공공 기관들 간의 AI를 위한 데이터 공유

00:05 [music]
00:13 

00:13 >> Across the Asia Pacific, governments are
00:15 

00:15 actively regulating how public agencies
00:17 

00:17 share data.
00:20 

00:20 Singapore's Public [music] Sector
00:21 

00:21 Government Act, Australia's Data
00:23 

00:23 Availability and Transparency Act,
00:25 

00:25 China's Data Security Law and Public
00:27 

00:27 Sector Data Directives all govern how
00:29 

00:29 public sector [music] data may be shared
00:31 

00:31 and protected under heavy safeguard.
00:34 

00:34 Here's the challenge.
00:36 

00:36 To ensure inter-agency collaboration.
00:39 

00:39 How do you share live, highly governed,
00:41 

00:41 and semantically unified data across
00:43 

00:43 agencies for human and AI consumption
00:45 

00:45 respecting these data sharing laws?
00:48 

00:48 Denodo provides the AI data layer
00:51 

00:51 enabled by active context for data
00:53 

00:53 sharing and for agents.
00:55 

00:55 Four pillars make it work. Universal
00:58 

00:58 connectivity, zero copy delivery,
00:60 

00:60 [music] governed access, and semantic
01:02 

01:02 trust.
01:03 

01:03 Hold on a few minutes and you will see
01:05 

01:05 [music] it working across an actual
01:06 

01:06 smart city.
01:08 

01:08 Let's start with the Denodo data
01:10 

01:10 marketplace, which is an intuitive shop
01:12 

01:12 for data experience with semantic trust
01:14 

01:14 [music] built in.
01:16 

01:16 A modern city has several independent
01:18 

01:18 agencies like emergency [music]
01:19 

01:19 services, police, public health, and
01:22 

01:22 transport.
01:24 

01:24 Here, each agency [music] owns and
01:26 

01:26 governs its own data products, metadata,
01:28 

01:28 and business definitions inside its own
01:30 

01:30 boundary called a domain.
01:33 

01:33 These domains can be the foundation of
01:34 

01:34 governed inter-agency sharing. This is a
01:37 

01:37 shared operations domain.
01:39 

01:39 It combines authorized information from
01:41 

01:41 all four agencies.
01:43 

01:43 Live, zero copy, [music] and the
01:45 

01:45 governance policies are enforced right
01:47 

01:47 here in the data layer.
01:49 

01:49 For example, through this shared
01:51 

01:51 operations domain, the police agency
01:53 

01:53 will not have access to data owned by
01:55 

01:55 public health while engaging in shared
01:57 

01:57 operations.
01:59 

01:59 It does not matter which data sources
02:01 

02:01 each individual agencies are using.
02:04 

02:04 The Denodo platform federates authorized
02:06 

02:06 information from all the four agencies.
02:09 

02:09 This is universal connectivity.
02:11 

02:11 Metric views are reusable metric objects
02:13 

02:13 defined once with sources, dimensions,
02:16 

02:16 and metrics.
02:18 

02:18 So, a measure like active incidents
02:20 

02:20 across the city has one consistent
02:22 

02:22 definition.
02:23 

02:23 This makes sure that every human and AI
02:25 

02:25 reads the same certified number for
02:27 

02:27 active incidents or average severity.
02:31 

02:31 This is semantic trust in action.
02:33 

02:33 Now that we have seen the foundations,
02:35 

02:35 let us take a visit to Meridian City
02:37 

02:37 [music] and explore how the Denodo
02:39 

02:39 platform helps the local authorities
02:40 

02:40 deal with a sudden tier one emergency.
02:44 

02:44 This is Meridian City Smart City
02:46 

02:46 Operations Center.
02:48 

02:48 The three-dimensional situation map of
02:50 

02:50 Meridian City has colored columns, which
02:52 

02:52 are incidents, and the height of each is
02:54 

02:54 risk severity.
02:56 

02:56 The red pulse is the critical incident
02:58 

02:58 we are investigating in this example.
03:00 

03:00 >> [music]
03:01 

03:01 >> A hazardous material spill just hit the
03:03 

03:03 Meridian interchange.
03:05 

03:05 It touches all four agencies at once.
03:09 

03:09 Let's see how we can arrange a
03:10 

03:10 coordinated response across all the
03:12 

03:12 agencies governed by interagency data
03:14 

03:14 sharing regulation.
03:17 

03:17 Here is the AI operations agent.
03:20 

03:20 The AI has analyzed the incident and
03:22 

03:22 proposed a coordinated response, and
03:24 

03:24 every action is assigned to the agency
03:26 

03:26 that owns it.
03:28 

03:28 Now, let's ask this agent what to do
03:29 

03:29 next as a fictional operations
03:31 

03:31 commander.
03:33 

03:33 The agent has come up with a coordinated
03:35 

03:35 response plan.
03:37 

03:37 For example, secure perimeter, dispatch
03:39 

03:39 ambulances, set up air [music]
03:41 

03:41 monitoring, etc.
03:43 

03:43 Now, if we switch the operating agency
03:45 

03:45 to police and ask the same question.
03:49 

03:49 The AI's answer changed completely.
03:52 

03:52 You can see that the actions listed are
03:53 

03:53 applicable only to the police department
03:55 

03:55 or agency.
03:57 

03:57 Let's try the same experiment with the
03:59 

03:59 public health agency.
04:01 

04:01 Again, you can see that the action plan
04:03 

04:03 recommended has been tailored
04:05 

04:05 specifically for public health.
04:08 

04:08 Now, let's look at a fire test of this
04:09 

04:09 governed access layer.
04:12 

04:12 As the public health agency, let's ask a
04:14 

04:14 question that a public health persona
04:16 

04:16 may not necessarily need to know the
04:17 

04:17 answer to.
04:19 

04:19 For example, who is the suspect?
04:22 

04:22 As you can see the AI answers, but
04:24 

04:24 strictly [music] within policy.
04:27 

04:27 Police investigation details are not
04:29 

04:29 available within the public health
04:30 

04:30 operating context.
04:32 

04:32 However, the police has secured the
04:34 

04:34 scene.
04:35 

04:35 It told public health exactly what they
04:37 

04:37 are allowed to know.
04:39 

04:39 This is one step above using the simple
04:41 

04:41 [music] masking or reduction you see in
04:43 

04:43 many other platforms.
04:45 

04:45 Active context here is the enabler of
04:47 

04:47 governance rather than simple data
04:49 

04:49 protection.
04:51 

04:51 As the operations commander, let's
04:53 

04:53 approve the response [music] plan, which
04:55 

04:55 has been recommended by the AI
04:56 

04:56 operations agent.
04:59 

04:59 Response plan has been approved and is
05:00 

05:00 immediately auditable [music] with
05:02 

05:02 timestamps included.
05:04 

05:04 Four agencies just coordinated one
05:06 

05:06 response over live governed data that
05:08 

05:08 never left each agency's ownership.
05:12 

05:12 In many platforms, AI reasons over a
05:14 

05:14 centralized copy of data.
05:17 

05:17 What you saw is AI reasoning over the
05:18 

05:18 Denodo I data layer, which is enabled by
05:21 

05:21 context, delivering zero copy across
05:24 

05:24 [music] four agency domains in real time
05:26 

05:26 with policies enforced.