
00:09:01.440 --> 00:09:05.159
was taking note of what you were
saying earlier about deep fakes, I mean,

125
00:09:05.200 --> 00:09:07.399
is there anything we can do in
general or just not turn on our

126
00:09:07.440 --> 00:09:11.759
phones ever? Ever? Yeah,
you've got to be careful with your information.

127
00:09:13.080 --> 00:09:16.679
I personally don't participate in social media
for legal reasons. I think it's

128
00:09:16.720 --> 00:09:20.480
a huge mistake and people are going
to be growing profiles of you not I

129
00:09:20.480 --> 00:09:24.360
mean, we know, for example, China does this just across the board

130
00:09:24.440 --> 00:09:28.080
for people. But I think it's
being done. I'm guessing at a mass

131
00:09:28.080 --> 00:09:33.440
scale by corporations and others. You
don't know who's selling your information. I

132
00:09:33.480 --> 00:09:39.399
would be very hesitant to put my
information online. Doctor bard Costco AI expert

133
00:09:39.559 --> 00:09:43.440
author and attorney as well as a
professor. His book is Fuzzy Thinking,

134
00:09:43.639 --> 00:09:48.519
probably the best primer around for learning
what fuzzy logic is all about, and

135
00:09:48.519 --> 00:09:52.840
it's available rereleased digitally. Don't have
to give your personal information to get it.

136
00:09:52.879 --> 00:09:54.519
But I thank you doctor for joining
us a bit. Thank you,

137
00:09:56.000 --> 00:10:01.279
thanks for listening to Later with Lee
Matthews, The Lee Met Matthews podcast and

138
00:10:01.399 --> 00:10:07.159
remember to listen to The Drive Live
weekday afternoons from five to seven and iHeartMedia presentation